JP2020060514A - Medical image processing apparatus and medical image processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate image data of image quality equivalent to that in a normal state even when a detector is out of order. A medical image processing apparatus according to an embodiment includes an acquisition unit and a processing unit. The acquisition unit acquires first medical data that does not include data regarding the first PET detector among the plurality of PET detectors arranged in a ring shape. The processing unit generates second medical data including data regarding the first PET detector from the acquired first medical data according to a machine learning model. [Selection diagram]

Description

  Embodiments of the present invention relate to a medical image processing apparatus and a medical image processing method.

  The nuclear medicine diagnostic apparatus detects radiation emitted from a subject administered with a radioisotope by a plurality of detectors in a detector ring. The plurality of detectors are arranged in a ring shape in the detector ring. As a device of this type, a positron emission tomography (PET) device and a PET-CT (computer tomography) device are typical. In the nuclear medicine diagnostic device, when any one of the plurality of detectors in the detector ring fails, the number of Lines of Response (LOR) decreases as compared to the normal state, so that the system collects data. Or the reconstructed image is distorted.

U.S. Patent Application Publication No. 2018/0032641 U.S. Patent Application Publication No. 2017/0372193 U.S. Patent Application Publication No. 2006/0214110

  The problem to be solved by the invention is to generate image data having an image quality corresponding to a normal state even in a state where a detector is out of order.

  The medical image processing apparatus according to the embodiment includes an acquisition unit and a processing unit. The acquisition unit acquires first medical data that does not include data regarding the first PET detector among the plurality of PET detectors arranged in a ring shape. The processing unit generates second medical data including data on the first PET detector from the acquired first medical data according to a machine learning model.

FIG. 1 is a diagram showing an example of the configuration of a positron emission tomography (PET) apparatus according to the first embodiment. FIG. 2 is a schematic vertical sectional view of a detector ring mounted on the gantry of FIG. FIG. 3 is a flowchart showing an example of learning data generation processing executed in the PET apparatus of FIG. FIG. 4 is a diagram for explaining learning and / or reinforcement learning of a machine learning model using a learning data set generated in the learning data generation process of FIG. FIG. 5 is a flowchart showing an example of the degenerate operation process executed in the PET apparatus of FIG. FIG. 6 is a diagram for explaining image correction using a machine learning model executed in the degenerate operation process of FIG. FIG. 7 is a diagram for explaining learning and / or reinforcement learning of the machine learning model and image correction using the machine learning model according to the second embodiment. FIG. 8 is a diagram for explaining learning and / or reinforcement learning of the machine learning model and image correction using the machine learning model according to the third embodiment.

[First Embodiment]
Hereinafter, a medical image processing apparatus and a medical image processing method according to the present embodiment will be described with reference to the drawings. In the following description, constituent elements having the same or substantially the same functions as those described above with reference to the drawings described above are designated by the same reference numerals, and will be redundantly described only when necessary. Even if the same portion is shown, the dimensions and ratios may be different depending on the drawings.

  The medical image processing apparatus according to this embodiment is installed in, for example, a nuclear medicine diagnostic apparatus. A nuclear medicine diagnostic device detects radiation emitted from a subject administered a radioisotope by a plurality of detector modules in a detector ring. The plurality of detector modules are arranged in a ring shape in the detector ring. The nuclear medicine diagnostic apparatus includes, for example, a Positron Emission Tomography (PET) apparatus, a PET / Computer Tomography (CT) apparatus, a PET / Magnetic Resonance Imaging (MRI) apparatus, and a single photon emission computer tomography apparatus. An imaging (single photon emission CT: SPECT) device, a SPECT / CT device, and a SPECT / MRI device. The nuclear medicine diagnostic apparatus equipped with the medical image processing apparatus may be referred to as a medical image processing apparatus.

  Hereinafter, a medical image processing apparatus installed in a PET apparatus and a medical image processing method in the PET apparatus will be described as an example.

  FIG. 1 is a diagram showing an example of the configuration of a PET device 1 according to this embodiment. FIG. 2 is a schematic vertical cross-sectional view of the detector ring 11 mounted on the gantry 10 of FIG.

  As shown in FIG. 1, the PET device 1 includes a gantry 10, a bed 50, and a console 70. For convenience of explanation, a plurality of gantry 10 is drawn in FIG. Typically, the gantry 10 and the bed 50 are installed in a common examination room, and the console 70 is installed in a control room adjacent to the examination room.

  The gantry 10 is an imaging device for PET imaging the subject P. As shown in FIGS. 1 and 2, the gantry 10 has a plurality of detector rings 11, a plurality of signal processing circuits 13 and a coincidence counting circuit 15.

  As shown in FIG. 2, the plurality of detector rings 11 are arranged in the gantry 10 along the central axis Z of the circumference. In FIG. 2, three detector rings 11 are shown for purposes of illustration. An image field of view (FOV) is formed in the opening of the detector ring 11. The top plate 53 on which the subject P is placed is inserted into the opening of the detector ring 11 so that the imaging region of the subject P enters the FOV. The subject P is placed on the top plate 53 so that the body axis thereof coincides with the central axis Z. A drug labeled with a radioisotope is injected into the subject P for PET imaging. As shown in FIG. 2, each detector ring 11 has a plurality of detectors 110 arranged on the circumference around the central axis Z. Here, the detector 110 is an example of a PET detector. The array of the plurality of detectors 110 is not limited to the ring shape shown in FIGS. 1 and 2, and may be, for example, an elliptical shape, a triangular shape, or the like. Each of the plurality of detectors 110 detects the pair annihilation gamma rays emitted from the inside of the subject P and generates a pulsed electric signal according to the light amount of the detected pair annihilation gamma rays. Specifically, the plurality of detectors 110 have a plurality of scintillators 113 and a plurality of photomultiplier tubes 115.

  Each of the plurality of scintillators 113 receives pair annihilation gamma rays derived from the radioisotope injected into the subject P and generates scintillation light. In the detector 110, the plurality of scintillators 113 are arranged, for example, two-dimensionally. Each scintillator 113 is arranged such that the long axis direction of each scintillator 113 substantially coincides with the radial direction of the detector ring 11. The scintillator 113 according to this embodiment may be formed of any existing type of scintillator material. For example, the scintillator 113 is NaI (sodium iodide), BGO (bismuth acid germanate), LSO (lutetium silicate with a certain amount of cerium added), LaBr3: Ce, LYSO (mixed crystal of LSO and yttrium silicate). ) Or other scintillator material. A lutetium crystal is often used as the material of the scintillator 113. In addition to the materials described above, the scintillator 113 may be formed of, for example, a gallium-based crystal or a garnet-based crystal.

