TWI893485B - Prognosis prediction system and its operation method - Google Patents

Abstract

The present disclosure provides an operating method of a prognosis prediction system, which includes steps as follows. A pre-processed resting-state fMRI image and a pre-processed T1-weighted image are used to build a dynamic causal model to calculate multiple effective connections between motor brain regions; a multivariate regression model is constructed based on the relationship between the multiple effective connections and a rehabilitation scale.

Description

Prognostic prediction system and its operation method

The present invention relates to a system and its operating method, and more particularly to a prognosis prediction system and its operating method.

Stroke is a major illness, primarily classified as ischemic, hemorrhagic, and transient ischemic. Ischemic stroke accounts for approximately 80% of stroke cases. Stroke can present with varying clinical manifestations depending on the region of the brain affected. Besides the common symptoms of hemiplegia, hemiplegia, and a crooked mouth, lack of prompt treatment can lead to serious consequences.

In addition to medical treatment in the early stages of a stroke, the sequelae of a stroke require rehabilitation therapy to recover, so prediction and judgment after a stroke are very important.

The present invention proposes a prognosis prediction system and its operation method to improve the problems of the prior art.

In one embodiment of the present invention, the prognostic prediction system includes a storage device and a processor. The storage device stores at least one instruction, and the processor is electrically connected to the storage device. The processor is configured to access and execute at least one instruction to: use resting-state functional magnetic resonance imaging (fMRI) images and T1 weighted images to establish a dynamic causal model to calculate multiple valid links between multiple motor brain regions; and establish a multivariate regression model based on the relationship between the multiple valid links and rehabilitation scales.

In one embodiment of the present invention, the processor is configured to access and execute at least one instruction to pre-process resting functional magnetic resonance imaging images before establishing a dynamic causal model.

In one embodiment of the present invention, a processor is used to pre-process resting-state functional MRI images to: perform slice timing correction on the resting-state functional MRI images to obtain a set of slice images, and realign the set of slice images; perform spatial standardization on the set of slice images to obtain a normalized image; perform spatial smoothing on the normalized image to obtain a smoothed image; and perform bandpass filtering on the smoothed image to obtain a filtered image as a pre-processed resting-state functional MRI image.

In one embodiment of the present invention, the processor is used to access and execute at least one instruction to: generate a generalized linear model by mapping a pre-processed resting-state functional magnetic resonance imaging image through statistical parameter mapping; pre-process the T1 weighted image to spatially normalize the T1 weighted image to obtain a pre-processed T1 weighted image; apply a cerebrospinal fluid mask and a white matter spatial mask segmented from the pre-processed T1 weighted image to remove the cerebrospinal fluid from the generalized linear model The system removes the influence of fluid and white matter, leaving only signals from the brain's gray matter. Spatial masks are applied to multiple motor brain regions to capture multiple gray matter signals from each of these regions, including the primary motor cortex, premotor area, motor supplementary area, and premotor supplementary area of the left brain, as well as the primary motor cortex, premotor area, motor supplementary area, and premotor supplementary area of the right brain. A dynamic causal model is established based on these multiple gray matter signals from multiple motor brain regions.

In one embodiment of the present invention, a processor is configured to access and execute at least one instruction to: aggregate multiple validity link data for multiple cases, each of the multiple validity link data comprising multiple validity links, and the multiple cases each having multiple rehabilitation scales; calculate a correlation parameter between the value of each validity link in the multiple validity links for each of the multiple cases and the score of each corresponding item in the multiple rehabilitation scales; and extract at least one validity link from the multiple validity link data that is significantly correlated with the multiple rehabilitation scales as input to a multivariate regression model for training, wherein the correlation parameter corresponding to the at least one validity link that is significantly correlated with the multiple rehabilitation scales meets a preset correlation standard.

In one embodiment of the present invention, each of the multiple rehabilitation scales has the age, gender, and infarct volume of each case, and the processor is used to access and execute at least one instruction to add the age, gender, and infarct volume of each case to a multivariate regression model for training.

