RU2840869C1 - Systems and methods for processing retinal signal data and identifying conditions - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: group of inventions relates to systems and methods for processing retinal signal data generated by light stimulation. Method and a system for predicting the probability that a patient is susceptible to one or more conditions are disclosed. Retinal signal data corresponding to the patient is received. Retinal signal features can be extracted from retinal signal data. Data can be applied to the first mathematical model and the second mathematical model. First mathematical model can correspond to the first state. Second mathematical model can correspond to the second state. First predicted probability for the first state may be generated. Second predicted probability for the second state may be generated. First predicted probability and a second predicted probability may be output.

EFFECT: group of inventions provides processing of retinal signal data and identification of conditions.

14 cl, 58 dwg

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[01] This application claims the benefit of U.S. Provisional Patent Application No. 63/000,055, filed March 26, 2020, U.S. Provisional Patent Application No. 63/038,257, filed June 12, 2020, and U.S. Provisional Patent Application No. 63/149,508, filed February 15, 2021, each of which is hereby incorporated by reference in its entirety.

FIELD OF TECHNOLOGY TO WHICH THE INVENTION RELATES

[02] The present invention relates to systems and methods for processing retinal signal data generated by light stimulation.

LEVEL OF TECHNOLOGY

[03] Clinicians may need to determine whether a patient is susceptible to a medical condition, such as a psychiatric condition or a neurological condition. The clinician may compare the patient to known criteria to determine which condition the patient is susceptible to. In some cases, a patient may meet more than one condition, and it may be difficult or impossible for the clinician to distinguish between the conditions. It may be preferable for the clinician to have a tool to assist in determining and/or confirming whether a patient is susceptible to a medical condition and/or distinguishing between the conditions.

[04] The object of the present invention is to overcome at least some of the limitations of the prior art.

ESSENCE OF THE INVENTION

[05] Embodiments of the present invention are developed based on the developers' concerns regarding some of the shortcomings of existing systems for determining medical conditions.

[06] Identification of biomarkers and/or biosignatures of conditions, such as psychiatric, mental health, or neurological conditions such as schizophrenia, bipolar disorder, or depression, may allow healthcare professionals to make an earlier determination of the condition, identify the condition to which a patient is susceptible when there are multiple possible conditions, and/or implement early and possibly preventive intervention. This early determination of a medical condition may improve patient treatment and/or prognosis.

[07] Embodiments of the present invention are developed based on the observation made by the developers that data obtained from electroretinograms (ERGs) allow for improved detection of medical conditions. However, existing methods for collecting and analyzing electroretinograms (ERGs) allow for the collection and analysis of only a limited amount of information from the captured electrical signals. It was found that increasing the amount of collected information regarding the retinal response to light stimulation allowed for the generation of retinal signal data with a higher information density, an increased amount of information, and/or additional types of information. This retinal signal data allows for multimodal display of electrical signals and/or other data and allows for the detection of additional multimodal display features characteristic of certain conditions. The multimodal display may include multiple parameters of the retinal signal data, such as time, frequency, light stimulation parameters, and/or any other parameter.

[08] Embodiments of the present invention form the basis of an improved methodology for determining medical conditions based on processing retinal signal data that provides a greater volume of information, a greater information density, and/or additional types of information detail compared to traditional ERG data, and which is referred to herein as “retinal signal processing and analysis” (RSPA). These retinal signal data allow, in some embodiments, mathematical modeling of data sets containing large volumes of information, identifying retinal signal features, making it possible to identify biomarkers and/or biosignatures in retinal signal data using, for example, retinal signal features. Some non-essential embodiments of the present invention also provide methods for collecting retinal signal data that have a greater volume of information, a greater information density, and/or additional types of information compared to ERG data.

[09] In some embodiments of the present invention, more accurate detection of certain medical conditions or more discriminative discrimination between medical conditions can be achieved. Detection of medical conditions or discriminative discrimination between conditions can be achieved in a variety of related (e.g., gender, age, disease onset, retinal pigmentation, iris color) and/or confounding factors (e.g., onset of conditions, drug use, results of certain treatment procedures, episodes of psychosis, anxiety, depression, overlap of signs and symptoms common to several disorders). Increasing the level of detail and/or increasing the number of retinal signal features to capture and analyze has a direct impact on the ability to identify biosignatures using retinal signal data for better discrimination between conditions, such as pathological conditions, as well as better description of conditions compared to a control (which may be referred to as a non-pathological condition). For example, embodiments of the present methods may be based on retinal signal data captured at a higher sampling rate and/or over a longer period of time compared to traditional ERG. The retinal signal data may include additional features recorded with electrical signals, such as, but not limited to, impedance, wavelength of light, spectrum of light, or intensity of light reaching the retina. Capturing additional information, at a higher sampling rate and/or data collected with an extended range of retinal light stimulation over an extended period of time, and a multidimensional representation thereof, may be referred to as "high-density retinal signal data." The retinal signal data captured at high density may contain more information than the data captured previously during an ERG (referred to as a "traditional ERG"). The retinal signal data may be voltage independent and/or time independent, unlike traditional ERG.

[10] In some embodiments, more efficient processing of retinal signal data is possible using high-density retinal signal data. The advantage of high-density retinal signal data over traditional ERG data is an increased amount of information associated with electrical signals and additional features of the retinal signal, and thus a more detailed biosignature. Thus, a higher level of detail allows for better discrimination between different conditions and the preparation of a series of classifiers representing the features of the biosignature of each condition.

[11] According to a first aspect of the present invention, there is provided a method for generating a mathematical model corresponding to a first condition, wherein the method is executed by at least one processor of a computer system, wherein the method comprises: generating a mathematical model corresponding to the first condition, wherein the method is executed by at least one processor of a computer system, wherein the method comprises: receiving a plurality of data arrays of labeled retinal signal data corresponding to a plurality of patients, wherein each data array comprises retinal signal data of the patient and a label, wherein the label indicates whether the patient is susceptible to the first condition; extracting a set of features from the retinal signal data; selecting a subset of features from the set of features, wherein the subset of features corresponds to biomarkers of the first condition; and determining, based on the subset of features, one or more classifiers that make it possible to distinguish the first condition from the second condition.

[12] In some embodiments of the method, the set of features comprises a voltage, a circuit impedance, a signal acquisition time, a sampling frequency, a light stimulation synchronization time, a light stimulation offset, or indications of illuminated areas of the retina.

[13] In some embodiments of the method, the set of features comprises the position of the eye, the size of the pupil, the intensity of the applied illumination, the frequency of the light stimulation, the sampling frequency of the retinal signal, the wavelength of the illumination, the time of illumination, the background wavelength, or the background illumination.

[14] In some embodiments of the method, one or more classifiers allow distinguishing a biosignature of a first state from a biosignature of a second state.

[15] In some embodiments of the method, the method further comprises ranking the set of features based on the relevance of each feature of the biosignature of the first state, and wherein selecting the subset of features comprises selecting the highest ranked features from the set of features.

[16] In some embodiments of the method, the method further comprises: receiving clinical information cofactors corresponding to a plurality of patients, wherein each data set comprises cofactors of the patient's clinical information; and selecting a subset of the clinical cofactors, wherein the clinical cofactors in the subset of the clinical cofactors influence the detection of biomarkers.

[17] In some embodiments of the method, the clinical information cofactors indicate the age, gender, skin pigmentation, or iris color of the respective patient.

[18] In some embodiments of the method, determining one or more classifiers comprises determining, based on a subset of clinical cofactors, one or more classifiers.

[19] In some embodiments of the method, the method further comprises generating a mathematical model based on one or more classifiers.

[20] In some embodiments of the method, the method further comprises: inputting into the mathematical model retinal signal data and clinical information cofactors corresponding to the patient; and outputting from the mathematical model a predicted probability that the patient is susceptible to the first condition.

[21] In some embodiments of the method, the method further comprises: inputting into the mathematical model retinal signal data and clinical information cofactors corresponding to the patient; and outputting from the mathematical model a predicted probability that the patient is not susceptible to the first condition.

[22] In some embodiments of the method, the first condition is schizophrenia, bipolar disorder, major depressive disorder, or psychosis.

[23] In some embodiments of the method, the first condition is post-traumatic stress disorder, seizure, substance abuse, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, or attention deficit disorder.

[24] In some embodiments of the method, the retinal signal data has a sampling frequency of 4 to 24 kHz.

[25] In some embodiments of the method, the retinal signal data is collected over a signal collection time of 200 milliseconds to 500 milliseconds.

[26] In some embodiments of the method, the retinal signal data comprises an impedance component of the receiving circuit that is continuously recorded during acquisition of the retinal signal data.

[27] In some embodiments of the method, the retinal signal data comprises one or more optical parameters.

[28] In some embodiments of the method, the optical parameters comprise the intensity of light stimulation of the retina or the size of the pupil.

[29] According to another aspect of the present invention, there is provided a method for predicting the probability that a patient is susceptible to one or more conditions, wherein the method is executed by at least one processor of a computer system, wherein the method comprises the steps of: receiving retinal signal data corresponding to the patient; extracting, from the retinal signal data, one or more features of the retinal signal; extracting, from the features of the retinal signal, one or more descriptors; applying the one or more descriptors to a first mathematical model and a second mathematical model, wherein the first mathematical model corresponds to a first condition and the second mathematical model corresponds to a second condition, thereby generating a first predicted probability for the first condition and a second predicted probability for the second condition; and outputting the first predicted probability and the second predicted probability.

[30] In some embodiments of the method, the method further comprises displaying an interface comprising the first predicted probability and the second predicted probability.

[31] In some embodiments of the method, the method further comprises storing the first predicted probability and the second predicted probability.

[32] In some embodiments of the method, the method further comprises an array of retinal signal data.

[33] In some embodiments of the method, the method further comprises: obtaining clinical information cofactors extracted from clinical information corresponding to the patient; and applying the clinical information cofactors to the first mathematical model and the second mathematical model.

[34] In some embodiments of the method, the clinical information cofactors correspond to the age, gender, skin pigmentation, or iris color of the patient.

[35] In some embodiments of the method, the retinal signal data has a sampling frequency of 4 to 24 kHz.

[36] In some embodiments of the method, retinal signal data is collected over a signal collection time of 200 milliseconds to 500 milliseconds.

[37] In some embodiments of the method, the retinal signal data comprises an impedance component of the receiving circuit that is continuously recorded during acquisition of the retinal signal data.

[38] In some embodiments of the method, the retinal signal data comprises one or more optical parameters.

[39] In some embodiments of the method, the optical parameters comprise the intensity of light stimulation of the retina or the size of the pupil.

[40] In some embodiments of the method, the first condition is a medical condition and the second condition is a control condition.

[41] In some embodiments of the method, the first condition or the second condition is schizophrenia, bipolar disorder, major depressive disorder, or psychosis.

[42] In some embodiments of the method, the first condition or the second condition is post-traumatic stress disorder, seizure, substance abuse, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, or attention deficit disorder.

[43] In some embodiments of the method, the method further comprises receiving user input indicating a selection of the first state and the second state.

[44] In some embodiments of the method, the method further comprises: selecting a drug based on the first predicted probability and the second predicted probability; and administering the drug to the patient.

[45] According to another aspect of the present invention, there is provided a method for determining a biosignature of a condition, wherein the method is executed by at least one processor of a computer system, wherein the method comprises: receiving a plurality of data arrays of labeled retinal signal data corresponding to a plurality of patients, wherein each data array comprises retinal signal data of the patient and a label, wherein the label indicates whether the patient is susceptible to the condition; extracting a set of features from the retinal signal data; selecting a subset of features from the set of features, wherein the subset of features corresponds to biomarkers of the condition; and determining, based on the subset of features, one or more classifiers that identify a biosignature of the condition.