  Each of the plurality of photomultiplier tubes 115 is provided at one end of the scintillator 113 in the radial direction orthogonal to the central axis Z. A light guide (not shown) is typically provided between the scintillator 113 and the photomultiplier tube 115. The plurality of scintillators 113 and the plurality of photomultiplier tubes 115 included in the detector ring 11 are arranged in a concentric circle (concentric cylinder) shape. The scintillation light generated in the scintillator 113 propagates in the scintillator 113 and goes to the photomultiplier tube 115. The photomultiplier tube 115 generates a pulsed electric signal according to the amount of scintillation light. The generated electric signal is supplied to the signal processing circuit 13, as shown in FIG. Further, the photomultiplier tube 115 may be a silicon photomultiplier element (SiPM).

  The number of rows of detector rings 11, the number of detectors 110 in the detector ring 11, the number of scintillators 113 in the detector 110, and the number of photomultiplier tubes 115 in the detector 110 are as shown in FIG. It is not limited to the number shown in 2.

  The plurality of signal processing circuits 13 generate single event data based on the electric signals of the plurality of detectors 110, respectively. Specifically, the signal processing circuit 13 performs detection time measurement processing, position calculation processing, and energy calculation processing. The signal processing circuit 13 includes an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), and a composite configured so as to be capable of performing detection time measurement processing, position calculation processing, and energy calculation processing. It is realized by a programmable logic device (Complex Programmable Logic Device: CPLD), a simple programmable logic device (Simple Programmable Logic Device: SPLD), and the like.

  The signal processing circuit 13 measures the detection time of the gamma ray by the detector 110 by the detection time measurement processing. Specifically, the signal processing circuit 13 monitors the peak value of the electric signal from the detector 110, and measures the time when the peak value exceeds a preset threshold value as the detection time. In other words, the signal processing circuit 13 electrically detects annihilation gamma rays by detecting that the peak value exceeds the threshold value.

  The signal processing circuit 13 calculates the incident position of the pair annihilation gamma ray based on the electric signal from the detector 110 by the position calculation process. The incident position of the annihilation gamma ray corresponds to the position coordinate of the scintillator on which the annihilation gamma ray is incident.

  The signal processing circuit 13 calculates the energy value of the detected pair annihilation gamma ray based on the electric signal from the detector 110 by the energy calculation process. The detection time data, the position coordinate data, and the energy value data regarding a single event are associated with each other. A combination of energy value data, position coordinate data, and detection time data relating to a single event is called single event data. Single event data is generated one after another every time an annihilation gamma ray is detected. The generated single event data is supplied to the coincidence counting circuit 15.

  The coincidence counting circuit 15 performs coincidence counting processing on the single event data from the signal processing circuit 13. As a hardware resource, the coincidence counting circuit 15 is realized by an ASIC, an FPGA, a CPLD, an SPLD, or the like configured to be capable of executing the coincidence counting process. The coincidence counting circuit 15 repeatedly identifies the single event data relating to two single events within a predetermined time frame from the repeatedly supplied single event data by the coincidence counting process. It is presumed that the single event of this pair originates from pair annihilation gamma rays generated from the same pair annihilation point. The paired single events are collectively referred to as a coincidence event. The line connecting the pair of detectors 110 (more specifically, the scintillator) that has detected this pair annihilation gamma ray is called a line of response (LOR). The event data related to the pair of events forming the LOR is called coincidence counting event data. The coincidence counting event data and the single event data are transmitted to the console 70. In addition, when the coincidence counting event data and the single event data are not particularly distinguished, they are referred to as PET event data.

  In the above configuration, the signal processing circuit 13 and the coincidence counting circuit 15 are included in the gantry 10, but the present embodiment is not limited to this. For example, both the coincidence counting circuit 15 or the signal processing circuit 13 and the coincidence counting circuit 15 may be included in a device separate from the gantry 10. Further, one coincidence counting circuit 15 may be provided for the plurality of signal processing circuits 13 mounted on the gantry 10, or the plurality of signal processing circuits 13 mounted on the gantry 10 may be divided into a plurality of groups. However, one coincidence counting circuit 15 may be provided for each group.

  The bed 50 mounts the subject P to be scanned and moves the mounted subject P. The bed 50 includes a base 51, a support frame 52, a top plate 53, and a bed driving device 54. The base 51 is installed on the floor. The base 51 is a housing that supports the support frame 52 so as to be movable in the vertical direction (Y-axis direction) with respect to the floor surface. The support frame 52 is a frame provided above the base 51. The support frame 52 supports the top plate 53 slidably along the central axis Z. The top plate 53 is a flexible plate on which the subject P is placed. The bed driving device 54 is housed in the housing of the bed 50. The bed driving device 54 is a motor or an actuator that generates power for moving the support frame 52 on which the subject P is placed and the top plate 53. The bed driving device 54 operates under the control of the console 70 or the like.

  The console 70 has a computer that controls the gantry 10 and the bed 50. The console 70 has a processing circuit 71, a PET data memory 72, an input circuit 73, a display circuit 74, and a memory 75. Data communication among the processing circuit 71, the PET data memory 72, the input circuit 73, the display circuit 74, and the memory 75 is performed via, for example, a wired bus or wireless. Here, the processing circuit 71 is an example of a medical image processing apparatus. The console 70 may be expressed as an example of the medical image processing apparatus.

  The processing circuit 71 controls the overall operation of the PET device 1. The processing circuit 71 executes a program related to correction of a failure image obtained when the detector 110 is in a failure state (hereinafter referred to as a failure image correction program) to perform processing related to machine learning model learning and / or reinforcement learning. And a process related to correction of a failure image using a machine learning model. The processing circuit 71 has, as hardware resources, a processor such as a Central Processing Unit (CPU), a Micro Processing Unit (MPU), and a Graphics Processing Unit (GPU), and a Read Only Memory (ROM) and a Random Access Memory (RAM). And a memory.