In one embodiment of the present invention, the operating method of the prognostic prediction system proposed in the present invention includes the following steps: (A) using pre-processed resting-state functional magnetic resonance imaging images and pre-processed T1 weighted images to establish a dynamic causal model to calculate multiple validity links between multiple motor brain regions; (B) establishing a multivariate regression model based on the relationship between the multiple validity links and the rehabilitation scale.

In one embodiment of the present invention, step (A) comprises: mapping the pre-processed resting-state functional magnetic resonance imaging image through statistical parameters to generate a generalized linear model; applying the cerebrospinal fluid mask and white matter space mask segmented from the pre-processed T1 weighted image to remove the influence of cerebrospinal fluid and white matter from the generalized linear model, leaving the signal of the gray matter of the brain; applying the multiple motor brain regions Spatial masking captures multiple gray matter signals from multiple motor brain regions, including the primary motor cortex, premotor area, motor supplementary area, and premotor supplementary area of the left brain, as well as the primary motor cortex, premotor area, motor supplementary area, and premotor supplementary area of the right brain. A dynamic causal model is established based on the multiple gray matter signals from multiple motor brain regions.

In one embodiment of the present invention, step (B) includes: aggregating multiple validity link data for multiple cases, each of the multiple validity link data includes multiple validity links, and the multiple cases respectively have multiple rehabilitation scales; calculating a correlation parameter between the value of each validity link in the multiple validity links for each of the multiple cases and the score of each corresponding item in the multiple rehabilitation scales; extracting at least one validity link from the multiple validity link data that is significantly correlated with the multiple rehabilitation scales as an input to a multivariate regression model for training, wherein the correlation parameter corresponding to the at least one validity link that is significantly correlated with the multiple rehabilitation scales meets a preset correlation standard.

In one embodiment of the present invention, step (B) further comprises: adding the age, gender, and infarct volume of each case to a multivariate regression model for training.

In summary, the technical solutions of the present invention offer significant advantages and beneficial effects compared to existing technologies. The prognosis prediction system and its operating method of the present invention measure signals from specific neural systems (e.g., motor brain regions) to calculate their correlation with specific functions. Based on this, a multivariate regression model is established that can predict the rehabilitation prognosis of patients (e.g., stroke patients) using neural signals obtained through neuroimaging.

The following will describe the above description in detail with an implementation method and provide a further explanation of the technical solution of the present invention.

To make the description of the present invention more detailed and complete, reference is made to the accompanying drawings and various embodiments described below, in which the same numbers represent the same or similar elements. On the other hand, well-known elements and steps are not described in the embodiments to avoid unnecessary limitations on the present invention.

Referring to Figure 1 , the technical aspect of the present invention is a prognostic prediction system 100 , which can be applied to predict the rehabilitation prognosis of stroke patients or widely applied in related technical fields. This technical aspect of the prognostic prediction system 100 can achieve significant technical advancements and has broad industrial application value. The following will illustrate the specific implementation of the prognostic prediction system 100 with reference to Figure 1 .

It should be understood that various embodiments of the prognostic prediction system 100 are described in conjunction with Figure 1. In the following description, for ease of explanation, many specific details are further set to provide a comprehensive description of one or more embodiments. However, the present technology can be implemented without these specific details. In other examples, in order to effectively describe these embodiments, known structures and devices are shown in block diagram form. The term "for example" used here means "as an example, instance or illustration." Any embodiment described herein as "for example" is not necessarily interpreted as better or superior to other embodiments.

In practice, in one embodiment of the present invention, the prognostic prediction system 100 can be a server, a computer host, or other computer equipment. In terms of servers, many technologies that have been developed or are under development can manage the operation of computer servers, generally providing accessibility, consistency, and efficiency. Remote management allows the removal of input and output interfaces for servers (e.g., display screen, mouse, keyboard, etc.) and the need for network administrators to physically access each server. For example, large data centers containing many computer servers are generally managed using a variety of remote management tools to configure, monitor, and debug server hardware and software.