[46] In some embodiments of the method, the method further comprises: receiving clinical information cofactors corresponding to a plurality of patients, wherein each data set comprises cofactors of the patient's clinical information; and selecting a subset of clinical cofactors, wherein the clinical cofactors in the subset of clinical cofactors influence the detection of biomarkers, and wherein determining the one or more classifiers comprises determining, based on the subset of clinical cofactors, the one or more classifiers.

[47] In some embodiments of the method, the method further comprises ranking the set of features based on the relevance of each feature of the biosignature of the first state, and wherein selecting the subset of features comprises selecting the highest ranked features from the set of features.

[48] According to another aspect of the present invention, there is provided a system for predicting the probability that a patient is susceptible to one or more conditions, the system comprising: a light stimulator; one or more sensors; a computer system comprising the system comprising at least one processor and a memory storing a plurality of executable instructions that, when executed by the at least one processor, cause the system to: cause the light stimulator to apply light stimulation signals to the retina of the patient; collect, via the one or more sensors, electrical signals in response to the light stimulation; generate, based on the electrical signals, retinal signal data corresponding to the patient; extract, from the retinal signal data, one or more features of the retinal signal; extract, from the features of the retinal signal, one or more descriptors; apply the one or more descriptors to a first mathematical model and a second mathematical model, wherein the first mathematical model corresponds to a first state and the second mathematical model corresponds to a second state, thereby generating a first predicted probability for the first state and a second predicted probability for the second state; and output the first predicted probability and the second predicted probability.

[49] In some embodiments of the system, the retinal signal data comprises a wavelength component of light recorded during acquisition of the retinal signal data.

[50] In some embodiments of the system, the retinal signal data comprises a component of the light spectrum recorded during acquisition of the retinal signal data.

[51] In some embodiments of the system, the retinal signal data comprises a light intensity component recorded during acquisition of the retinal signal data.

[52] In some embodiments of the system, the retinal signal data comprises a component of the illuminated retinal surface recorded during acquisition of the retinal signal data. The retinal surface component may indicate an illuminated region of the retinal surface.

[53] According to another aspect of the present invention, there is provided a system for predicting the probability that a patient is susceptible to one or more conditions, wherein the system comprises a computer system comprising at least one processor and a memory storing a plurality of executable instructions that, when executed by the at least one processor, cause the computer system to: receive retinal signal data corresponding to the patient; extract, from the retinal signal data, one or more features of the retinal signal; extract, from the features of the retinal signal, one or more descriptors; apply the one or more descriptors to a first mathematical model and a second mathematical model, wherein the first mathematical model corresponds to a first state and the second mathematical model corresponds to a second state, thereby generating a first predicted probability for the first state and a second predicted probability for the second state; and output the first predicted probability and the second predicted probability.

[54] In some embodiments of the system, the system further comprises a light stimulator and one or more sensors, and instructions, when executed by at least one processor, cause the computer system to: cause the light stimulator to apply light stimulation signals to the retina of the patient; collect, via the one or more sensors, electrical signals in response to the light stimulation; and generate, based on the electrical signals, retinal signal data.

[55] In some embodiments of the system, the system further comprises a display, and wherein the instructions, when executed by the at least one processor, cause the system to output, via the display, an interface comprising the first predicted probability and the second predicted probability.

[56] According to another aspect of the present invention, a method for monitoring a patient's condition is provided, the method being executed by at least one processor of a computer system, the method comprising: receiving retinal signal data corresponding to the patient; extracting, from the retinal signal data, one or more features of the retinal signal; extracting, from the features of the retinal signal, one or more descriptors; applying the one or more descriptors to a mathematical model corresponding to the condition, thereby generating a predicted probability for the condition; and outputting the predicted probability.

[57] In some embodiments of the method, the method further comprises: selecting a drug based on the predicted probability; and administering the drug to the patient.

[58] In some embodiments of the method, the retinal signal data comprises retinal signal data captured during treatment of a patient for a condition.

[59] In some embodiments of the method, the condition is schizophrenia, bipolar disorder, major depressive disorder, psychosis, post-traumatic stress disorder, seizure, substance abuse, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, or attention deficit disorder.

[60] In the context of the present description of the invention, unless expressly stated otherwise, a computer system means, but is not limited to, an “electronic device”, “operating system”, “system”, “computer system”, “control unit”, “control device” and/or any combination thereof suitable for solving the current problem.

[61] In the context of the present description of the invention, unless expressly stated otherwise, the expression “computer-readable medium” and “memory” are intended to include media of any nature and kind, non-limiting examples of which include RAM, ROM, disks (CD-ROMs, DVDs, floppy disks, hard disks, etc.), USB keys, flash memory cards, solid state disks and tape drives.

[62] In the context of the present description of the invention, a “database” is any structured data array, regardless of its specific structure, database management software, or computer hardware on which data is stored, implemented, or otherwise made available for use. A database may be located on the same hardware as the process that stores or uses the information stored in the database, or may be located on separate hardware, such as a dedicated server or multiple servers.

[63] In the context of the present description of the invention, unless expressly stated otherwise, the words “first”, “second”, “third”, etc. are used only as adjectives to distinguish similar nouns from each other, but not for the purpose of describing any specific relationship between these nouns.

[64] Each of the embodiments of the present invention has at least one of the above-mentioned problems and/or aspects, not necessarily all of them. It should be understood that some aspects of the present invention, due to attempts to solve the above-mentioned problem, may not correspond to this problem and/or may correspond to other problems not specifically mentioned herein.

[65] Additional and/or alternative features, aspects and advantages of embodiments of the present invention will be apparent from the following description, the accompanying drawings and the following claims.

BRIEF DESCRIPTION OF DRAWINGS

[66] For a better understanding of the present invention, as well as its other aspects and additional features, reference is made to the following description, which is to be used in conjunction with the accompanying drawings, where:

[67] Fig. 1 is a block diagram of an illustrative computing environment in accordance with various embodiments of the present invention;

[68] Fig. 2 is a block diagram of a retinal signal processing system in accordance with various embodiments of the present invention;

[69] Fig. 3 is an illustrative functional architecture of a method for processing information resulting in the construction of mathematical functions for predicting whether a patient is susceptible to a condition in accordance with various embodiments of the present invention;

[70] Fig. 4 is a flow chart of operations of a method for predicting the likelihood of a medical condition in accordance with various embodiments of the present invention;

[71] Fig. 5 is a flow chart of operations of a method for generating a mathematical model for predicting whether a patient is susceptible to a condition in accordance with various embodiments of the present invention;

[72] Fig. 6 is a flow chart of operations of a method for training a machine learning algorithm (MLA) for predicting the likelihood of a medical condition in accordance with various embodiments of the present invention;

[73] Fig. 7 is a flow chart of operations of a method of using MLA to predict the likelihood of a medical condition in accordance with various embodiments of the present invention.

[74] Fig. 8-27 are examples of time-frequency analysis and selection of discriminant regions based on statistical significance of higher frequency in accordance with various embodiments of the present invention;

[75] Figs. 28-42 are examples of selecting retinal signal features and displaying the statistical significance of the selected retinal signal features in accordance with various embodiments of the present invention; and

[76] Figs. 43-58 are examples of selecting and displaying descriptors based on the amount of discrimination between two states, in accordance with various embodiments of the present invention.

[77] It should be noted that unless expressly stated herein, the drawings are not drawn to scale.

DETAILED DESCRIPTION

[78] Some aspects and embodiments of the present invention relate to methods and systems for processing retinal signals. In a broad sense, some aspects and embodiments of the present invention are based on higher density retinal signal data, which can be obtained by any method, such as increasing the number of states for light stimulation (e.g., the number and range of light intensities), adding dynamic resistance (impedance) to the circuit used to collect electrical signals, capturing retinal signal data over a longer period of time, and/or capturing retinal signal data at a higher frequency. The present computer-implemented methods are provided for analyzing retinal signal data and extracting retinal signal features that are used together to further decipher biomarkers and/or biosignatures to minimize, reduce, or eliminate the limitations of the prior art. In some optional embodiments, methods and systems for capturing high density retinal signal data are provided.

[79] Some aspects and embodiments of the present invention provide methods and systems that can analyze retinal signal data and provide a predicted probability for specific conditions given a number of different cofactors of clinical information. The conditions may be psychiatric conditions, psychological conditions, neurological conditions, and/or any other type of medical condition. The predicted probability may indicate that a patient is currently susceptible to the condition and/or that the patient is at risk of developing the condition. For example, if a patient's parent is susceptible to the condition, the patient's retinal signal data may be analyzed to determine how likely it is that the patient is susceptible to the same condition as his or her parent.

[80] The systems and methods described herein allow for full or at least partial automation to minimize clinician input in determining a medical condition or a treatment plan for a medical condition. Predictions derived from the systems and methods described herein can be used to assist a clinician in determining a medical condition and/or developing a treatment plan for a patient.

[81] The systems and methods described herein may comprise generating a mathematical model for identifying a distance of a patient from a biosignature of a condition by: 1) collecting retinal signal data from patients, 2) tagging each retinal signal data of a patient with a label indicating a potential medical condition of the patient (that may be diagnosed by a clinician), 3) extracting retinal signal features from the retinal signal data, 4) selecting a subset of features that correspond to biomarkers of the condition, and/or 5) defining classifiers that distinguish the biosignature of the condition from the biosignature of other conditions. A biosignature may comprise portions of the retinal signal data that are characteristic of a condition. The retinal signal data may comprise multiple biosignatures, where each biosignature is characteristic of a condition. The condition may be, but is not limited to, schizophrenia, bipolar disorder, major depressive disorder, psychosis, post-traumatic stress disorder, seizure, substance abuse disorder, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, attention deficit disorder, and/or any other condition. These steps may be used to build a mathematical model of any condition that has a biomarker embedded in the retinal signal data. The retinal signal data collection steps described herein may be applied to the analysis of features characteristic of any condition that are expressed in the retinal signal data.

[82] The systems and methods described herein may comprise predicting whether a patient is susceptible to a condition by: 1) collecting retinal signal data of the patient, 2) receiving a selection of conditions for examination, 3) extracting mathematical models corresponding to the selected conditions, 4) extracting retinal signal features from the retinal signal data, 5) extracting descriptors from the retinal signal features, where the descriptors refer to biosignatures of the selected conditions in the retinal signal data, 6) applying the descriptors to the mathematical models, and/or 7) outputting predicted probabilities that the patient is susceptible to each condition. Clinical information of the patient may be collected. Cofactors of the clinical information may be generated using the clinical information. Cofactors of the clinical information may also be applied to the mathematical models.

[83] The systems and methods described herein may be based on retinal signal data having a higher level of information compared to data captured by traditional ERG. The collected retinal signal data may be analyzed using mathematical and statistical computations to extract specific features of the retinal signal. The retinal signal features may comprise parameters of the retinal signal data and/or features generated using the retinal signal data. Descriptors may be extracted from the retinal signal features. Graphical representations of the obtained data may be processed and output and may provide visual support for the choices made in selecting relevant features and/or descriptors of the retinal signal. Applications may apply mathematical and/or statistical analysis of the results, allowing for the quantification of these features and/or descriptors of the retinal signal and comparisons of different conditions. Based on the retinal signal data and/or any other clinical information, classifiers may be constructed that describe the biosignature of the condition identified in the retinal signal data. The patient's retinal signal data can be collected and the distance between the patient's retinal signal data and the identified biosignatures can be determined, for example, using classifiers.