  The processing circuit 71 according to the present embodiment executes the reconstruction function 711, the image processing function 712, the imaging control function 713, the display control function 714, the correction processing function 716, and the like by the failure image correction program. Further, the processing circuit 71 executes the model generation function 715 and the like by the learning program. Here, the processing circuit 71 that realizes the reconstruction function 711 is an example of an acquisition unit. The processing circuit 71 that realizes the imaging control function 713 is an example of a specifying unit. The processing circuit 71 that realizes the correction processing function 716 is an example of a processing unit.

  In the reconstruction function 711, the processing circuit 71 reconstructs a PET image showing the distribution of the positron emitting nuclides administered to the subject P, based on the coincidence counting event data acquired from the PET data memory 72. Here, the PET image includes a failure image and a simulated failure image. For example, the processing circuit 71 reconstructs a PET image (hereinafter referred to as a failure image) based on the failure data acquired from the PET data memory 72. The failure data is coincidence counting event data obtained when one of the detectors 110 has a failure. For example, the processing circuit 71 reconstructs a failure image (hereinafter referred to as simulated failure image) based on the failure data (hereinafter referred to as simulated failure data) generated by the model generation function 715. As the image reconstruction algorithm, an existing image reconstruction algorithm such as a Filtered Back Projection (FBP) method or a successive approximation reconstruction method may be used. The processing circuit 71 outputs the generated PET image, simulated failure image, and failure image to the memory 75 and / or the display circuit 74. Here, the failure image is an example of first medical data. Further, the detector 110 that has failed at the time of PET imaging in which the failure data used for reconstructing the failure image is collected is an example of the first PET detector.

  In the image processing function 712, the processing circuit 71 performs various image processing on the PET image reconstructed by the reconstruction function 711. For example, the processing circuit 71 performs three-dimensional image processing such as volume rendering, surface volume rendering, pixel value projection processing, Multi-Planer Reconstruction (MPR) processing, Curved MPR (CPR) processing on the PET image to display the display image. To generate.

  In the imaging control function 713, the processing circuit 71 synchronously controls the gantry 10 and the bed 50 to perform PET imaging. The processing circuit 71 can execute the positioning scan by the gantry 10 and the setting of the imaging range. The processing circuit 71 also controls the operation of each unit during the degenerate operation. During the degenerate operation, the processing circuit 71 performs control related to PET imaging with the failed detector 110 stopped. The processing circuit 71 stops the failed detector 110 based on the error message output from the gantry 10.

  In the display control function 714, the processing circuit 71 displays various information on the display circuit 74. For example, the processing circuit 71 displays the PET image reconstructed by the reconstruction function 711 on the display circuit 74. Further, the processing circuit 71 displays a GUI for various settings and inputs on the display circuit 74. Further, during the degenerate operation, the processing circuit 71 displays, on the display circuit 74, a notification that the image has been corrected using the machine learning model, the detector number of the detector 110 that has failed, and the like.

  In the model generation function 715, the processing circuit 71 uses the collected data read from the PET data memory 72 to generate simulated failure data for each detector. The processing circuit 71 generates learning data using the simulated failure image reconstructed from the simulated failure data, the normal image reconstructed from the original collected data, and the detector number corresponding to the simulated failure data. . The processing circuit 71 uses the generated learning data to learn and / or reinforce the machine learning model. Here, the simulated failure image is an example of the first medical data for learning. The normal image is an example of the second medical data for learning. Further, the detector 110 indicated by the detector number corresponding to the simulated failure data is an example of the first PET detector.

  In the correction processing function 716, the processing circuit 71 selects the machine learning model stored in the memory 75 according to the acquired failure detector number. The processing circuit 71 inputs the failure image generated by the reconstruction function 711 to the selected machine learning model. The processing circuit 71 acquires, as a corrected image, the output of the learned machine learning model to which the failure image has been input. The processing circuit 71 outputs the generated corrected image to the memory 75 and / or the display circuit 74. Here, the corrected image is an example of second medical data.

  The processing circuit 71 may be realized by an ASIC or FPGA. Further, the processing circuit 71 may be realized by CPLD or SPLD.

  The functions 711 to 716 are not limited to being realized by a single processing circuit. A plurality of independent processors may be combined to configure a processing circuit, and each processor may execute a program to realize each of the functions 711 to 716.

  The PET data memory 72 is a storage device that stores the single event data and the coincidence counting event data transmitted from the gantry 10. The PET data memory 72 is a storage device such as a hard disk drive (HDD), a solid state drive (SSD), and an integrated circuit storage device. The PET data memory 72 reads / writes various information from / to a portable storage medium such as a Compact Disc (CD) -ROM drive, a Digital Versatile Disc (DVD) drive, and a flash memory, in addition to the HDD and SSD. It may be a driving device or the like. The storage area of the PET data memory 72 may be in an external storage device connected by a network. The event data stored in the PET data memory 72 may be in any data format such as list mode data or histogram data. In the present embodiment, these data stored in the PET data memory 72 are collectively referred to as collected data.

  The input circuit 73 receives various input operations from the operator, converts the received input operations into electric signals, and outputs the electric signals to the processing circuit 71. A keyboard, a mouse, various switches, or the like can be used as the input circuit 73. The input circuit 73 supplies the output signal from the input device to the processing circuit 71 via the bus. Specifically, the input circuit 73 inputs examination contents such as an examination name and imaging conditions according to an operator's instruction prior to performing PET imaging on the subject P. The examination content may be examination order information input from a radiology information system (RIS) via a network. At this time, the processing circuit 71 extracts the inspection content from the order information. The input circuit 73 may also input the use or image quality of the PET image according to an instruction from the operator.

  The display circuit 74 displays various information under the control of the processing circuit 71 that realizes the display control function 714. The display circuit 74 displays, for example, a Graphical User Interface (GUI) related to input of inspection content and imaging conditions. The display circuit 74 displays the PET image generated by the processing circuit 71 that realizes the reconstruction function 711 and the medical image generated by the processing circuit 71 that realizes the image processing function 712. The display circuit 74 may display, for example, a GUI relating to an input such as image quality or usage. The display circuit 74 has a display interface circuit and a display device. The display interface circuit converts data (display information) representing a display target into a video signal. The display signal is supplied to the display device. The display device displays a video signal representing a display target. As the display device, various arbitrary one or two or more displays can be appropriately used. For example, as a display device, a CRT (Cathode Ray Tube) display, a liquid crystal display (Liquid Crystal Display: LCD), an organic EL display (Organic Electro Luminescence Display: OELD), a light emitting diode (LED) display, or a plasma display can be appropriately used. is there.