It should be understood that the terms "about," "approximately," or "substantially" used herein are used to modify any quantity that may vary slightly, but such slight variations do not alter its essence. Unless otherwise specified in the embodiments, the error range of the value modified by "about," "approximately," or "substantially" is generally allowed to be within 20%, preferably within 10%, and more preferably within 5%.

In practice, in one embodiment of the present invention, the prognostic prediction system 100 can selectively establish a connection with the magnetic resonance imaging device 190. It should be understood that in the embodiments and patent scope, the description of "connection" can generally refer to one component indirectly communicating with another component through other components by wire and/or wireless communication, or one component being physically connected to another component without being connected through other components. For example, the prognostic prediction system 100 can indirectly communicate with the magnetic resonance imaging device 190 through other components by wire and/or wireless communication, or the prognostic prediction system 100 can be physically connected to the magnetic resonance imaging device 190 without being connected through other components. Those skilled in the art should flexibly choose this option based on their current needs.

FIG1 is a block diagram of a prognostic prediction system 100 according to an embodiment of the present invention. As shown in FIG1 , prognostic prediction system 100 includes a storage device 110, a processor 120, and a display 130. For example, storage device 110 may be a hard drive, a flash memory device, or other storage medium; processor 120 may be a central processing unit (CPU); and display 130 may be a built-in display or an external screen.

Architecturally, the prognosis prediction system 100 is electrically connected to the MRI device 190, the storage device 110 is electrically connected to the processor 120, and the processor 120 is electrically connected to the display 130. It should be understood that in the embodiments and patent claims, the term "electrically connected" may generally refer to one component being indirectly electrically coupled to another component through other components, or one component being directly electrically connected to another component without going through other components. For example, the storage device 110 may be a built-in storage device directly electrically connected to the processor 120, or the storage device 110 may be an external storage device indirectly connected to the processor 120 via a network device.

During use, the magnetic resonance imaging (MRI) device 190 can acquire resting-state functional MRI (rs-fMRI) and T1-weighted images of the brain. In practice, for example, the MRI device 190 can acquire resting-state functional MRI and T1-weighted images of a patient seven days, one month, and three months after a stroke. Although only one MRI device 190 is depicted in Figure 1, this is not a limitation of the present invention. In practice, the MRI device 190 may refer to one or more MRI devices, and those skilled in the art should flexibly select one based on their current needs.

In some embodiments of the present invention, storage device 110 stores the aforementioned images and at least one instruction, and processor 120 is configured to access and execute the at least one instruction to: use resting-state functional MRI images and T1-weighted images to establish a dynamic causal model to calculate multiple validity links between multiple motor brain regions; and to establish a multivariate regression model based on the relationships between the multiple validity links and rehabilitation scales. Display 130 may display the prediction results of the multivariate regression model for predicting a patient's rehabilitation prognosis. Thus, the prognosis prediction system 100 measures the signals of a specific neural system (e.g., motor brain area) to calculate its correlation with a specific function, and based on this, establishes a multivariate regression model that can predict the rehabilitation prognosis of patients (e.g., stroke patients) using neural signals obtained through neuroimaging.

In a controlled experiment, a predictive model based on the brain connectome was used to calculate the correlation between functional connections between different brain regions and brain behavior. This was then used as a feature for repeated training to establish a predictive model to predict the rehabilitation of stroke patients. However, the controlled experiment explored functional connections without directionality. Functional connections describe the statistical dependencies between brain regions and do not consider causal relationships. The validity connection of the present invention describes the causal relationship between brain regions. By observing neural activity at different times and spaces, the causal interaction between brain regions is inferred. Therefore, the validity connection has directionality and facilitatory and inhibitory effects, which can more clearly describe the nature of the relationship between brain regions.