Computing environment

[84] Fig. 1 shows a computing environment 100 that can be used to implement and/or perform any of the methods described herein. In some embodiments, the computing environment 100 can be implemented by any of a conventional personal computer, a network device, and/or an electronic device (for example, but not limited to, a mobile device, a tablet device, a server, a control unit, a control device, etc.) and/or any combination thereof suitable for solving the current problem. In some embodiments, the computing environment 100 contains various hardware components, including one or more single- or multi-core processors, collectively represented by a processor 110, a solid-state disk 120, RAM 130, and an input/output interface 150. The computing environment 100 can be a computer specially designed to execute a machine learning algorithm (MLA). The computing environment 100 can be a general-purpose computer system.

[85] In some embodiments, the computing environment 100 may also be a subsystem of one of the above-mentioned systems. In some other embodiments, the computing environment 100 may be a “off-the-shelf” general-purpose computer system. In some embodiments, the computing environment 100 may also be distributed among multiple systems. The computing environment 100 may also be specially designed to implement the present invention. One skilled in the art may come up with several embodiments of the computing environment 100 that do not depart from the scope of the present invention.

[86] It will be apparent to those skilled in the art that the processor 110 typically represents the processing capability. In some embodiments, instead of or in addition to one or more traditional central processing units (CPUs), one or more specialized processing cores may be provided. For example, in addition to or in place of one or more CPUs, one or more graphics processing units (GPUs), tensor processing units (TPUs), and/or other so-called accelerated processors (or processing accelerators) may be provided.

[87] The system memory typically includes the random access memory 130, but is more generally intended to encompass any type of system memory, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), read-only memory (ROM), or a combination thereof. The solid state disk 120 is shown as an example of a mass storage device, but more generally such a mass storage device may comprise any type of storage device configured to store data, programs, and other information, and to make the data, programs, and other information accessible via the system bus 160. For example, the mass storage device may comprise one or more of a solid state disk, a hard disk, a magnetic disk drive, and/or an optical disk drive.

[88] Communication between the various components of the computing environment 100 may be provided by a system bus 160 comprising one or more internal and/or external buses (e.g., a PCI bus, a universal serial bus, an IEEE 1394 “Firewire” bus, a SCSI bus, a Serial-ATA bus, an ARINC bus, etc.) to which the various hardware components are electronically connected.

[89] The input/output interface 150 provides network capabilities, such as wired or wireless access. As an example, the input/output interface 150 may comprise a network interface, such as, but not limited to, a network port, a network socket, a network interface controller, etc. Those skilled in the art may provide several options for implementing the network interface. For example, the network interface may implement specific physical layer and data link layer standards, such as Ethernet, Fibre Channel, Wi-Fi, Token Ring, or serial communication protocols. A specific physical layer and data link layer may provide the basis for a complete network protocol stack, enabling communication between small groups of computers on the same local area network (LAN) and large-scale network communication via routed protocols, such as the Internet Protocol (IP).

[90] The input/output interface 150 may be connected to the touch screen 190 and/or one or more internal and/or external buses 160. The touch screen 190 may be part of a display. In some embodiments, the touch screen 190 is a display. The touch screen 190 may also be referred to as a screen 190. According to the embodiments illustrated in Fig. 1, the touch screen 190 comprises haptic hardware 194 (e.g., pressure-sensitive cells embedded in the display layer to detect physical interaction between the user and the display) and a haptic input/output controller 192 that communicates with the display interface 140 and/or one or more internal and/or external buses 160. In some embodiments, the input/output interface 150 may be connected to a keyboard (not shown), a mouse (not shown), or a touch pad (not shown), allowing the user to interact with the computing device 100 in addition to or instead of the touch screen 190.

[91] According to some implementations of the present invention, the solid-state disk 120 stores program instructions suitable for loading into the random access memory 130 and executable by the processor 110 to perform the actions of one or more methods described herein. For example, at least some of the program instructions may be part of a library or an application.

Retinal signal processing system

[92] Fig. 2 shows a block diagram of a retinal signal processing system 200 according to various embodiments of the present invention. The retinal signal processing system 200 may collect retinal signal data from a patient. As described above, compared to a traditional ERG, the retinal signal data captured using the retinal signal processing system 200 may contain additional features and/or data, such as impedance, a higher measurement frequency, an extended range of retinal light stimulation, and/or a longer measurement time. The retinal signal processing system 200 may process and/or analyze the collected data. The retinal signal processing system 200 may output a predicted probability that the patient is susceptible to a given condition, such as a medical condition.

[93] It should be clearly understood that the described system 200 is merely an illustrative implementation of the present invention. Accordingly, the following description presents merely illustrative examples of the present invention. This description does not serve to define the scope or limits of the present invention. In a number of instances, supposedly useful examples of modifications to the system 200 may also be set forth below. This is done merely as an aid to understanding, and again, is not intended to define the scope or limits of the present invention. This list of modifications is not exhaustive, and as one skilled in the art will appreciate, other modifications are possible. Additionally, where this is not done (i.e., where no example of modifications is given), it should not be assumed that no modifications are possible and/or that the only way to implement that element of the present invention has been described. As one skilled in the art will appreciate, this is likely not the case. In addition, it should be understood that the system 200 may provide a simple implementation of the present invention in some cases, and if so, they were presented in this manner as an aid to understanding. As will be understood by those skilled in the art, various implementations of the present invention may be more complex.

[94] The retinal signal processing system 200 may comprise a light stimulator 205, which may be an optical stimulator, for providing light stimulation signals to the retina of a patient. The retinal signal processing system 200 may comprise a sensor 210 for collecting electrical signals that occur in response to the optical stimulation. The retinal signal processing system 200 may comprise a data acquisition system 215, which may be a computing environment 100, for controlling the light stimulator 205 and/or collecting data measured by the sensor 210. For example, the light stimulator 205 and/or the sensor 210 may be a commercially available ERG system, such as the Espion Visual Electrophysiology System from DIAGNOSYS, LLC or UTAS and RETEVAL systems manufactured by LKC TECHNOLOGIES, INC.

[95] The light stimulator 205 may be any type of light source or sources that, individually or in combination, can generate light in a specified range of wavelength, intensity, frequency and/or duration. The light stimulator 205 may direct the generated light onto the retina of the patient. The light stimulator 205 may comprise light emitting diodes (LEDs) in combination with other light sources, such as one or more xenon lamps. The light stimulator 205 may provide a background light source.

[96] The light stimulator 205 may be configured to apply a light stimulation signal to the retina of the patient. The collected retinal signal data may depend on the light stimulation conditions. In order to maximize the ability to generate relevant retinal signal features in the retinal signal data, the light stimulator 205 may be configured to provide a wide variety of lighting conditions. The light stimulator 205 may be configured to control background light and/or stimulation light directed at the retina as flashes of light.

[97] The light stimulator may comprise any light sources capable of generating light beams that differ in wavelength (e.g., from about 300 to about 800 nanometers), light intensity (e.g., from about 0.001 to about 3000 cd⋅s/ ^m2 ), illumination time (e.g., from about 1 to about 500 milliseconds), time between successive flashes of light (e.g., from about 0.2 to about 50 seconds) with different background wavelengths (e.g., from about 300 to about 800 nanometers) and background illuminance (e.g., from about 0.01 to about 900 cd/ ^m2 ).

[98] The retinal signal processing system 200 may include a sensor 210. The sensor 210 may be configured to detect electrical signals from the retina. The sensor 210 may include one or more electrodes. The sensor 210 may be an electroretinographic sensor. The ground electrode may be located on the skin in the middle of the forehead. The reference electrode for each eye may be located on the earlobes or temporal areas near the eyes or other areas of the skin.

[99] Electrical signals from the retina may be initiated by light stimulation from the light stimulator 205 and collected by the sensor 210 as retinal signal data. The retinal signal data may be collected by the sensor 210, for example, an electrode located on the eyeball or in the periocular regions. Light may initiate a low-amplitude electrical signal generated by the retinal cells of the patient. Depending on the nature of the light (e.g., intensity, wavelength, spectrum, frequency, and duration of flashes) and the conditions of the light stimulation (e.g., background light, dark adaptation, or light of the individual undergoing this process), different electrical signals may be generated since different types of retinal cells will be initiated. This signal propagates in the eye and eventually reaches the visual area of the brain via the optic nerve. However, like any electrical signal, it propagates in all possible directions depending on the electrical conductivity of the tissues. In this way, the electrical signal can be collected in tissues outside the eyeball that are accessible from the outside, such as the conjunctiva.

[100] There are several types of electrodes that can be used to collect the electrical signal. These are based on the specific electrical conductivity and geometry of the wire. It should be understood that a variety of recording electrode designs are possible and that any suitable design or combination of designs of the sensor 210 can be used. The sensor 210 can include a contact lens, gold foil, gold wire, corneal wick, wire loops, microfibers, and/or skin electrodes. Each type of electrode has its own recording characteristics and its own artifacts.

[101] In addition to the sensor 210, the system 200 may also include other devices for monitoring the position of the eye and/or the size of the pupil, which affect the amount of light reaching the retina and, thus, the electrical signal initiated in response to this stimulus. The system 200 may include sensors for recording the wavelength of light, the spectrum of light and/or the intensity of light, such as a spectrometer and/or a photosensor.

[102] An electrical signal may be obtained between an active eye electrode (located on the eye) and a reference electrode, with differential registration from a ground electrode. The electrodes of the sensor 210 may be connected to a data acquisition system 215, which may include a recording device. The data acquisition system 215 may be capable of amplifying electrical signals and/or converting electrical signals into digital signals for further processing. The data acquisition system 215 may implement frequency filtering processes that may be applied to electrical signals from the sensor 210. The data acquisition system 215 may store electrical signals in a database in a voltage format depending on points in time.

[103] The data acquisition system 215 may be configured to receive measured electrical signals of the patient, such as from the sensor 210, and/or data of the stimulating light, such as from the light stimulator 205, and store these collected data as retinal signal data. The data acquisition system 215 may be operatively connected to the light stimulator 205, which may be configured to initiate electrical signals and provide data to the data acquisition system 215. The data acquisition system 215 may synchronize the light stimulation with the capture and recording of the electrical signal.

[104] The collected data may be fed to the data collection system 215 in any suitable manner, such as via a storage device (not shown) and/or a network. The data collection system 215 may be connected to the sensor 210 and/or the light stimulator 205 via a communication network (not shown). The communication network may be the Internet and/or an intranet. Those skilled in the art may suggest several embodiments of the communication network.

[105] The retinal signal data may comprise electrical characteristic data (e.g., voltage and circuit impedance) collected over multiple signal acquisition times (e.g., 5 to 500 milliseconds) at multiple sampling rates (e.g., 0.2 to 24 kHz) with the synchronization time of the light stimulation (flash time) and/or offset (baseline voltage and impedance before the light stimulation). The data acquisition system 215 may collect retinal signal data at frequencies (i.e., sampling rate) from 4 to 16 kHz or higher. This frequency may be higher than in a traditional ERG. The electrical characteristic data may be collected continuously or intermittently.

[106] The data acquisition system 215 may comprise a sensor processor for measuring the impedance of an electrical circuit used to acquire retinal signal data. The impedance of the electrical circuit may be recorded simultaneously with the acquisition of other electrical signals. The collected impedance data may be stored in the retinal signal data. A method for determining the impedance of the circuit simultaneously with the acquisition of electrical signals may be based on the process of injecting a reference signal of known frequency and amplitude through a channel for recording electrical signals. This reference signal may then be filtered separately and processed. By measuring the magnitude of the output at the frequency of the excitation signal, the impedance of the electrode may be calculated. The impedance may then be used as a co-variable to increase the signal density with the resistance of the circuit at each time point in the recording of electrical signals.