  The display circuit 74 may be provided at any place in the control room. The display circuit 74 may be a desktop type or a tablet terminal or the like capable of wireless communication. Further, one or more projectors may be used as the display device of the display circuit 74.

  The memory 75 is a storage device such as an HDD, SSD, or integrated circuit storage device. Further, the memory 75 may be a drive device that reads and writes various information from and to a portable storage medium such as a CD-ROM drive, a DVD drive, and a flash memory, in addition to the HDD and SSD. The storage area of the memory 75 may be in an external storage device connected by a network.

  The memory 75 stores various conditions and information input via the input circuit 73, a program corresponding to each of a plurality of functions executed in the processing circuit 71, and data being processed when each function is executed in the processing circuit 71. Memorize The memory 75 also stores various PET images generated by the reconstruction function 711, display images generated by the image processing function 712, and simulated failure data and learning data generated by the model generation function 715. The memory 75 also stores a machine learning model. The machine learning model according to the present embodiment includes a plurality of models corresponding to the plurality of detectors 110.

  Hereinafter, an operation example of each unit of the PET device 1 according to the present embodiment will be described in detail with reference to the drawings.

  First, generation of the learning data set according to this embodiment will be described. FIG. 3 is a flowchart showing an example of the learning data generation process executed in the PET device 1 of FIG. FIG. 4 is a diagram for explaining learning and / or reinforcement learning of the machine learning model 820 using the learning data set 810 generated in the learning data generation process of FIG.

  It is assumed that this processing is performed after PET imaging is performed in the PET device 1. That is, it is assumed that the collected data obtained by PET imaging is stored in the PET data memory 72 when this processing is started. Here, it is assumed that the collected data is event data after simultaneous counting. Further, it is assumed that the acquired data is a set of event data from the start to the end of PET imaging, which is obtained by PET imaging in a state where all detectors 110 are operating normally.

  Hereinafter, for simplicity of explanation, in the PET device 1 according to the present embodiment, each of the plurality of detector rings 11 has N detectors 110. It is assumed that the N detectors 110 are assigned detector numbers 1 to N, respectively.

  Further, the following description will be made with respect to an arbitrary detector ring 11 among the plurality of detector rings 11. At this time, the other detector rings 11 are also treated in the same manner. For example, the description regarding any one detector 110 can be regarded as the description regarding the plurality of detectors 110 arranged in the body axis direction.

  In step S101, the processing circuit 71 that realizes the model generation function 715 reads the collected data from the PET data memory 72.

  In step S102, the processing circuit 71 that realizes the model generation function 715 uses the collected data read in step S101 to generate simulated failure data. Here, the simulated failure data is data that simulates the collected data obtained when any one of the plurality of detectors 110 fails. The processing circuit 71, which realizes the model generation function 715, generates N simulated fault data for the N detectors 110 by excluding the collected data of the respective detectors 110. For example, the processing circuit 71 removes the LOR event data regarding the arbitrary detector 110 that has artificially failed from the collected data to generate the simulated failure data regarding the relevant detector.

  In step S103, the processing circuit 71 that realizes the reconstruction function 711 generates a normal image 813 by performing image reconstruction on the collected data read in step S101. That is, the normal image 813 is a reconstructed image obtained by reconstructing the original collected data. That is, the normal image 813 is a reconstructed image based on the event data collected when all the detectors 110 are operating normally.

  In step S104, the processing circuit 71 that realizes the reconstruction function 711 generates the simulated failure image 817 by performing image reconstruction on each of the N simulated failure data generated in step S102. That is, the simulated failure image 817 is a reconstructed image obtained by reconstructing the simulated failure data.

  In step S105, the processing circuit 71 that realizes the model generation function 715 generates the learning data set 810. As shown in FIG. 4, the processing circuit 71 that implements the model generation function 715 converts each of the N simulated fault images 817 generated in step S104 into the normal image generated in each detector number 815 and step S103. In association with 813, N pieces of learning data 811 are generated. The processing circuit 71 that realizes the model generation function 715 generates a learning data set 810 including the N pieces of generated learning data 811.

  In step S106, the processing circuit 71 that realizes the model generation function 715 stores the learning data set 810 generated in step S105 in the memory 75.

  Although a case has been described as an example where the learning data generation process is performed on arbitrary collected data, the learning data set 810 includes a plurality of sets of learning data 811 generated from a plurality of sets of collected data. . The plurality of sets of learning data 811 generated from the plurality of sets of collected data may be associated with each detector number 815 or may be grouped and stored.

  Here, learning and / or reinforcement learning of the machine learning model 820 according to this embodiment will be described. The processing circuit 71 that realizes the model generation function 715 reads out the learning data set 810 generated by the learning data generation process from the memory 75. As shown in FIG. 4, the processing circuit 71 that implements the model generation function 715 uses the simulated failure image 817 and the normal image 813 to learn and / or reinforce the machine learning model 820 for each detector number 815. . Specifically, the processing circuit 71 that realizes the model generation function 715 inputs the simulated failure image 817 to the input end of the machine learning model 821 and the normal image 813 to the output end for each detector number 815. In this way, a plurality of machine learning models 821 corresponding to a plurality of detector numbers 815 are generated. That is, the machine learning model 821 can be expressed as the machine learning model 820 for each detector number 815.

  Next, an operation example of each unit in the degenerate operation performed when a failure of any one of the plurality of detectors 110 is detected will be described. FIG. 5 is a flowchart showing an example of the degenerate operation process executed in the PET apparatus 1 of FIG. FIG. 6 is a diagram for explaining image correction using the machine learning model 820 executed in the degenerate operation process of FIG.

  In step S201, the processing circuit 71 that realizes the imaging control function 713 determines whether to perform the degenerate operation. In this determination, for example, when the error message is received from the gantry 10, it is determined that the degenerate operation is performed. The error message is output when the output of any of the detectors 110 is equal to or less than a predetermined threshold, for example, the output of any of the detectors 110 cannot be detected. It is assumed that the predetermined threshold value is preset and stored in the storage area in the signal processing circuit 13 of the gantry 10, the memory 75, or the like. The process repeats the process of this step until it is determined that the degenerate operation is performed. At this time, the PET apparatus 1 performs normal PET imaging and the like. The process proceeds to step S202 when it is determined that the degenerate operation is performed.