To quantify a patient's post-stroke recovery, the prognosis prediction system 100 utilizes motor function-related rehabilitation scales to assess the patient's various motor indicators. In some embodiments of the present invention, the rehabilitation scale may be a stroke rehabilitation scale, such as the Modified Rankin Scale (MRS), Barthel Index (BI), Berg Balance Scale (BBS), and Fugl-Meyer Assessment (FMA), which respectively assess a patient's disability level, ability to care for themselves, balance, and functional recovery.

The prognostic prediction system 100 uses resting-state functional magnetic resonance imaging (fMRI) images to establish a dynamic causal model to calculate the effectiveness of linkages between motor brain regions. Based on the relationship between the effectiveness of linkages and the stroke rehabilitation scale at seven days, one month, and three months after the stroke, a multivariate regression model is established to predict the motor function rehabilitation status of ischemic stroke patients at different stages after the stroke.

Dynamic causal modeling (DCM) is a method used to infer neural processes based on temporal measurements, such as functional magnetic resonance imaging (fMRI) data. The core concept is to transform the modeled neurodynamics into hemodynamic responses that approximate real-world conditions and to calculate the parameters of a realistic neural model to predict the blood-oxygen-level-dependent (BOLD) signal.

DCM models the temporal changes of the neural state vector and describes the activity state of the brain region using a state equation that includes the current state z, external input u, and other neural parameters θ, as shown in the following equation (1):

In equation (1), the state z and input u are time-dependent, while the parameter θ is time-independent. Therefore, F can be expressed as a bilinear equation as shown in equation (2):

Among them, A, Bj, and C can be expressed as partial differentials of F:

(4)

(5)

These three parameters describe the potential influences on a neuron's state under natural conditions. A relates solely to the neuron's state and represents the degree of connectivity between brain regions when there is no external input (u = 0), i.e., the effective connectivity in the resting state. B represents the effect of j stimuli on the connectivity between brain regions, or the modulatory effect of experimental manipulation of external input variables on the effective connectivity. C represents the direct effect of external input signals on the neuron's state.

To further explain the operating method of the aforementioned prognostic prediction system 100, please refer to Figures 1 and 2. Figure 2 is a flow chart of an operating method 200 of the prognostic prediction system 100 according to an embodiment of the present invention. As shown in Figure 2, operating method 200 includes steps S201-S211 (it should be understood that, except for those steps specifically described in the order, the steps mentioned in this embodiment can be adjusted in order according to actual needs and can even be executed simultaneously or partially simultaneously).

The operating method 200 may take the form of a computer program product on a non-transitory computer-readable recording medium having a plurality of computer-readable instructions embodied in the medium. Suitable recording media may include any of the following: non-volatile memory, such as read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), and electrically erasable programmable read-only memory (EEPROM); volatile memory, such as static access memory (SRAM), dynamic access memory (DRAM), and double data rate random access memory (DDR-RAM); optical storage devices, such as compact disc read-only ROM (CD-ROM) and digital versatile disc read-only ROM (DVD-ROM); and magnetic storage devices, such as hard drives and floppy drives.

In step S201, the MRI device 190 can obtain resting-state functional MRI images of the brain, for example, resting-state functional MRI images taken seven days, one month, and three months after a stroke. Before establishing the dynamic causal model, in step S202, the processor 120 accesses and executes at least one instruction to pre-process the resting-state functional MRI images.

Regarding the image pre-processing of step S202, in some embodiments of the present invention, the processor 120 is used to pre-process the resting-state functional MRI image to: perform slice timing correction on the resting-state functional MRI image to obtain a set of slice images, and realign the set of slice images; perform spatial standardization on the set of slice images to obtain a normalized image; perform spatial smoothing on the normalized image to obtain a smoothed image; and perform bandpass filtering on the smoothed image to obtain a filtered image as a pre-processed resting-state functional MRI image.