[107] The data analysis system 220 may process the data collected by the data acquisition system 215. The data analysis system 220 may extract features and/or descriptors of the retinal signal from the retinal signal data, and/or perform any other processing on the retinal signal data. The data analysis system 220 may receive clinical information of the patient and/or extract cofactors of clinical information from the clinical information.

[108] The prediction output system 225 may receive data from the data analysis system 220 and generate an output for use by a clinician. The output may be an output user interface, a report, or other document, etc. The output may indicate a predicted probability that a patient is susceptible to one or more conditions. For each condition, the output may indicate a predicted probability that the patient is susceptible to that condition. The output may indicate the patient's susceptibility to a pathology. The output may be used by a clinician to assist in determining whether a patient is susceptible to a medical condition and/or determining which medical condition the patient is susceptible to.

[109] The data collection system 215, the data analysis system 220, and/or the prediction output system 225 may be accessed by one or more users, such as through their respective clinics and/or through a server (not shown). The data collection system 215, the data analysis system 220, and/or the prediction output system 225 may also be connected to appointment management software that may schedule appointments or follow-up care based on the determination of a condition by embodiments of the system 200.

[110] The data acquisition system 215, the data analysis system 220, and/or the prediction output system 225 may be distributed among multiple systems and/or combined into a system or multiple systems. The data acquisition system 215, the data analysis system 220, and/or the prediction output system 225 may be geographically distributed.

[111] The systems and methods described herein may comprise: 1) extracting retinal signal data collected from multiple individuals, such as from the memory of a computer system, 2) extracting and/or generating retinal signal features from the retinal signal data, such as voltage, circuit impedance, signal acquisition times, sampling rates, timing and/or offset of light stimulation, and/or any other type of data that can be extracted from the retinal signal data or generated using the retinal signal data, 3) combining the retinal signal features with cofactors from clinical information related to conditions observed in the relevant individual, 4) selecting retinal signal features from the extracted features and the combined features and defining a hierarchy with a ranking depending on the relevance of the retinal signal features, 5) assembling this information into mathematical descriptors, 6) evaluating classifiers from these mathematical descriptors, 7) constructing mathematical domains of classifiers related to these conditions, and/or 8) obtaining density functions from these classifiers.

[112] Mathematical descriptors may be mathematical functions that combine features from the retinal signal data and/or clinical cofactors. Descriptors may indicate a feature of the retinal signal that is characteristic of a condition or population due to additional discrimination between patient groups. Examples of descriptors that may be used include skew, kurtosis, compactness, eigenvectors, centroid coordinates, local binary patterns, time series regression coefficients, spectral entropy, quantum entropy of any form, Renyi entropy, von Neumann entropy, Hartley entropy, Tsallis entropy, uniform entropy, Hu's moments, Haralick features, and/or eigenvalue functions. Descriptors may be used to derive classifiers. Classifiers may be mathematical or statistical functions that use descriptors to map data into a category or class of information by ranking the descriptors according to their statistical significance. Descriptors can be ranked based on their relevance to depicting specific components of a biosignature. Descriptors can be grouped into a descriptor catalog. The descriptor catalog can be used during MLA training.

[113] The systems and methods described herein may comprise: 1) extracting retinal signal data collected from an individual, such as from the memory of a computer system, 2) extracting and/or generating retinal signal features from the retinal signal data, such as voltage, circuit impedance, signal acquisition times, sampling rates, timing and/or offset of light stimulation, and/or any other type of data that can be extracted from the retinal signal data or generated using the retinal signal data, 3) combining the retinal signal features with cofactors from clinical information related to conditions observed in the individual, 4) calculating probabilities that the individual belongs to one or more domains (i.e., is susceptible to the condition), and/or 5) determining the mathematical proximity to those domains for the individual as a predicted probability that the patient is susceptible to the condition corresponding to the domain.

[114] Clinical information may include information indicating the general health status of the individual, such as information regarding comorbidities, medical procedures, previous medical conditions, coffee, alcohol or tobacco use, substance abuse, and/or any other general health status data. Clinical information may include information related to specific psychiatric conditions, such as information from structured questionnaires specific to psychiatric illnesses. These structured questionnaires may include questions related to anxiety, components of emotion and mood, cognitive impairment, feelings, habits, hallucinations, behavioral reactions, and/or other questions related to psychiatric illnesses. Cofactors of clinical information may be extracted from clinical information, such as cofactors of clinical information indicating age, gender, iris color and/or skin pigmentation instead of retinal pigmentation, etc.

[115] The retinal signal data may contain multiple biosignatures, where each biosignature is characteristic of a condition. The biosignature corresponding to a condition for a given patient may be identified using classifiers. The condition may be a psychiatric condition, such as, but not limited to, bipolar disorder, schizophrenia, and depression. The condition may also be a neurological condition, a non-psychiatric condition, or a threat of such a condition. The retinal signal data analysis steps described herein may be applied to the analysis of retinal signal features characteristic of any conditions that are expressed in the retinal signal data.

[116] Graphical representations of the obtained data can be processed and displayed, and can provide visual support for the choices made in selecting retinal signal features for use in mathematical models. Applications can employ mathematical and/or statistical analysis of the results to evaluate the suitability of the data, the robustness of the information, and the accuracy of the results generated by the analysis process.

[117] The retinal signal features extracted from the recorded retinal signal data may include electrical parameters such as circuit voltage and impedance, signal acquisition times (e.g., 5 to 500 milliseconds), sampling rates (e.g., 0.2 to 24 kHz), light stimulation timing (flash time) and offset (baseline voltage and impedance before light stimulation), illuminated retinal areas, and/or other retinal signal features that affect the retinal signal data. The retinal signal features may be generated based on the extracted retinal signal features, such as by performing a mathematical operation on one or more of the extracted retinal signal features. The retinal signal features extracted from the retinal signal data may include data related to the retinal signal, such as the position of the eye, the size of the pupil, the distance from the light source to the eye or part of the eye (pupil, retina) and/or the parameters of the applied illumination (intensity, wavelength, spectrum, frequency of light stimulation, sampling frequency of the retinal signal, wavelength, illumination time, background wavelength, background illumination). The retinal signal features may be voltage independent and/or time independent. All or part of these features may be analyzed using the systems and methods described herein.

Methods of processing retinal signal data

[118] Fig. 3 shows an exemplary functional architecture of a method 300 for processing information that results in constructing mathematical functions for predicting whether a patient is susceptible to a condition. The method 300 may be used to model mathematical domains of information from retinal signal data collected from patients susceptible to conditions. The method 300 may be performed, in whole or in part, at an information input step 310, a feature extraction step 320, a feature selection and placement step 330, a feature weighting and assembly step 340, a classifier ranking and discrimination step 350, a matching step 360, and/or a prediction output step 370.

[119] Retinal signal data and collected clinical information can be processed differently depending on the level and specificity of the information. Retinal signal features can be classes or categories of indicators attached to attribute variables by explanatory variables. Descriptors can be relevant features characteristic of a condition or population that are determined after a discrimination process. Classifiers can be mathematical or statistical functions that use descriptors to map data into a category or class after a ranking process. Regions can be identified as a subset of ranges of mathematical functions or a collection of functions constructed with classifiers. A domain can be identified as a region characteristic of a condition.

[120] It should be clearly understood that the functional architecture depicted in Fig. 3 is merely an illustrative implementation of the present invention. Accordingly, the following description provides merely illustrative examples of the present invention. This description does not serve to define the scope or limits of the present invention.

[121] In a number of instances, supposedly useful examples of modifications to the functional architecture shown in Fig. 3 may also be set forth below. This is done merely as an aid to understanding, and again is not intended to define the scope or limits of the present invention. This list of modifications is not exhaustive, and other modifications are possible, as will be understood by those skilled in the art. Additionally, where this is not done (i.e., where no example of modifications is given), it should not be assumed that no modifications are possible and/or that the only way to implement this element of the present invention has been described. As will be understood by those skilled in the art, this is likely not the case. Furthermore, it should be understood that the functional architecture shown in Fig. 3 may provide, in a number of instances, a simple implementation of the present invention, and if so, they have been presented in this manner as an aid to understanding. As will be understood by those skilled in the art, various implementations of the present invention may be more complex.

[122] In one or more aspects, the method 300 or one or more of its steps may be performed by a computing system, such as the computing environment 100. The method 300 or one or more of its steps may be implemented in computer-executable instructions that are stored on a computer-readable medium, such as a mass storage device, loaded into a memory, and executed on a CPU. The method 300 is illustrative, and it should be understood that some steps or a portion of the steps in the diagram may be omitted and/or rearranged in a different order.

[123] In some embodiments, the ability to specifically model a state mathematically is supported by a graphical representation of state-specific information domains that were discussed during the demonstrations and then used to describe the concept of mathematical domains. The results of the robustness and accuracy analyses are used to demonstrate the relevance and specificity of the mathematical domains constructed according to the present invention.

[124] By extending the analysis strategy to additional computational processes, followed by generation of additional mathematical descriptors, the methods for deciphering the biosignature included in the retinal signal data can be enriched.

Entering information

[125] At step 311, the input of information may be performed using the collected retinal signal data with or without the removal of artifacts, which may include distorted signals, noise, and/or artifacts of any other type. Artifacts may arise from one or more of the following: unintentional capture of electrical signals not originating from the retina, shifts in the placement of the eye electrode, changes in the contact of the ground or reference electrode, blinking and/or eye movements.

[126] At step 312, the collected retinal signal data may be transformed, for example, by transposing the retinal signal data to another one-dimensional or multi-dimensional scale, such as scaling, shifting, transforming with elementary functions (e.g., logarithmic, polynomial, power, trigonometric), time series (resulting in a change in the shape of the distribution), wavelet transforms (resulting in scalegrams), empirical mode decomposition (resulting in intrinsic mode decomposition functions (IMF)), gradient transforms (resulting in a vector) and/or kernel decomposition (resulting in a change in the shape of the distribution and/or the length scales and variances of the signal). Filtering (finite impulse response, infinite impulse response) may be performed on the retinal signal data, such as a high-pass filter, a low-pass filter, a band-pass filter, notch filters. Hilbert-Huang transform (instantaneous frequency IMF).

[127] At step 313, the clinical information may be transformed, such as by transposing the clinical data to another unidimensional or multidimensional scale, such as scaling, shifting, transforming using elementary functions (e.g., logarithmic, polynomial, power, trigonometric), and/or regrouping variables into composite variables.

[128] At step 314, the suitability of the data for the data generated at step 310 of inputting information may be tested and confirmed, such as by ensuring that all data to be processed complies with a standardized format suitable for the upcoming processing.

Feature extraction

[129] In step 321, the features of the retinal signal that are suspected of containing components of the biosignature may be obtained, for example, by time-frequency analysis (magnitudes of a particular position on scalegrams, positions of minima and maxima of the magnitude), kernel decomposition, principal component analysis (PCA), geometric (algebraic) operations in various time-frequency intervals, such as min (e.g., a-wave), max (e.g., b-wave), latency, slope, gradient, curvature, integral, energy (sum of squared amplitudes), variance, cohesion-variance (homogeneity, density) and/or any other methods of extracting or generating features of the retinal signal from the retinal signal data.

[130] At step 322, potential cofactors of clinical information that are suspected of influencing the components of the biosignature may be obtained from clinical information, such as by multivariate analysis (chi-square combined with variances) and subsequent clustering, forward selection based on stepwise regression, best subset regression (using a specified set of cofactors), from clinical practice (i.e., cofactors to a condition, such as duration of illness, number of crises/hospitalizations), and/or any other methods for obtaining cofactors of clinical information from clinical information.