  When it is determined in this step that the degenerate operation is performed, the processing circuit 71 that realizes the display control function 714 may notify the user that the degenerate operation is performed. At this time, the processing circuit 71 that realizes the display control function 714 may further notify the user that the image correction is performed using the machine learning model.

  The error message is not limited to when the output of any of the detectors 110 is equal to or lower than the predetermined threshold value, and when the temperature of any of the detectors 110 is equal to or higher than the predetermined threshold value. It may be output when the output is equal to or more than a predetermined threshold value.

  It should be noted that this determination is not limited to whether or not an error message is received, and may be made based on the output of the input circuit 73 in response to a user operation. At this time, the processing circuit 71 that realizes the display control function 714 displays the GUI for operation prior to the notification that guides the implementation of the degenerate operation.

  Hereinafter, as an example, in step S201, a failure is detected with respect to the two detectors 110 having the detector number 1 and the detector number 3 among the N detectors 110 having the detector numbers 1 to N. Continue the explanation.

  In step S202, the processing circuit 71 that realizes the imaging control function 713 stops the detectors 110 having the detector numbers 1 and 3. Here, the stop of the detector 110 is realized, for example, by the signal processing circuit 13 not reading the detection value from the detector 110.

  Note that stopping the detector 110 may be realized, for example, by stopping the supply of power to the detector 110, by not performing coincidence counting, or by recollecting the collected data as an image. It may be realized by not using it in the configuration.

  In step S203, the processing circuit 71 that realizes the imaging control function 713 executes the PET imaging with the detectors 110 having the detector numbers 1 and 3 stopped. The PET imaging collects failure data regarding a state in which the detectors 110 having the detector numbers 1 and 3 are stopped. The failure data is stored in the PET data memory 72. The simulated failure data used for learning the machine learning model 820 can be expressed as collected data that simulates the failure data obtained in this step.

  In step S204, the processing circuit 71 that realizes the reconstruction function 711 generates the failure image 801. The failure image 801 is a reconstructed image obtained by reconstructing the failure data read from the PET data memory 72. In the present embodiment, the failure image 801 is a reconstructed image based on the failure data collected when the detectors 110 having the detector numbers 1 and 3 are stopped. That is, the failure image 801 is a reconstructed image based on the set of collected data in which the LOR event data regarding the detectors 110 having the detector numbers 1 and 3 is missing. Therefore, the failure image 801 has an artifact component due to the lack of the LOR for the detectors 110 of detector number 1 and detector number 3. The simulated failure image 817 used for learning the machine learning model 820 can be expressed as a reconstructed image that simulates the failure image 801 generated in this step.

  In step S205, the processing circuit 71 that realizes the correction processing function 716 acquires the failure detector number 803. The failure detector number 803 is the detector number of the detector 110 that has failed at the time of PET imaging in step S203. From this, the processing circuit 71 that realizes the correction processing function 716 can also be expressed as an example of the specifying unit.

  In step S206, the processing circuit 71 that realizes the correction processing function 716 selects the machine learning model 820 according to the failure detector number 803. For example, when the failure detector number 803 is the detector number 1 and the detector number 3, the machine learning model 820 selected according to the failure detector number 803 is the machine learning model 821 and the detector related to the detector number 1. It is a machine learning model 821 related to number 3. As shown in FIG. 6, in the selected machine learning model 823, the machine learning model 821 regarding the detector number 1 and the machine learning model 821 regarding the detector number 3 are connected in series.

  In step S207, the processing circuit 71 that realizes the correction processing function 716 generates the corrected image 805 by using the selected machine learning model 823. Specifically, the processing circuit 71 that realizes the correction processing function 716 inputs the failure image 801 to the machine learning model 821 regarding the detector number 1. The output of the machine learning model 821 regarding the detector number 1 is input to the machine learning model 821 regarding the detector number 3. The processing circuit 71 that realizes the correction processing function 716 acquires the output of the machine learning model 821 regarding the detector number 3 as the corrected image 805. As described above, the corrected image 805 is the output of the selected machine learning model 823 when the failure image 801 is input, and has an image quality equivalent to that of the normal image 813 used for learning of the machine learning model 820. It can be expressed as a possessed image. The corrected image 805 is based on the event data collected in a state where the detectors 110 having the detector numbers 1 and 3 are operating normally, that is, all the detectors 110 are operating normally. It has the same image quality as the reconstructed image.

  In step S208, the processing circuit 71 that realizes the display control function 714 displays the corrected image 805 generated in step S207 on the display circuit 74. Here, the display image displayed on the display circuit 74 is not limited to the corrected image 805, but a notification indicating that the corrected image 805 is an image corrected using the machine learning model 820 or a failure image 801 before correction, The failure detector number 803 and the like may be included.

  As described above, in the PET apparatus 1 which is an example of the nuclear medicine diagnostic apparatus in which the medical image processing apparatus according to the present embodiment is installed, the processing circuit 71 assumes a state in which one of the detectors 110 is out of order. The machine failure model 820 is learned and / or the reinforcement learning is performed for each detector number 815 by using the simulated failure image 817 and the normal image 813 in a state where all the detectors 110 are normal. Further, the processing circuit 71 selects the machine learning model 820 according to the failure detector number 803. Further, the processing circuit 71 corrects the failure image 801 using the machine learning model 820 selected according to the failed detector 110. According to this configuration, it is possible to generate image data having an image quality equivalent to that in a normal state even when the detector 110 is out of order.

  Note that, in the example shown in FIG. 6, the machine learning model 821 regarding the detector number 3 is connected to the subsequent stage of the machine learning model 821 regarding the detector number 1, but the invention is not limited to this. For example, the machine learning model 821 related to the detector number 1 may be connected to the subsequent stage of the machine learning model 821 related to the detector number 3. Further, the processing circuit 71 that implements the correction processing function 716 may determine the order of the machine learning model 821 based on the output of the input circuit 73 according to the input of the user. Even with such a configuration, the same effect as described above can be obtained.

  The processing circuit 71 that implements the correction processing function 716 may obtain a plurality of corrected images 805 from the machine learning model 821 connected in a plurality of orders. The processing circuit 71 that realizes the display control function 714 displays the acquired plurality of corrected images 805 on the display circuit 74. According to such a configuration, in addition to the effects described above, the user can confirm the plurality of corrected images 805 and select a more appropriate corrected image 805.