Specifically, the brain is divided into gray matter and white matter. Gray matter is home to neuronal nuclei, while white matter contains neuronal synapses and axons. To better calculate the connectivity between brain regions, pre-processing is required to correct for physiological noise in resting-state functional MRI images. In practice, for example, resting-state functional MRI images have multiple sections, which can be acquired in descending order from top to bottom, ascending order from bottom to top, or alternating ascending and descending orders. Slice timing correction is used to reconstruct three-dimensional (3D) resting-state functional MRI images according to the spatial order of the sections. Realignment is used to correct for head movement during subsequent MRI scans. This correction is accomplished by translating and rotating the original image signals along the x, y, and z axes to align the image with the midpoint of the slice sequence. Because the shape of each individual's brain image varies, all images are typically mapped to a reference space of the same size, a process known as spatial normalization. Spatial smoothing can reduce noise in the image by using a Gaussian kernel to perform a weighted average of the intensities of neighboring voxels. However, excessive smoothing can result in loss of image detail, so the degree of smoothing needs to be determined based on the voxel size. The degree of smoothing is determined by the kernel's full width at half maximum (FWHM), which is typically approximately 1.5 times the voxel size of the original image. Resting-state fMRI images often contain noise other than neural signals, such as heartbeat (0.15 Hz) and respiration (0.3 Hz). Temporal filtering aims to remove this noise to improve the signal-to-noise ratio (SNR). A 0.01-0.08 Hz bandpass filter is typically used to remove noise from resting-state fMRI images.

In step S203, the MRI device 190 acquires a T1-weighted image of the brain. In step S204, the processor 120 accesses and executes at least one instruction to pre-process the T1-weighted image, thereby spatially normalizing the T1-weighted image to obtain a pre-processed T1-weighted image. This allows the pre-processed T1-weighted image and the pre-processed resting-state functional MRI image to be mapped to the same reference space.

In step S205, the processor 120 accesses and executes at least one instruction to: generate a generalized linear model (GLM) by using statistical parametric mapping (SPM) on the pre-processed resting-state functional magnetic resonance imaging image; apply the cerebrospinal fluid mask and white matter spatial mask segmented from the pre-processed T1 weighted image to remove the influence of cerebrospinal fluid and white matter from the GLM, leaving only the signal of the cerebral gray matter; apply the spatial mask of multiple motor brain regions to respectively capture multiple gray matter signals of the multiple motor brain regions, including the primary motor cortex (M1) of the left cerebrum, the premotor cortex (PM) of the left cerebrum, and the supplementary motor area (SMA) of the left cerebrum. SMA), the pre-supplementary motor area (preSMA) of the left brain, the primary motor cortex of the right brain, the premotor area of the right brain, the motor supplementary area of the right brain, and the pre-supplementary motor area of the right brain.

In step S206, the processor 120 is used to access and execute at least one instruction to establish a dynamic causal model based on the plurality of gray matter signals of the plurality of motor brain regions.

Specifically, step S205 analyzes the signals of the motor brain areas of interest. The motor brain areas selected include the primary motor cortex, premotor area, motor supplementary area, and premotor supplementary area, with a total of eight brain areas symmetrically distributed between the left and right hemispheres as the areas of interest.

In order to calculate the effective links between brain regions, a dynamic causal model needs to be specified. In practice, for example, the pre-processed resting-state functional magnetic resonance imaging image of a single case is first input into a statistical parameter mapping application (such as SPM12) to generate a generalized linear model. Gray matter is the location of neuronal nuclei. In order to more accurately capture the signal of neuronal activity, step S205 needs to filter out the signal outside the gray matter. The cerebrospinal fluid and white matter spatial masks segmented from the pre-processed T1 weighted image are applied to capture the blood (BOLD) signals of the cerebrospinal fluid and white matter, respectively. Then, the generalized linear model obtained in step 1 is used to remove the influence of cerebrospinal fluid and white matter, leaving the signal of the gray matter of the brain. In step S205, the spatial mask of the motor brain region of interest is applied to respectively capture the gray matter signals of the eight motor brain regions of interest. The spatial mask of the motor brain region can be segmented from the pre-processed T1 weighted image, for example, but the present invention is not limited thereto.