[131] At step 323, the robustness of the information and the accuracy of the results generated at step 320 of feature extraction may be assessed.

Selecting and Placing Features

[132] In step 331, the features of the retinal signal that contain the most significant components of the biosignature may be selected, for example, by time-frequency visual analyses or scalegrams (in this case, the features of the retinal signal are the absolute values of the signal in some time-frequency windows that have proven to be discriminant), stepwise regression with cross-validation, cofactor-adjusted or unadjusted sparse representation classification (SRC) with predetermined thresholds, selection and combination of the most relevant features of the retinal signal to generate principal components, for example, in a supervised PCA (SPCA) process, least absolute shrinkage and selection operator (LASSO), ridge regression, flexible network, Bayesian or "spike and slab" methods and/or any other selection method.

[133] At step 332, a hierarchy of cofactors (confounding, mediator, moderator) may be determined, i.e., the direction of influence and hierarchy of the most important features of the retinal signal and cofactors of clinical information as components of the biosignature that together make a major contribution to the analysis models may be assessed, such as by means of i) confounding methods (randomization, constraint, and matching); stratification accompanied by the Mantel-Haenszel estimator; and/or multivariate methods, such as ANCOVA, linear and logistic regressions, and/or ii) mediation and moderation methods: such as the Baron and Kenny method; the Fairchild and MacKinnon method, and/or other suitable methods.

[134] At step 333, clinical information cofactors with high contributions to the mathematical models may be selected from the clinical information, i.e., those clinical information cofactors that influence the components of the biosignature and collectively contribute highly to the mathematical models (thus becoming ‘qualified’ cofactors).

[135] At step 334, the stability of the information and the accuracy of the results generated at step 330 of selecting and placing features may be assessed.

Weighing and assembling features

[136] At step 341, relevant features of the retinal signal may be restructured and/or weighted to generate descriptors. For example, the contribution of features of the retinal signal may be determined and/or displayed. Features of the retinal signal that contain biosignature descriptors and influence models may be identified, for example, by analyzing multivariate regression or related processes.

[137] At step 342, the descriptors may be assembled with the most significant cofactors of the clinical information (i.e., the qualified cofactors). The descriptors may then be selected to obtain the components of the biosignature that together contribute most to the mathematical models, such as by matching when merging the descriptors and cofactors using mathematical expressions or relationships, using, for example, PCA, SPCA, or other methods used in selecting and/or combining features of the retinal signal data.

[138] At step 343, the robustness of the information and the accuracy of the results generated at step 340 of weighting and assembling features may be assessed.

Ranking and discrimination of classifiers

[139] At step 351, the classifiers may be evaluated using the collected descriptors and/or cofactors of the clinical information to train the classifiers. The classifiers may be selected and/or ranked based on their performance, such as by logistic regressions, probit regressions, probabilistic modeling (Gaussian process, kernel estimation models, Bayesian modeling), best subset regression (using a specified set of predictors), SVM, neural network methods, decision tree, random forest, weighted voting, boosting and bagging, Kaplan-Meier analysis, Cox regression, and/or other selection or ranking methods.

[140] At step 352, zones and domains may be constructed by selecting and mapping state-specific classifiers to mathematical functions that represent their most specific biosignature, such as by visualizing regression results enhanced by mathematical constructs (e.g., wavelets, Kullback-Leibler divergence, etc.), according to the Neyman-Pearson criterion for discrimination between domains and/or other related methods.

[141] At step 353, the robustness of the information and the accuracy of the results generated at step 350 of ranking and discriminating classifiers may be assessed.

Comparison of the obtained data from the decoding of the retinal signal

[142] At step 361, density functions may be obtained and compared, i.e., by obtaining mathematical functions that contain the highest density of biosignature components from the retinal signal data and clinical cofactors and comparing these functions between states, such as by a histogram; kernel density estimation (e.g., Parzen-Rosenblatt windows, band selection), characteristic function density, and other relevant estimators, data clustering methods including vector quantization, reconstruction methods, or based on the cumulative probability distribution of the sample or based on sample moments, and/or other methods.

Completion of retinal signal decoding

[143] At step 371, the probabilities may be calculated from mathematical functions that contain the corresponding density of the biosignature components from the retinal signal data and the clinical information cofactors as compared to those obtained in different states, such as using a density function; by estimating a CPF (cumulative probability function) and/or other related methods.

[144] At step 372, the distances from the zones may be determined by identifying the components of the biosignature that are most relevant to the distances between zones or high-density domains in different states and computing the distances as mathematical expressions, such as by comparing probabilities with Bayesian statistical prior distributions. Entropy methods, the Kullback-Leibler divergence parameter for estimating the direction of change, and the Neyman-Pearson criterion for estimating distances between zones. The semiparametric maximum likelihood estimation procedure may use the adjacent intruder pooling (PAVA) algorithm, as well as other methods of testing the medical state assigned to each individual part of the model (i.e., zones) and re-evaluating their membership in that model using IPWE (inverse probability weighting estimator) or related methods.

[145] At step 373, the stability of the information and the accuracy of the results generated at completion step 370 may be assessed.

Methods of prediction

[146] Fig. 4 shows a flow chart of operations of a method 400 for predicting the likelihood that a patient is susceptible to a condition in accordance with various embodiments of the present invention. All or part of the method 400 may be performed by the data acquisition system 215, the data analysis system 220, and/or the prediction output system 225. In one or more aspects, the method 400 or one or more of its steps may be performed by a computing system, such as the computing environment 100. The method 400 or one or more of its steps may be implemented in computer-executable instructions that are stored on a computer-readable medium, such as a mass storage device, loaded into a memory, and executed on a CPU. The method 400 is illustrative, and it should be understood that some steps or portions of the steps in the flow chart may be omitted and/or rearranged in a different order.

[147] The method 400 comprises performing various actions, such as extracting retinal signal features from retinal signal data, selecting retinal signal features that are most relevant to particular conditions, combining and comparing these retinal signal features to generate mathematical descriptors that are most discriminant to the conditions to be analyzed or compared, generating a multimodal display, identifying biomarkers and/or biosignatures of the conditions, and/or predicting the likelihood that a patient is susceptible to any of the conditions, as will be described in more detail below.

[148] At step 405, retinal signal data may be captured from a patient. The retinal signal data may be captured using a predetermined acquisition protocol. The retinal signal data may include measured electrical signals captured by electrodes placed on the patient. The retinal signal data may include system parameters used to capture the retinal signal data, such as light stimulation parameters. The retinal signal data may include an impedance of a receiving electrical circuit used in a device that measures electrical signals. In some embodiments, this step 405 is omitted.

[149] The retinal signal data may include impedance measurements and/or other electrical parameters. The retinal signal data may include optical parameters, such as changes in pupil size and/or parameters of the applied illumination (intensity, wavelength, spectrum, frequency of light stimulation, sampling frequency of the retinal signal).

[150] To generate retinal signal data, the retina of the patient may be stimulated, for example, using a light stimulator 205, which may be one or more optical stimulators. The retinal signal data may be collected by a sensor, for example, a sensor 210, which may include one or more electrodes and/or other sensors.

[151] The light stimulator may comprise any light sources capable of generating light beams that differ in wavelength (e.g., from about 300 to about 800 nanometers), light intensity (e.g., from about 0.01 to about 3000 cd⋅s/ ^m2 ), illumination time (e.g., from about 1 to about 500 milliseconds), time between successive flashes of light (e.g., from about 0.2 to about 50 seconds) with different background wavelengths (e.g., from about 300 to about 800 nanometers) and background illuminance (e.g., about 0.1 to about 800 cd/ ^m2 ).

[152] The retinal signal data may comprise electrical characteristic data (e.g., voltage and circuit impedance) collected over multiple signal collection times (e.g., 5 to 500 milliseconds) at multiple sampling rates (e.g., 0.2 to 24 kHz) with a light stimulation synchronization time (flash time) and an offset (baseline voltage and impedance before light stimulation). Thus, step 405 may comprise collecting retinal signal data at frequencies from 4 to 16 kHz.

[153] After collecting the retinal signal data, such as by a medical practitioner, the retinal signal data may be uploaded to a server, such as a data analysis system 220, for analysis. The retinal signal data may be stored in the memory 130 of the computer system.

[154] At step 410, if step 405 is omitted, retinal signal data may be extracted from memory 130. Retinal signal features may be extracted from the retinal signal data. Extraction of retinal signal features may be based on processing the retinal signal data and/or transforming it using several signal analysis methods, such as polynomial regressions, wavelet transforms and/or empirical mode decomposition (EMD). Extraction of retinal signal features may be based on parameters derived from these analyses or specific modeling, such as principal components and most discriminant predictors, parameters from linear or nonlinear regression functions, higher frequency, Kullback-Leibler distance of the difference, features of Gaussian kernels, log-likelihood of the difference and/or high energy regions. These analyses may be used to determine the contribution of each specific feature of the retinal signal and statistically compare features of the retinal signal.

[155] The features of the retinal signal to be extracted may be determined in advance. The features of the retinal signal to be extracted may be determined by analyzing labeled data sets of retinal signal data for multiple patients. Each patient represented in the data sets may have one or more associated medical conditions to which the patient is susceptible and/or one or more medical conditions to which the patient is not susceptible. These medical conditions may be a label of each patient's data set. By analyzing a set of retinal signal data from patients sharing a medical condition, the features of the retinal signal to be extracted may be determined. A multimodal map may be generated based on the features of the retinal signal. Domains may be defined based on the multimodal map.

[156] At step 415, descriptors may be extracted from the retinal signal features. Mathematical descriptors may be mathematical functions that combine features from the retinal signal data and/or clinical cofactors. The descriptors may indicate a retinal signal feature that is characteristic of a condition or population due to additional discrimination between patient groups. As described above at step 342 of the method 300, descriptors may be selected to obtain components of the biosignature that together contribute the most to mathematical models, such as by matching when merging descriptors and cofactors using mathematical expressions or relationships, using, for example, PCA, SPCA, or other methods used in selecting and/or combining features of the retinal signal data.

[157] At step 420, clinical information of the patient may be received. The clinical information may include medical records and/or any other collected data regarding the patient. The clinical data may include the results of a survey and/or clinical examination by a healthcare practitioner.

[158] At step 425, clinical information cofactors may be generated using clinical information. The clinical information cofactors may be selected based on their impact on the retinal signal data. The clinical information cofactors may include indications of the patient's age, gender, skin pigmentation, which may be used in place of retinal pigmentation, and/or any other clinical information relevant to the patient.

[159] At step 430, the clinical information cofactors and/or descriptors may be applied to mathematical models of conditions. Any number of mathematical models may be used. The clinician may use mathematical models of his or her choice. Each model may correspond to a specific condition or management.

[160] At step 435, each model may determine the distance between the patient and the biosignature of the model state. The principal components of the retinal signal data may be located in domains corresponding to the states. Descriptors and/or cofactors of the clinical information may be compared with the biosignature of each model.

[161] At step 440, each model may output a predicted probability that the patient is susceptible to the model condition. The probability that the patient is susceptible to the condition may be predicted based on the level of statistical significance in comparing the magnitude and position of the individual's descriptors to the model's. The predicted probability may be binary and indicate whether the biosignature of the condition is present or absent in the patient's retinal signal data. The predicted probability may be a percentage indicating how likely it is that the patient is susceptible to the condition.

[162] At step 445, the predicted probability that the patient is susceptible to each condition may be output. An interface and/or a report may be output. The interface may be output on a display. The interface and/or a report may be output to a clinician. The output may indicate the probability that the patient is susceptible to one or more conditions. The output may indicate the patient's susceptibility to the pathology. The predicted probabilities may be stored.