  Further, the memory 75 may store a machine learning model for model selection. The processing circuit 71 that implements the model generation function 715 uses the fault detector number 803 and the connection order of the machine learning model 821 used to generate the corrected image 805 selected by the user to select a model. Learn and / or reinforce machine learning models. At this time, the processing circuit 71 that realizes the correction processing function 716 determines the selected machine learning model 823 according to the output of the machine learning model for model selection to which the failure detector number 803 is input. According to such a configuration, in addition to the effects described above, there is an effect that an appropriate corrected image 805 can be easily generated.

  Further, the machine learning model 820 according to the present embodiment may include the machine learning model for model selection described above. At this time, the processing circuit 71 that realizes the model generation function 715 uses the failure image 801, the failure detector number 803, the correction image 805 selected by the user, and the machine used to generate the selected correction image 805. The machine learning model 820 is learned and / or reinforcement-learned using the connection order of the learning model 821. At this time, the processing circuit 71 that realizes the correction processing function 716 acquires the output of the machine learning model 820 to which the failure image 801 and the failure detector number 803 are input as the correction image 805. According to such a configuration, in addition to the effects described above, there is an effect that an appropriate corrected image 805 can be generated more easily.

[Second Embodiment]
Hereinafter, a medical image processing apparatus and a medical image processing method according to the present embodiment will be described with reference to the drawings. Here, differences from the first embodiment will be mainly described. In the following description, constituent elements having the same or substantially the same functions as those in the first embodiment are designated by the same reference numerals and will be redundantly described only when necessary.

  In the first embodiment, the machine learning model 821 of the machine learning model 820 that has been trained and / or reinforced for each detector number 815 is selected according to the failure detector number 803, and the selected machine learning model 823 is selected. The case where the failure image 801 is corrected by using is described as an example. On the other hand, the machine learning model may be trained and / or reinforcement-learned with respect to the combination of the detector numbers according to the assumed failure pattern.

  First, the generation of the learning data set 830 according to this embodiment will be described. FIG. 7 is a diagram for explaining learning and / or reinforcement learning of the machine learning model 840 and image correction using the machine learning model 840 according to this embodiment. Note that the following description will be given while comparing with the learning data generation processing according to the first embodiment of FIG.

  The processing circuit 71 that realizes the model generation function 715 reads the collected data from the PET data memory 72 in the same manner as in step S101. The processing circuit 71 that realizes the model generation function 715 generates simulated failure data using the read collected data. In this embodiment, unlike step S102 in which simulated failure data is generated for each detector 110, a plurality of simulated failure data corresponding to a plurality of assumed failure patterns is generated. The assumed failure pattern includes, for example, a state in which any one detector 110 is out of order and a state in which any two or more detectors 110 are out of order. It is assumed that the detector 110 or the combination of the detectors 110 according to the assumed failure pattern is preset and stored in the memory 75. The processing circuit 71 that realizes the reconstruction function 711 generates a normal image 833 and a plurality of simulated failure images 837 in the same manner as in step S103 and step S104. That is, the normal image 833 and the simulated failure image 837 are reconstructed images obtained by image reconstruction of the original collected data and the simulated failure data, respectively. The processing circuit 71 that realizes the model generation function 715 generates the learning data set 830 as in step S105. As shown in FIG. 7, the learning data set 830 includes a plurality of learning data 831 corresponding to a plurality of failure patterns. Similar to the learning data 811, each learning data 831 includes a normal image 833, a detector number 835, and a simulated failure image 837. The generated learning data set 830 is stored in the memory 75.

  Next, learning and / or reinforcement learning of the machine learning model 840 according to this embodiment will be described with reference to FIG. 7. The processing circuit 71 that realizes the model generation function 715 reads out the learning data set 830 generated by the learning data generation process from the memory 75. As shown in FIG. 7, the processing circuit 71 that implements the model generation function 715 uses the learning data 831 to learn and / or reinforce the machine learning model 840. Specifically, the processing circuit 71 that realizes the model generation function 715 inputs the detector number 835 and the simulated failure image 837 to the input end of the machine learning model 840 for each learning data 831 and outputs the normal image to the output end. Enter 833. In this way, the machine learning model 840 corresponding to the plurality of failure patterns is generated.

  Next, an operation example of each unit in the degenerate operation according to the present embodiment will be described with reference to FIG. 7. The following description will be given in comparison with the degenerate operation process according to the first embodiment of FIG. Hereinafter, a case where a failure is detected with respect to two detectors 110 having a detector number 3 and a detector number 5 among the N detectors 110 having the detector numbers 1 to N will be described as an example. Here, the two detectors 110 having the detector number 3 and the detector number 5 are examples of the first PET detector and the second PET detector.

  The processing circuit 71 that realizes the imaging control function 713 determines whether to perform the degenerate operation, as in step S201. After the degenerate operation is started, the processing circuit 71 that realizes the imaging control function 713 stops the two detectors 110 of the detector number 3 and the detector number 5 in which the failure is detected, as in step S202. . The processing circuit 71 that implements the imaging control function 713 executes PET imaging and acquires failure data, as in step S203. The processing circuit 71 that realizes the reconstruction function 711 generates a failure image 801 by image reconstruction regarding the acquired failure data in the same manner as in step S204. The processing circuit 71 that realizes the correction processing function 716 acquires the failure detector number 803 in the same manner as in step S205.

  The machine learning model 840 according to the present embodiment is not learned for each detector number, unlike the machine learning model 820 according to the first embodiment. Therefore, in the degenerate operation process according to the present embodiment, the process corresponding to step S206 is not executed.

  After that, the processing circuit 71 that realizes the correction processing function 716 generates the corrected image 805 using the machine learning model 840. Specifically, the processing circuit 71 that realizes the correction processing function 716 acquires the output of the machine learning model 840 to which the failure image 801 and the failure detector number 803 are input as the correction image 805. The processing circuit 71 that realizes the display control function 714 displays the generated corrected image 805 and the like on the display circuit 74, as in step S208.