Next, in step S206, the motor brain area signals are input into SPM12's Dynamic Causal Modeling (DCM) analysis to establish a dynamic causal model. The model's parameter settings include: Modulatory effect: bilinear, State per region: one, Stochastic effects: no, Center input: no, Fit timeseries or CSD: CSD, and Connection: fully connected model.

Based on the above settings, in step S207, the calculated dynamic causal model is retrieved using the review function in SPM12. Coupling(A) represents the effective connectivity between brain regions (i.e., matrix A). In the topological model of Coupling(A), nodes represent the motor brain regions of interest, and the values between nodes represent the effective connectivity between these regions. The effective connectivity matrix is organized according to the direction of the arrows and the magnitude of the values.

Repeating steps S201-S207, processor 120 processes each patient's resting-state functional MRI images. In step S208, storage device 110 stores each patient's rehabilitation score. In step S209, each patient's effectiveness score is linked to its corresponding rehabilitation scale as input and output of the prediction model to establish a multivariate regression model.

In some embodiments of the present invention, in steps S207-S209, the processor is configured to access and execute at least one instruction to: aggregate multiple validity link data (e.g., validity link matrices) for multiple cases, each of the multiple validity link data comprising multiple validity links, and the multiple cases each having multiple rehabilitation scales; calculate correlation parameters between the values of each validity link in the multiple validity links for each of the multiple cases and the scores of each corresponding score in the multiple rehabilitation scales; and extract at least one validity link from the multiple validity link data that is significantly correlated with the multiple rehabilitation scales as input to a multivariate regression model for training, wherein the correlation parameter corresponding to the at least one validity link that is significantly correlated with the multiple rehabilitation scales meets a preset correlation standard.

Regarding the default correlation criterion, in some embodiments of the present invention, the correlation parameter may be a p-value. If the p-value is less than a threshold (e.g., 0.1 or 0.05), the correlation parameter meets the default correlation criterion. Alternatively, the correlation parameter may be a correlation coefficient (e.g., r-value). If the absolute value of the correlation coefficient is greater than a threshold (e.g., 0.4), the correlation parameter meets the default correlation criterion.

In some embodiments of the present invention, each of the multiple rehabilitation scales provided in step S208 includes the age, gender, and infarct volume of each case. In step S209, the processor 120 accesses and executes at least one instruction to add the age, gender, and infarct volume of each case to a multivariate regression model for training.

The multivariate regression in step S209 is an enhanced version of simple regression, which is used to explore the relationship between multiple variables (such as one or more significantly correlated validity links, age, gender, and infarct volume) and a response variable (such as the score on the rehabilitation scale) and to establish a multivariate regression model to predict the response variable of interest.

In practice, for example, the present invention uses the Regression Learner App in MATLAB software to conduct multivariate regression analysis, applying linear and nonlinear regression models to multiple traits and rehabilitation scales. The trained models are then validated using a leave-one-out cross-validation method, comparing the performance of different models and selecting the model with the highest coefficient of determinant as the optimal model.

During machine learning, the features used in training may include irrelevant and redundant features. Removing these redundant features does not result in loss of information necessary for training, but irrelevant features can negatively impact the model and affect its performance. Therefore, the present invention minimizes irrelevant features in the training features to improve model performance. Furthermore, fewer features can save computational time and money.

The predictors used in the present invention are 64 elements of the validity link matrix (i.e., 8×8=64 validity links), which include characteristics unrelated to the rehabilitation scale. Including all elements in the training would be time-consuming and the model's regression coefficient would generally be low. To effectively improve the model's effectiveness, in step S209, the correlation coefficient between the motor brain region validity links and the rehabilitation scale is calculated. Validity links that are significantly correlated with the rehabilitation scale (e.g., p<0.05 and |r|>0.4) are selected as inputs to the regression model. Age, gender, and infarct volume are also added as clinical characteristics to the training.