[163] The output may include a definition of the medical condition, a predicted probability of the medical condition, and/or a degree of agreement between the patient's retinal signal data and the condition and/or other conditions. The predicted probability may be expressed as a percentage of agreement for the medical condition, which may provide an objective neurophysiological measure to further assist the clinician in medical diagnosis.

[164] The output may be used in conjunction with the clinician's preliminary medical diagnosis to increase the comfort level with determining the clinician's medical condition and/or to initiate an earlier or more effective treatment plan. The output may be used to initiate treatment earlier rather than spending additional time clarifying the medical condition and treatment plan. The output may reduce the clinician's and/or the patient's level of uncertainty about the clinician's preliminary medical diagnosis. The output may be used to select a medication to be prescribed to the patient. The selected medication may then be prescribed to the patient.

[165] The method 400 may be used to monitor a patient's condition. The patient's condition may be diagnosed in advance. The method 400 may be used to monitor the progress of the condition. The method 400 may be used to monitor and/or change a treatment plan for the condition. For example, the method 400 may be used to monitor the effectiveness of a drug used to treat the condition. Retinal signal data may be collected before, during, and/or after the patient undergoes treatment for the condition.

[166] The method 400 may be used to identify and/or track neurological symptoms of an infection, such as a viral infection. For example, the method 400 may be used to identify and/or track neurological symptoms of patients infected with COVID-19. Retinal signal data may be collected from patients who are or have been infected with COVID-19. The retinal signal data may be evaluated using the method 400 to determine whether a patient is suffering from neurological symptoms, the severity of the neurological symptoms, and/or to develop a treatment plan for the neurological symptoms.

Generating mathematical models

[167] Fig. 5 shows a flow chart of operations of a method 500 for generating a mathematical model for predicting whether a patient is susceptible to a condition in accordance with various embodiments of the present invention. All or part of the method 500 may be performed by a data acquisition system 215, a data analysis system 220, and/or a prediction output system 225. In one or more aspects, the method 500 or one or more of its steps may be performed by a computing system, such as the computing environment 100. The method 500 or one or more of its steps may be implemented in computer-executable instructions that are stored on a computer-readable medium, such as a mass storage device, loaded into a memory, and executed on a CPU. The method 500 is illustrative, and it should be understood that some steps or portions of the steps in the flow chart may be omitted and/or rearranged in a different order.

[168] The method 500 can process data sets of retinal signal data and/or other clinical information to create a classification model based on domains characteristic of various conditions. Mathematical modeling of the retinal signal data can be performed using several sequential analyses that process the retinal signal data and/or clinical information cofactors to generate classification metrics based on domains that are characteristic of the conditions. The analysis has several mechanisms that combine the retinal signal data and clinical information cofactors, such as specific descriptors in retinal biosignatures, to select the most discriminant components as descriptors. These descriptors, which can be combined with clinical information cofactors, can be used to train classifiers to subsequently select them based on their performance while providing a probability factor for the patient's susceptibility to the condition. The descriptors can be used to construct density functions characteristic of each of the conditions. Domain clusters can be formed using multimodal analysis of discriminant features, principal components, and mappings of descriptors of high statistical significance.

[169] At step 505, retinal signal data may be collected from patients. The collected retinal signal data may comprise impedance measurements and/or other electrical parameters. The retinal signal data may comprise optical parameters, such as the distance from the light source to the eye or part of the eye (pupil, retina), changes in pupil size, and/or parameters of the applied illumination (light intensity, light wavelength, light spectrum, light stimulation frequency, retinal signal sampling frequency). In some embodiments, step 505 may be omitted.

[170] At step 510, each retinal signal data of the patient may be labeled with a condition. The label may be one or more conditions that may be diagnosed in the patient. The label may indicate that the patient is not diagnosed with any of the conditions, in which case the patient may be referred to as a control subject. The condition may be schizophrenia, bipolar disorder, major depressive disorder, psychosis, post-traumatic stress disorder, seizure, substance abuse, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, attention deficit disorder, and/or any other condition. If step 505 is omitted, step 510 may comprise retrieving stored retinal signal data from a memory, such as memory 130 of computing environment 100.

[171] At step 515, clinical information cofactors for patients may be received. The clinical information cofactors may be extracted from medical records. The clinical information cofactors may include indications of population parameters, such as age, gender, and/or skin pigmentation instead of retinal pigmentation, iris color, etc. The clinical information cofactors may be received for all or a subset of patients for whom retinal signal data was collected.

[172] At step 520, a data set may be generated for each patient. The data set may include retinal signal data of the patient, a label or labels assigned to the patient, and/or cofactors of the patient's clinical information. The data set may be stored in a database and/or saved in any other suitable format.

[173] At step 525, retinal signal features may be extracted from retinal signal data in data sets. Any suitable mathematical selection process may be used to select retinal signal features for extraction, such as principal component analysis (PCA), generalized linear models (GLM), sparse representation classification (SRC), least absolute shrinkage and selection operator (LASSO), and/or a combination of several models designed to transform the selected retinal signal features and generate descriptors of interest in terms of cohesion and density (information density) in order to select the most discriminative retinal signal features and generate statistical significance displays. Various methods can be used to analyze the collected retinal signal data, such as, but not limited to, time-frequency analysis, prototype mother wavelets generated from reference data sets, probabilistic modeling, information cohesion and dispersion, principal component analysis (PCA), which can be supervised or unsupervised, generalized linear models (GLM), sparse representation classification (SRC), least absolute shrinkage and selection operator (LASSO) and/or similar selection methods or a combination of several methods. Mappings can be used to delineate regions of the retinal signal data (n-dimensional) that have the greatest discriminant power to separate the domain of each state and the position of the individual in these domains (distances).

[174] A state biosignature may contain portions of the retinal signal data that are specific to the state. One or more of the most discriminative features of the retinal signal may be selected as descriptors of some state. The descriptors may be combined in a multimodal computation.

[175] The possibility of concrete mathematical modeling is supported by a graphical representation of the domains of information characteristic of the states that were observed during the demonstrations and then used to describe the domains. Clusters of relevant predictors can be specified between the features of the retinal signal obtained using multimodal analysis.

[176] By extending the analysis strategy to additional computational processes, followed by generation of additional retinal signal features, the components used to decipher the biosignature included in the retinal signal data can be enriched.

Frequency (spectral) analysis

[177] Time-frequency analysis can be performed using mother wavelets, such as discrete wavelet transforms (DWT) or continuous wavelet transforms (CWT), and/or empirical mode decomposition (EMD) filtering. Commonly used features such as higher frequency, time of occurrence, frequency mapping (scalegrams), time-frequency domains, wavelet coefficient as the output of the wavelet transform, relative frequency of occurrence of the wavelet coefficient, and/or composite parameters can be visualized and/or calculated.

[178] Specific mother wavelets (prototype wavelets) can be constructed and tuned to the nature of the data (either control groups or a group of individuals with specific pathologies). Specific mother wavelets can be validated (wavelet construction) with preprocessing and filter frequencies that are not considered as significant features. This strategy can be used to tune a known pre-defined mother wavelet and perform more specific time-frequency analysis with or without clustering time (i.e. time-frequency analysis with a mother wavelet tuned to the portion of the retinal waveform being analyzed).

[179] Applications can be used to compare analyses and demonstrate differences between retinal signal data from patients with conditions, such as psychiatric conditions, and retinal signal data from individuals not diagnosed with these conditions (control subjects).

Principal Component Analysis

[180] Principal component analysis (PCA) can be performed. PCA can be supervised (SPCA) or unsupervised (PCA) to extract the most discriminative retinal signal features from the retinal signal data. PCA can identify so-called “principal components”, which may be the retinal signal features with the maximum dependence on the feature variables, in this case the suspicious states. PCA can analyze the dependence between the retinal signal features and the states and allow the selection of these retinal signal features with the maximum dependence.

[181] Retinal signal features can be analyzed in conjunction with additional demographic or clinical variables characteristic of associated pathologies or medical conditions. These analyses can yield specific data vectors (principal component scores) used to define the retinal signal features, which can then be combined and reanalyzed (“matched”) to produce a classification of the most discriminatory retinal signal features based on their statistical significance at a pre-specified threshold. Such a classification can allow the selection of retinal signal features that can be used to construct vectorized domains of specific conditions.

Probabilistic modeling

[182] Probabilistic modeling can be used to find regions of interest in the retinal signal data. A standard Gaussian and/or non-Gaussian modeling process can be used to assign probabilities to events in the collected data set. The features of the retinal signal selected for building models (for training the model) can be selected based on data sets that are specific to retinal signal data from patients with specific conditions and controls (data sets of patients who are not exposed to conditions).

[183] At step 530, a subset of retinal signal features that correspond to the biomarker of the state may be selected. The retinal signal features may be ranked based on their relevance to the state. Some or all of the highest ranked retinal signal features may be selected in the subset.

[184] At step 535, a classifier may be determined that distinguishes the biosignature of the condition from the biosignature of other conditions. The classifier may include a subset of retinal signal features and/or cofactors of clinical information. The classifier may be a mathematical model that specifies the biosignature. To predict the likelihood that a patient is susceptible to the condition, a distance between the patient's retinal signal data and the condition classifier may be determined.

[185] Fig. 6 shows a flow chart of a method 600 for training a machine learning algorithm (MLA) to predict the likelihood of a medical condition, such as a psychiatric condition or a neurological condition, in accordance with various embodiments of the present invention. All or part of the method 600 may be performed by the data acquisition system 215, the data analysis system 220, and/or the prediction output system 225. In one or more aspects, the method 600 or one or more of its steps may be performed by a computing system, such as the computing environment 100. The method 600 or one or more of its steps may be implemented in computer-executable instructions that are stored on a computer-readable medium, such as a mass storage device, loaded into a memory, and executed on a CPU. The method 600 is illustrative, and it should be understood that some steps or portions of the steps in the flow chart may be omitted and/or rearranged in a different order.

[186] At step 605, data sets may be extracted from the retinal signal data. The data sets may be generated at steps 505-520 of the method 500, as described above. Each data set may include retinal signal data of a patient, one or more labels corresponding to conditions diagnosed in the patient, and clinical information cofactors corresponding to the patient. The retinal signal data may correspond to multiple patients. The retinal signal data may be selected based on a specific patient population. For example, all retinal signal data corresponding to a specified gender and age range may be extracted.

[187] The extracted data sets may contain impedance measurements and/or other electrical parameters. The extracted data sets may contain optical parameters, such as pupil size changes and/or parameters of the applied illumination (light intensity, wavelength, spectrum, frequency of light stimulation, sampling frequency of the retinal signal). The extracted data sets may contain population parameters, such as age, gender, iris color and/or skin pigmentation instead of retinal pigmentation, etc.

[188] The retinal signal data may be labeled. For each patient represented in the data, one or more conditions diagnosed in the patient may be listed. For patients who have not been diagnosed with any of the available conditions, no condition may be listed, or a label may be provided indicating that the patient is a control subject.

[189] At step 610, retinal signal features may be generated. Retinal signal features may be generated using a transform process. Retinal signal features may be generated for each set of retinal signal data extracted at step 605.

[190] At step 615, descriptors may be extracted from features of the retinal signal in the data arrays. The descriptors may be features selected as the most representative states.

[191] At step 620, the descriptors may be ranked. Each descriptor may be ranked based on the level of statistical significance of the descriptor when testing the contribution of the descriptor to discrimination between conditions and/or between a condition and a control (i.e., in the absence of a diagnosed condition).

[192] At step 625, the descriptors may be selected based on their ranks. A predetermined number of highest-ranking descriptors may be selected.