  As described above, in the PET apparatus 1 which is an example of the nuclear medicine diagnostic apparatus equipped with the medical image processing apparatus according to the present embodiment, the processing circuit 71 includes the simulated failure image 837 assuming a plurality of failure patterns and the detector. The machine learning model 840 is trained and / or reinforcement-learned using the number 835 and the normal image 833 in a state where all the detectors 110 are normal. Further, the processing circuit 71 corrects the failure image 801 by using the machine learning model 840 learned according to the failure pattern. According to this configuration, since the machine learning model 840 is learned using the learning data 831 corresponding to the combination of the failed detectors 110, it is possible to more accurately generate the image data of the image quality corresponding to the normal time. You can

[Third Embodiment]
Hereinafter, a medical image processing apparatus and a medical image processing method according to the present embodiment will be described with reference to the drawings. Here, differences from the first embodiment will be mainly described. In the following description, constituent elements having the same or substantially the same functions as those in the first embodiment are designated by the same reference numerals and will be redundantly described only when necessary.

  In the first embodiment, the case where the machine learning model 820 is learned and / or the reinforcement learning is performed for each detector number 815 has been described as an example. On the other hand, the machine learning model may be trained and / or reinforcement learned for any one detector 110.

  First, the generation of the learning data set 850 according to this embodiment will be described. FIG. 8 is a diagram for explaining learning and / or reinforcement learning of the machine learning model 860 according to this embodiment and image correction using the machine learning model 860. Note that the following description will be given while comparing with the learning data generation processing according to the first embodiment of FIG. Hereinafter, a case where the machine learning model is learned and / or the reinforcement learning is assumed assuming a failure of the detector 110 having the detector number 3 will be described as an example.

  The processing circuit 71 that realizes the model generation function 715 reads the collected data from the PET data memory 72 in the same manner as in step S101. The processing circuit 71 that realizes the model generation function 715 generates simulated failure data using the read collected data. In this embodiment, unlike step S102 in which simulated fault data for each detector 110 is generated, simulated fault data for detector number 3 is generated by excluding the collected data for detector 110 having detector number 3. It The processing circuit 71 that realizes the reconstruction function 711 generates the normal image 853 and the simulated failure image 857 in the same manner as in step S103 and step S104. That is, the normal image 853 and the simulated failure image 857 are reconstructed images obtained by image reconstruction of the original collected data and the simulated failure data regarding the detector number 3, respectively. The processing circuit 71 that realizes the model generation function 715 generates the learning data 851 regarding the detector number 3. The learning data set 850 according to the present embodiment may include a plurality of learning data 851 generated assuming a failure of the same detector 110 with respect to a plurality of collected data. The learning data 851 includes a normal image 853 and a simulated failure image 857. The learning data 851 according to the present embodiment is generated assuming a failure of any one detector 110. Therefore, the learning data 851 does not include the detector number, unlike the learning data 811 generated in step S105. However, in the example shown in FIG. 8, the detector number 855 is shown in the learning data 851 in order to show the detector 110 that is assumed to have a failure when the learning data 851 is generated. The generated learning data set 850 is stored in the memory 75.

  Next, learning and / or reinforcement learning of the machine learning model 860 according to this embodiment will be described with reference to FIG. The processing circuit 71 that realizes the model generation function 715 reads out the learning data set 850 generated by the learning data generation process from the memory 75. As shown in FIG. 8, the processing circuit 71 that realizes the model generation function 715 uses the learning data 851 to learn and / or reinforce the machine learning model 860. Specifically, the processing circuit 71 that realizes the model generation function 715 inputs the simulated failure image 857 to the input end of the machine learning model 860 and the normal image 853 to the output end of the learning data 851. In this way, the machine learning model 840 assuming a failure of any one detector 110 is generated.

  Next, an operation example of each unit in the degenerate operation according to the present embodiment will be described with reference to FIG. The following description will be given in comparison with the degenerate operation process according to the first embodiment of FIG. Hereinafter, a case where a failure is detected with respect to the detector 110 having the detector number 5 out of the N detectors 110 having the detector numbers 1 to N will be described as an example.

  The processing circuit 71 that realizes the imaging control function 713 determines whether to perform the degenerate operation, as in step S201. After the degenerate operation is started, the processing circuit 71 that realizes the imaging control function 713 stops the detector 110 of the detector number 5 in which the failure is detected, as in step S202. The processing circuit 71 that implements the imaging control function 713 executes PET imaging and acquires failure data, as in step S203. The processing circuit 71 that realizes the reconstruction function 711 generates a failure image 801 by image reconstruction regarding the acquired failure data in the same manner as in step S204. The processing circuit 71 that realizes the correction processing function 716 acquires the failure detector number 803 in the same manner as in step S205.

  Note that the machine learning model 860 according to the present embodiment is not learned for each detector number, unlike the machine learning model 820 according to the first embodiment. Therefore, in the degenerate operation process according to the present embodiment, the process corresponding to step S206 is not executed. On the other hand, in the degenerate operation process according to the present embodiment, the detector 110 that is assumed to have a failure when the learning data 851 is generated may be different from the actually failed detector 110 indicated by the failure detector number 803.

  Therefore, the processing circuit 71 that realizes the correction processing function 716 according to the present embodiment converts the failure image 801 into a failure image 802 corresponding to the detector number 855 of the detector 110 that is assumed to have a failure when the learning data 851 is generated. Convert. Here, the processing circuit 71 that realizes the correction processing function 716 according to the present embodiment is an example of a correction unit. Specifically, the processing circuit 71, which realizes the correction processing function 716, performs a rotation process for rotating the coordinate system on the failure image 801 relating to the detector number 5, so that the failure image 802 corresponding to the detector number 3 is obtained. To generate. The entire failure image 801 may be rotated, or only the artifact component included in the failure image 801 due to the lack of the LOR regarding the detector 110 having the detector number 5 may be rotated. Here, the rotation process is performed with respect to the attachment angle (angle around the central axis of the detector ring 11) in the detector ring 11 between the detector 110 having the detector number 3 and the detector 110 having the detector number 5. It is based on the difference. The mounting angle of the detector 110 on the detector ring 11 can be defined by, for example, the direction from each detector 110 toward the center of the detector ring 11. The detector number 855 of the detector 110 and the attachment angle of each detector 110, which is assumed to have a failure when the learning data 851 is generated, are stored in the memory 75, for example.