In steps S210-S211, to investigate the impact of different infarct regions on prediction results, processor 120 grouped the cases according to stroke location, primarily into two groups: those with infarcts in the cortex, corona radiata, and basal ganglia, and those in the midbrain, pons, and medulla oblongata. Furthermore, processor 120 also analyzed the groups based on whether the infarct region was located in the left or right hemisphere of the brain.

For example, each case group was analyzed based on information from different time periods: onset effectiveness linking and clinical characteristics predicting rehabilitation scale one month after stroke; onset effectiveness linking and clinical characteristics predicting rehabilitation scale three months after stroke; and one month after stroke effectiveness linking and clinical characteristics predicting rehabilitation scale three months after stroke. In practice, the trained multivariate regression model was able to predict rehabilitation outcomes well.

In summary, the technical solutions of the present invention offer significant advantages and beneficial effects compared to existing technologies. The prognosis prediction system 100 and its operating method 200 of the present invention measure signals from specific neural systems (e.g., motor brain regions) to calculate their correlation with specific functions. Based on this, a multivariate regression model is established that can predict the rehabilitation prognosis of patients (e.g., stroke patients) using neural signals obtained through neuroimaging.

Although the present invention has been disclosed in the form of embodiments as described above, it is not intended to limit the present invention. Anyone skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the scope of the attached patent application.

To facilitate understanding of the above and other objects, features, advantages, and embodiments of the present invention, the accompanying symbols are described as follows: 100: Prognosis prediction system 110: Storage device 120: Processor 130: Display 190: Magnetic resonance imaging device 200: Operation method S201-S211: Steps

To facilitate a clearer understanding of the above and other objects, features, advantages, and embodiments of the present invention, the accompanying drawings are described as follows: Figure 1 is a block diagram of a prognostic prediction system according to an embodiment of the present invention; and Figure 2 is a flow chart of an operating method of a prognostic prediction system according to an embodiment of the present invention.

200: How it works

S201~S211: Steps

Claims (10)