[193] At step 630, the MLA may be trained using the selected descriptors. The data sets extracted at step 605 may be filtered to remove descriptors other than the descriptors selected at step 625. All or a portion of the data sets may then be used to train the MLA. The MLA may be trained to predict, based on a set of descriptors corresponding to a patient, the likelihood that the patient is susceptible to the condition.

[194] After training the MLA, the MLA may be used to predict, based on measured retinal signal data, the likelihood that a patient is susceptible to a condition. Fig. 7 shows a flow chart of a method 700 for using the MLA to predict the likelihood of a psychiatric condition in accordance with various embodiments of the present invention. All or part of the method 700 may be performed by the data acquisition system 215, the data analysis system 220, and/or the prediction output system 225. In one or more aspects, the method 700 or one or more steps thereof may be performed by a computing system, such as the computing environment 100. The method 700 or one or more steps thereof may be implemented in computer-executable instructions that are stored on a computer-readable medium, such as a mass storage device, loaded into memory, and executed on a CPU. The method 700 is illustrative and it should be understood that some steps or portions of steps in the flow chart may be omitted and/or rearranged in a different order.

[195] At step 705, the patient's retinal signal data may be collected. The actions performed at step 505 may be similar to those described above in connection with step 405.

[196] At step 710, features and/or descriptors of the retinal signal may be extracted from the retinal signal data. The descriptors may be extracted in accordance with an MLA, such as an MLA generated using the method 600.

[197] At step 715, descriptors may be input into the MLA. Cofactors of the patient's clinical information may also be input into the MLA. The MLA may be configured to predict the likelihood that the patient is susceptible to the condition based on the descriptors and/or cofactors of the clinical information. The MLA may be an MLA generated using the method 600.

[198] At step 720, a determination may be made whether additional data needs to be input into the MLA. The MLA may determine that the input of data at step 715 is insufficient to make a prediction. The MLA may determine that the amount of data is insufficient. The MLA may determine that the data is not accurate enough to make a prediction, such as if there are errors in the data. The MLA may output a confidence level of the prediction, and a determination may be made that the confidence level is below a threshold confidence level, such as a predetermined threshold confidence level. If the MLA requires additional data, the additional data may be captured at step 705. In other words, if the MLA has a sufficient amount of data to make a prediction, the method 700 may proceed to step 725.

[199] At step 725, the MLA may output a predicted probability that the patient is susceptible to one or more conditions. For each condition, the MLA may output a predicted probability that the patient is susceptible to the condition. The patient's susceptibility to the pathology may be output by the MLA and/or may be determined based on the predictions output by the MLA. The patient's susceptibility may indicate the patient's distance from the biosignature of the condition. A user interface and/or a report may be generated and output to a clinician for use as an aid in determining the patient's medical condition and/or confirming the medical condition.

[200] Examples for time-frequency analysis and selection of discriminant regions based on statistical significance of higher frequency are illustrated in Figs. 8-27. In these examples, retinal signal data from patients with schizophrenia, bipolar disorder, or major depressive disorder, a youth population at risk for psychosis (i.e., young people whose parents have schizophrenia, bipolar disorder, or depression), and from control subjects not affected by these conditions were time-frequency analyzed (wavelet transform with different mother wavelets) and compared. Time-frequency analysis is performed with any known mother wavelet or with a specific mother wavelet prepared (prototype mother wavelets) from reference data sets (control subject or individuals with specific conditions).

[201] Examples for selecting and displaying statistical significance of retinal signal features are illustrated in Figs. 28-42. In these examples, retinal signal features obtained by frequency analysis of the signal from patients with schizophrenia, bipolar disorder, or major depressive disorder, a youth population at risk for psychosis (i.e., young people whose parents suffer from schizophrenia, bipolar disorder, or depression), and from control subjects not affected by these conditions were selected and compared. Displays of the most discriminatory features were then prepared from the selection of features.

[202] Examples for selecting and displaying the statistical significance of retinal signal descriptors are illustrated in Figs. 43-58. In these examples, retinal signal descriptors obtained from patients with schizophrenia, bipolar disorder, or major depressive disorder, a youth population at risk for psychosis (i.e., young people whose parents have schizophrenia, bipolar disorder, or depression), and from control subjects not affected by these conditions were pooled and their potential for discrimination across conditions was assessed. From this selection, displays of the most discriminatory descriptors were then prepared based on their discriminatory potential and their position in the information volume.

[203] Figure 8 shows a comparison of frequency analysis of retinal signal data from patients with schizophrenia (lighter dots) and individuals not diagnosed with mental health disorders (control subjects; darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[204] Figure 9 shows a comparison of frequency analysis of retinal signal data from patients with schizophrenia (lighter dots) and individuals not diagnosed with mental health disorders (control subjects; darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[205] Figure 10 shows a comparison of frequency analysis of retinal signal data from patients with bipolar disorder (lighter dots) and individuals not diagnosed with mental health disorders (control subjects; darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[206] Figure 11 shows a comparison of frequency analysis of retinal signal data from patients with bipolar disorder (lighter dots) and individuals not diagnosed with mental health disorders (control subjects; darker dots) using a discrete mother wavelet approximation of a particular design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[207] Figure 12 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and individuals not diagnosed with mental health disorders (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[208] Figure 13 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and from individuals not diagnosed with mental health disorders (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[209] Figure 14 shows a comparison of frequency analysis of retinal signal data from patients at risk of psychosis (lighter dots) and from individuals not diagnosed with mental health disorders (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[210] Figure 15 shows a comparison of frequency analysis of retinal signal data from patients at risk of psychosis (lighter dots) and from individuals not diagnosed with mental health disorders (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[211] Figure 16 shows a comparison of frequency analysis of retinal signal data from patients with schizophrenia (lighter dots) and from patients with bipolar disorder (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[212] Figure 17 shows a comparison of frequency analysis of retinal signal data from patients with schizophrenia (lighter dots) and from patients with bipolar disorder (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[213] Figure 18 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and from patients with bipolar disorder (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[214] Figure 19 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and from patients with bipolar disorder (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[215] Figure 20 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and from patients with schizophrenia (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[216] Figure 21 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and from patients with schizophrenia (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[217] Figure 22 shows a comparison of frequency analysis of retinal signal data from patients with bipolar disorder (lighter dots) and from patients at risk of psychosis (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[218] Figure 23 shows a comparison of frequency analysis of retinal signal data from patients with bipolar disorder (lighter dots) and from patients at risk of psychosis (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[219] Figure 24 shows a comparison of frequency analysis of retinal signal data from patients with schizophrenia (lighter dots) and from patients at risk of psychosis (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[220] Figure 25 shows a comparison of frequency analysis of retinal signal data from patients with schizophrenia (lighter dots) and from patients at risk of psychosis (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[221] Figure 26 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and from patients at risk of psychosis (darker dots) using a discrete approximation of the Morlet waveform as the mother wavelet. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[222] Figure 27 shows a comparison of frequency analysis of retinal signal data from patients with major depressive disorder (lighter dots) and from patients at risk of psychosis (darker dots) using a discrete approximation of a mother wavelet of a certain design with retinal signal data sets from control subjects. The magnitude of the SRC coefficients as a function of time and frequency is 90% of the threshold.

[223] Figure 28 shows the most discriminant predictors from a selection of features (the frequency analysis components in this example) performed with sparse representation classification (SRC) at a wavelet coefficient threshold of p≤0.05 in patients with bipolar disorder compared to patients with schizophrenia using a discrete approximation of a mother wavelet of a particular design with retinal signal data sets from control subjects. The colors represent scales of the magnitude of the SRC at a particular position. Darker dots depict SRC positions where the SRC exceeds a given threshold (in the illustrated example, the threshold is from 70% to 100%).

[224] Figure 29 shows the most discriminant predictors from a selection of features (frequency analysis components in this example) performed with classification by principal component analysis based on statistical significance with a statistical significance threshold of p≤0.05 in patients with bipolar disorder compared to patients with schizophrenia, using a mother wavelet of a specific design with retinal signal datasets from control subjects.

[225] Figure 30 shows the displays of the most discriminant predictors and their significance levels (statistical p-values) from a selection of features (frequency analysis components in this example) performed with classification by principal component analysis ranked by statistical significance in patients with bipolar disorder versus patients with schizophrenia, using a mother wavelet of a particular design with retinal signal data sets from control subjects. In this example, the two most discriminant components were considered with a threshold of p≤0.05. The colors (in gray scale) are scales of the magnitude of discriminant power.

[226] Figure 31 shows the most discriminant predictors from a selection of features (the frequency analysis components in this example) performed with sparse representation classification (SRC) at a wavelet coefficient threshold of p≤0.05 in patients with bipolar disorder compared to patients with major depressive disorder, using a discrete approximation of a mother wavelet of a particular design with retinal signal data sets from control subjects. The shading illustrates the scales of the SRC magnitude at a particular position. Darker dots depict SRC positions where the SRC exceeds a given threshold (in the illustrated example, the threshold is from 70% to 100%).

[227] Figure 32 shows the most discriminant predictors from a selection of features (frequency analysis components in the illustrated example) performed with classification by principal component analysis based on statistical significance with a statistical significance threshold of p≤0.05 in patients with bipolar disorder compared to patients with major depressive disorder, using a mother wavelet of a specific design with retinal signal data sets from control subjects.

[228] Figure 33 shows the displays of the most discriminant predictors and their significance levels (statistical p-values) from a selection of features (frequency analysis components in this example) conducted with classification by principal component analysis ranked by statistical significance in patients with bipolar disorder compared to patients with major depressive disorder, using a mother wavelet of a particular design with retinal signal data sets from control subjects. In the illustrated example, the two most discriminant components were considered with a threshold of p≤0.05. The shading in the gray scale indicates the scales of magnitude of discriminant power.

[229] Figure 34 shows the most discriminant predictors from a selection of features (the frequency analysis components in this example) performed with sparse representation classification (SRC) at a wavelet coefficient threshold of p≤0.05 in patients with bipolar disorder compared to patients at risk of psychosis, using a discrete approximation of a mother wavelet of a particular design with retinal signal data sets from control subjects. Shading indicates the scale of the SRC magnitude at a particular position. Darker dots indicate SRC positions where the SRC exceeds a given threshold (in the illustrated example, the threshold is between 70% and 100%).

[230] Figure 35 shows the most discriminant predictors from a selection of features (frequency analysis components in this example) performed with classification by principal component analysis based on statistical significance with a statistical significance threshold of p≤0.05 in patients with bipolar disorder compared to patients at risk of psychosis, using a mother wavelet of a specific design with retinal signal datasets from control subjects.

[231] Figure 36 shows the displays of the most discriminant predictors and their significance levels (statistical p-values) from a selection of features (frequency analysis components in this example) conducted with classification by principal component analysis ranked by statistical significance in patients with bipolar disorder versus patients at risk of psychosis, using a mother wavelet of a particular design with retinal signal data sets from control subjects. In the illustrated example, the two most discriminant components were considered with a threshold of p≤0.05. Shading indicates scales of magnitude of discriminant power.

[232] Figure 37 shows the most discriminant predictors from a selection of features (the frequency analysis components in this example) performed with sparse representation classification (SRC) at a wavelet coefficient threshold of p≤0.05 in patients with schizophrenia compared to patients at risk of psychosis, using a discrete approximation of a mother wavelet of a particular design with retinal signal data sets from control subjects. Shading indicates the scale of the SRC magnitude at a particular position. Darker dots indicate SRC positions where the SRC exceeds a given threshold (in the illustrated example, the threshold is between 70% and 100%).