A prognosis prediction system comprises: a storage device storing at least one instruction; and a processor electrically connected to the storage device, wherein the processor is configured to access and execute the at least one instruction to: use a resting-state functional magnetic resonance imaging image and a T1 weighted image to establish a dynamic causal model to calculate multiple validity links between multiple motor brain regions, wherein parameter settings of the dynamic causal model include moderation, per-region state, random effect, central input setting, simulated time series, and connection setting; and establish a multivariate regression model based on the relationship between the multiple validity links and a rehabilitation scale; The dynamic causal model is modeled based on the temporal changes of the neural state vector to describe the activity states of the multiple motor brain regions, including the effects of resting states, the effects of multiple stimuli on the multiple motor brain regions, and the effects of external input signals. The prognostic prediction system of claim 1, wherein the processor is configured to access and execute the at least one instruction to: Pre-process the resting-state functional magnetic resonance imaging image before establishing the dynamic causal model. The prognostic prediction system of claim 2, wherein the processor is configured to pre-process the resting-state functional MRI image to: perform slice timing correction on the resting-state functional MRI image to obtain a set of slice images, and realign the set of slice images; perform spatial normalization on the set of slice images to obtain a normalized image; perform spatial smoothing on the normalized image to obtain a smoothed image; and perform bandpass filtering on the smoothed image to obtain a filtered image to serve as the resting-state functional MRI image after the pre-processing. The prognostic prediction system of claim 3, wherein the processor is configured to access and execute the at least one instruction to: Process the pre-processed resting-state functional MRI image through statistical parameter mapping to generate a generalized linear model; Pre-process the T1 weighted image to spatially normalize the T1 weighted image to obtain a pre-processed T1 weighted image; Apply a cerebrospinal fluid mask and a white matter spatial mask segmented from the pre-processed T1 weighted image to remove the effects of cerebrospinal fluid and white matter from the generalized linear model, leaving only the signal of the cerebral gray matter; Applying spatial masks to the multiple motor brain regions, a plurality of gray matter signals are captured from the multiple motor brain regions, including the primary motor cortex of the left brain, the premotor area of the left brain, the motor supplementary area of the left brain, the premotor supplementary area of the left brain, the primary motor cortex of the right brain, the premotor area of the right brain, the motor supplementary area of the right brain, and the premotor supplementary area of the right brain; and establishing the dynamic causal model based on the plurality of gray matter signals from the multiple motor brain regions. The prognostic prediction system of claim 1, wherein the processor is configured to access and execute the at least one instruction to: aggregate a plurality of validity link data for a plurality of cases, each of the plurality of validity link data comprising the plurality of validity links, the plurality of cases respectively having a plurality of rehabilitation scales; calculate a correlation parameter between the value of each validity link in the plurality of validity links for each of the plurality of cases and the score of each corresponding score in the plurality of rehabilitation scales; and remove at least one validity link from the plurality of validity link data that is significantly correlated with the plurality of rehabilitation scales as an input to the multivariate regression model for training, wherein the correlation parameter corresponding to the at least one validity link that is significantly correlated with the plurality of rehabilitation scales meets a preset correlation criterion. The prognostic prediction system of claim 5, wherein each of the plurality of rehabilitation scales includes age, gender, and infarct volume for each case, and the processor is configured to access and execute the at least one instruction to: Add the age, gender, and infarct volume of each case to the multivariate regression model for training. A method for operating a prognostic prediction system, wherein the prognostic prediction system includes a processor, the method comprising the following steps: (A) using a pre-processed resting-state functional magnetic resonance imaging image and a pre-processed T1 weighted image to establish a dynamic causal model by the processor to calculate multiple validity links between multiple motor brain regions, wherein the parameter settings of the dynamic causal model include moderation, each region state, random effect, central input setting, fitting time series, and connection setting; and (B) establishing a multivariate regression model by the processor based on the relationship between the multiple validity links and a rehabilitation scale; The dynamic causal model is modeled based on the temporal changes of the neural state vector to describe the activity states of the multiple motor brain regions, including the effects of resting states, the effects of multiple stimuli on the multiple motor brain regions, and the effects of external input signals. The method of claim 7, wherein step (A) comprises: Processing the pre-processed resting-state functional magnetic resonance imaging image through statistical parameter mapping to generate a generalized linear model; Applying the cerebrospinal fluid mask and white matter spatial mask segmented from the pre-processed T1-weighted image to the generalized linear model by the processor to remove the effects of cerebrospinal fluid and white matter, leaving only the signal of cerebral gray matter; The processor applies spatial masks to the plurality of motor brain regions to respectively acquire a plurality of gray matter signals from the plurality of motor brain regions, the plurality of motor brain regions comprising the primary motor cortex of the left brain, the premotor area of the left brain, the motor supplementary area of the left brain, the premotor supplementary area of the left brain, the primary motor cortex of the right brain, the premotor area of the right brain, the motor supplementary area of the right brain, and the premotor supplementary area of the right brain; and the processor establishes the dynamic causal model based on the plurality of gray matter signals from the plurality of motor brain regions. The operating method of claim 7, wherein step (B) comprises: aggregating, by the processor, a plurality of validity link data for a plurality of cases, each of the plurality of validity link data comprising the plurality of validity links, the plurality of cases respectively having a plurality of rehabilitation scales; calculating, by the processor, a correlation parameter between the value of each validity link in the plurality of validity links for each of the plurality of cases and the score of each corresponding item in the plurality of rehabilitation scales; and retrieving, by the processor, at least one validity link from the plurality of validity link data that is significantly correlated with the plurality of rehabilitation scales as an input to the multivariate regression model for training, wherein the correlation parameter corresponding to the at least one validity link that is significantly correlated with the plurality of rehabilitation scales meets a predetermined correlation criterion. The method of claim 9, wherein step (B) further comprises: Including, by the processor, the age, sex, and infarct volume of each case into the multivariate regression model for training.