[233] Figure 38 shows the most discriminant predictors from a selection of features (frequency analysis components in this example) performed with classification by principal component analysis based on statistical significance with a statistical significance threshold of p≤0.05 in patients with schizophrenia compared to patients at risk of psychosis, using a mother wavelet of a specific design with retinal signal datasets from control subjects.

[234] Figure 39 shows the plots of the most discriminant predictors and their significance levels (statistical p-values) from a selection of features (frequency analysis components in this example) performed with classification by principal component analysis ranked by statistical significance in patients with schizophrenia versus patients at risk of psychosis, using a mother wavelet of a particular design with retinal signal data sets from control subjects. In this example, the two most discriminant components were considered with a threshold of p≤0.05. Shading indicates scales of magnitude of discriminant power.

[235] Figure 40 shows the most discriminant predictors from a selection of features (the frequency analysis components in this example) performed with sparse representation classification (SRC) at a wavelet coefficient threshold of p≤0.05 in patients with major depressive disorder compared to patients at risk of psychosis, using a discrete approximation of a mother wavelet of a particular design with retinal signal data sets from control subjects. Shading indicates the scale of the SRC magnitude at a particular position. Darker dots indicate SRC positions where the SRC exceeds a given threshold (in the illustrated example, the threshold is between 70% and 100%).

[236] Figure 41 shows the most discriminant predictors from a selection of features (frequency analysis components in the illustrated example) performed with classification by principal component analysis based on statistical significance with a statistical significance threshold of p≤0.05 in patients with major depressive disorder compared to patients at risk of psychosis, using a mother wavelet of a specific design with retinal signal data sets from control subjects.

[237] Figure 42 shows the displays of the most discriminant predictors and their significance levels (statistical p-values) from a selection of features (frequency analysis components in this example) conducted with classification by principal component analysis ranked by statistical significance in patients with major depressive disorder versus patients at risk of psychosis, using a mother wavelet of a particular design with retinal signal data sets from control subjects. In this example, the two most discriminant components were considered with a threshold of p≤0.05. Shading indicates scales of magnitude of discriminant power.

[238] Figure 43 shows the display of the most discriminant selected descriptors (spectral entropy in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to individuals not diagnosed with mental health disorders (control subjects). Darker dots indicate the position of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[239] Figure 44 shows the display of the most discriminant selected descriptors (kurtosis in the illustrated example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to individuals not diagnosed with mental health disorders (control subjects). Darker dots are positions of higher discrimination, where SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[240] Figure 45 shows the display of the most discriminative selected descriptors (the Hu's moment in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to individuals not diagnosed with mental health disorders (control subjects). Darker dots indicate the position of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[241] Figure 46 shows the display of the most discriminant selected descriptors (three descriptors in this example: spectral entropy, kurtosis, and Hu's moment) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to individuals not diagnosed with mental health disorders (control subjects). Darker dots indicate the position of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%). The advantage of additional extracted specific descriptors of the retinal signal (both in position and statistical significance of the information) can be understood by comparing the descriptors with those presented in Figures 43, 44, and 45.

[242] Figure 47 shows the display of the most discriminant selected descriptors (spectral entropy in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with bipolar disorder. Darker dots are positions of higher discrimination, where SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[243] Figure 48 shows the display of the most discriminant selected descriptors (kurtosis in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with bipolar disorder. Darker dots are positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[244] Figure 49 shows the display of the most discriminant selected descriptors (Hu's moment in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with bipolar disorder. Darker dots are positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[245] Figure 50 shows the display of the most discriminant selected descriptors (three descriptors in this example: spectral entropy, kurtosis, and Hu's moment) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with bipolar disorder. The darker dots are the positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%). The advantage of additional extracted specific descriptors of the retinal signal (both in terms of position and statistical significance of the information) can be understood by comparing the descriptors with those presented in Figures 47, 48, and 49.

[246] Figure 51 shows the display of the most discriminant selected descriptors (spectral entropy in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with schizophrenia. Darker dots indicate the position of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[247] Figure 52 shows the display of the most discriminant selected descriptors (kurtosis in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with schizophrenia. Darker dots are positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[248] Figure 53 shows the mapping of the most discriminant selected descriptors (Hu's moment in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with schizophrenia. Darker dots are positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[249] Figure 54 shows the display of the most discriminant selected descriptors (three descriptors in this example: spectral entropy, kurtosis, and Hu's moment) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with schizophrenia. Darker dots are the positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%). The advantage of additional extracted specific descriptors of the retinal signal (both in terms of position and statistical significance of the information) can be understood by comparing the descriptors with those presented in Figures 51, 52, and 53.

[250] Figure 55 shows the display of the most discriminant selected descriptors (spectral entropy in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with major depressive disorder. Darker dots are positions of higher discrimination, where SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[251] Figure 56 shows the display of the most discriminant selected descriptors (kurtosis in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with major depressive disorder. Darker dots are positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[252] Figure 57 shows the display of the most discriminant selected descriptors (Hu's moment in this example) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with major depressive disorder. Darker dots are positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%).

[253] Figure 58 shows the display of the most discriminant selected descriptors (three descriptors in this example: spectral entropy, kurtosis, and Hu's moment) performed with sparse representation classification (SRC) with a threshold of p≤0.05 in patients at risk of psychosis compared to patients with major depressive disorder. Darker dots are the positions of higher discrimination, where the SRC exceeds the given threshold (in the illustrated example, the threshold is 80%). The advantage of additional extracted specific descriptors of the retinal signal (both in terms of position and statistical significance of the information) can be understood by comparing the descriptors with those presented in Figures 55, 56, and 57.

[254] It is readily understood that not all technical results mentioned herein are guaranteed in every embodiment of the present invention.

[255] Those skilled in the art may suggest modifications and improvements to the above-described embodiments of the present invention. The above description serves for illustration and not limitation. Accordingly, the scope of the present invention is limited only by the scope of the following claims.

Claims (42)

1. A method for determining medical conditions based on processing retinal signal data, wherein the method is executed by at least one processor of a computer system, wherein the method comprises the steps of: receive retinal signal data corresponding to the patient, while the retinal signal data is stimulated by applying a light stimulation signal to the retina of the patient via a light stimulator that emits light, and is collected by at least one electrode for detecting electrical signals, a photosensor for measuring the intensity of the emitted light from the light stimulator, and a spectrometer for measuring the spectrum of the emitted light from the light stimulator, wherein the retinal signal data is related to light stimulation of the patient's retina and includes at least one parameter that is voltage independent and/or time independent, wherein the at least one parameter comprises one of the intensity of the emitted light, the spectrum of the emitted light, or the impedance component of the receiving circuit; determining, based on the retinal signal data, one or more features of the retinal signal; applying one or more features of the retinal signal to a mathematical model, wherein the mathematical model corresponds to a medical condition, wherein the mathematical model comprises a machine learning algorithm trained using a data set of retinal signal data of patients with the medical condition, and wherein the data set comprises at least one parameter that is voltage independent and/or time independent; derive, by means of a mathematical model, a predicted probability of the patient's medical condition; and displaying an interface containing a predicted probability of a patient's medical condition, wherein the retinal signal data has a sampling frequency of 4 to 24 kHz, and the retinal signal data is collected for a signal collection time of no more than 500 milliseconds. 2. The method according to claim 1, wherein the impedance component is recorded continuously during the acquisition of retinal signal data. 3. The method according to claim 1, further comprising the steps of: obtain clinical information cofactors extracted from the clinical information relevant to the patient; and apply cofactors of clinical information to the mathematical model. 4. The method of claim 1, wherein the medical condition is one of: post-traumatic stress disorder, seizure, substance abuse disorder, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, schizophrenia, bipolar disorder, major depressive disorder, psychosis, or attention deficit disorder. 5. The method of claim 1, further comprising the steps of selecting a drug based on the predicted probability and administering the drug to the patient. 6. A method for determining the medical condition of a patient based on a biosignature of the medical condition of the patient, wherein the method is executed by at least one processor of a computer system, wherein the method comprises the steps of: receiving retinal signal data corresponding to a patient, wherein the retinal signal data is stimulated by applying a light stimulation signal to the retina of the patient via a light stimulator that emits light, and is collected by at least one electrode for detecting electrical signals, a photosensor for measuring the intensity of emitted light from the light stimulator, and a spectrometer for measuring the spectrum of emitted light from the light stimulator, wherein the retinal signal data is related to light stimulation of the patient's retina and includes at least one parameter that is voltage independent and/or time independent, wherein the at least one parameter comprises one of the intensity of the emitted light, the spectrum of the emitted light, or the impedance component of the receiving circuit; determining, based on the retinal signal data, one or more features of the retinal signal; determine the distance between one or more features of the retinal signal and a biosignature of a medical condition, wherein the biosignature is determined using retinal signal data from patients with a medical condition, wherein the retinal signal data comprises at least one parameter that is voltage independent and/or time independent; and display an interface indicating the distance between one or more features of the retinal signal and a biosignature of a patient's medical condition, wherein the retinal signal data has a sampling frequency of 4 to 24 kHz, and the retinal signal data is collected for a signal collection time of no more than 500 milliseconds. 7. The method of claim 6, wherein the impedance component is recorded continuously during the acquisition of retinal signal data. 8. The method of claim 6, wherein the medical condition is one of: post-traumatic stress disorder, seizure, substance abuse disorder, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, schizophrenia, bipolar disorder, major depressive disorder, psychosis, or attention deficit disorder. 9. The method of claim 6, further comprising the step of determining, based on the distance between one or more retinal signals and the biosignature of the medical condition, a treatment plan for the patient. 10. A system for determining medical conditions based on processing retinal signal data, the system comprising: a light stimulator that emits light; a photo sensor for measuring the intensity of light emitted from a light stimulator; a spectrometer for measuring the spectrum of light emitted from the light stimulator; and one or more electrodes for collecting retinal signal data corresponding to the patient, and a computer system comprising at least one processor and a memory storing executable instructions that, when executed by at least one processor, cause the computer system to: receive retinal signal data, wherein the retinal signal data is related to light stimulation of the retina of the patient and includes at least one parameter that is voltage independent and/or time independent, wherein the at least one parameter comprises one of: an intensity of emitted light, a spectrum of emitted light, or an impedance component of the receiving circuit; determine, based on the retinal signal data, one or more features of the retinal signal; apply one or more features of the retinal signal to a mathematical model, wherein the mathematical model corresponds to a medical condition, wherein the mathematical model comprises a machine learning algorithm trained using a data set of retinal signal data of patients with the medical condition, and wherein the data set comprises at least one parameter that is voltage independent and/or time independent; derive, by means of a mathematical model, the predicted probability of the patient's medical condition; and display an interface containing a predicted probability of a medical condition of a patient, wherein the retinal signal data has a sampling frequency of from 4 to 24 kHz, and the retinal signal data is collected for a signal collection time of no more than 500 milliseconds. 11. The system of claim 10, wherein the impedance component is recorded continuously during the acquisition of retinal signal data. 12. The system according to paragraph 10, wherein the instructions additionally prescribe the computer system to: obtain clinical information cofactors extracted from the clinical information relevant to the patient; and apply cofactors of clinical information to a mathematical model. 13. The system of claim 10, wherein the medical condition is one of: post-traumatic stress disorder, seizure, substance abuse disorder, obsessive-compulsive disorder, Alzheimer's disease, Parkinson's disease, multiple sclerosis, autism, schizophrenia, bipolar disorder, major depressive disorder, psychosis, or attention deficit disorder. 14. The system according to paragraph 10, wherein the instructions additionally prescribe the computer system to: select a drug based on the predicted probability and administer the drug to the patient.

Images