  The angle of attachment of each detector 110 is not limited to the angle of attachment of each detector 110, and the difference in the angle of attachment of each detector 110 to the detector 110 that is assumed to have a failure when the learning data 851 is generated may be stored. Further, when the relative angles of the two adjacent detectors 110 are constant over the entire circumference of the detector ring 11, the relative angles of the two adjacent detectors 110 may be stored. Further, the mounting angle (angle around the central axis of the detector ring 11) is correlated with the position of the detector 110 assuming a failure. Therefore, instead of the attachment angle, information about the attachment position (position within the detector ring 11) may be stored. The information about the mounting angle and the mounting position is an example of the position information.

  After that, the processing circuit 71 that realizes the correction processing function 716 generates the corrected image 804 corresponding to the detector number 3 by using the machine learning model 860. Specifically, the processing circuit 71 that realizes the correction processing function 716 outputs the output of the machine learning model 860 to which the failure image 802 corresponding to the detector number 3 is input as the corrected image 804 corresponding to the detector number 3. get. The processing circuit 71 that realizes the correction processing function 716 performs a rotation process on the correction image 804 corresponding to the detector number 3 to generate the correction image 805 for the detector number 5. In the rotation process performed on the correction image 804, the entire correction image 804 may be rotated according to the rotation process performed on the failure image 801, or the detector number included in the correction image 804 Only the artifact component due to the lack of detector 110 for 3 may be rotated. The processing circuit 71 that realizes the display control function 714 displays the generated corrected image 805 and the like on the display circuit 74, as in step S208.

  In the present embodiment, the case where the failed detector 110 is the one detector 110 with the detector number 5 has been described as an example, but the present technology is not limited to the case where two or more detectors have failed. Is also applicable. For example, assume that the fault detector number 803 includes the detector number 7 in addition to the detector number 5. At this time, the corrected image 805 is corrected with respect to the detector number 5 as described above, but is not corrected with respect to the detector number 7. Therefore, the processing circuit 71 that implements the correction processing function 716 further executes the correction processing using the machine learning model 860, using the corrected image 805 as the failure image 801 related to the detector number 7. At this time, the corrected image 805 is subjected to rotation processing based on the relative angle between the two detectors 110 having the detector number 3 and the detector number 7. The corrected image 804 corresponding to the detector number 3 may be used as the failure image 801 related to the detector number 7. At this time, the rotation process may be performed based on the difference between the relative angles of the detector number 5 and the detector number 7 with respect to the detector number 3.

  Note that the case has been described as an example where the learning data 851 is generated for any one of the plurality of detectors 110, but the present invention is not limited to this. For example, the simulated failure image corresponding to the detector number 3 may be generated by converting the simulated failure image obtained by reconstructing the simulated failure data regarding the detector number 9. According to this configuration, it is possible to generate the learning data 851 for the number of the detectors 110 from one collected data. That is, there is an effect that the learning data set 850 can be easily generated.

  In the present embodiment, the case where the failure image 802 corresponding to the detector number 3 is generated by rotating the failure image 801 related to the detector number 5 has been described as an example, but the present invention is not limited to this. For example, the failure image 802 corresponding to detector number 3 may be generated by image reconstruction of simulated failure data for detector number 5 rotated to correspond to detector number 3.

  As described above, in the PET apparatus 1 which is an example of the nuclear medicine diagnostic apparatus in which the medical image processing apparatus according to the present embodiment is installed, the processing circuit 71 includes the simulated failure image simulating the failure of any one detector 110. The machine learning model 860 is trained and / or the reinforcement learning is performed using 857 and the normal image 853 in the state where all the detectors 110 are normal. Further, the processing circuit 71 uses the machine learning model 840 learned about the failure of any one detector 110 to generate the failure image 802 corresponding to the detector number 855 that assumed the failure when the learning data 851 was generated. to correct. Further, the processing circuit 71 converts the corrected failure image 804 to generate a corrected image 805. With this configuration, the machine learning model 860 can be more easily learned and / or the reinforcement learning can be performed.

  In the medical image processing apparatus and the medical image processing method according to each of the above embodiments, as shown in FIG. 5, after PET imaging is performed in step S203, the failure detector number 803 is acquired in step S205. For this reason, even if the detector 110 newly fails during PET imaging, it is possible to generate image data having an image quality equivalent to that in a normal state. In each of the above embodiments, the case where the image correction using the machine learning model is executed during the degenerate operation has been described as an example, but the present invention is not limited to this. Even if the detector 110 fails during normal PET imaging, since the failure detector number 803 is acquired after PET imaging, it is possible to generate image data of image quality corresponding to normal times.

  In the medical image processing apparatus and the medical image processing method according to each of the above-described embodiments, the notification image, the failure image 801 before correction, and the failure detector number 803 are displayed together with the correction image 805. According to this configuration, the user can recognize that the image data having the image quality corresponding to the normal time is the image corrected by using the machine learning model. The user can also determine whether to perform PET imaging again after repairing the failed detector 110.

  In the medical image processing device and the medical image processing method according to each of the above-described embodiments, each machine learning model is a failure image 801 obtained during degenerate operation or the like, and a machine learning model to which the failure image 801 is input. Reinforcement learning may be performed using the output corrected image 805.

  In each of the above-described embodiments, a medical image processing apparatus and a medical image processing method for generating learning data using simulated failure data generated by excluding collected data relating to an arbitrary detector number will be described as an example. However, it is not limited to this. The simulated failure data may be collected by performing PET imaging with the detector 110 having an arbitrary detector number stopped. According to this configuration, since the simulated failure data can be obtained in the same state as that at the time of the failure, the simulated failure data closer to the failure data obtained at the time of the failure can be generated.

  In each of the above embodiments, the medical image processing apparatus and the medical image processing method in which the plurality of detectors 110 arranged in the body axis direction across the plurality of detector rings 11 are handled in the same manner have been described as examples. Not limited to The technique according to each of the above embodiments may be applied to all the detectors 110 of the plurality of detector rings 11. For example, in the PET apparatus 1 according to the first embodiment, learning data 811 regarding all the detectors 110 of the plurality of detector rings 11 is generated. For example, in the PET device 1 according to the second embodiment, a failure pattern is set for all the detectors 110 of the plurality of detector rings 11.

  In each of the above embodiments, the medical image processing apparatus and the medical image processing method for correcting the failure image 801 obtained at the time of failure using the machine learning model have been described as examples, but the invention is not limited to this. The failure image 801 may be corrected, for example, by probabilistically interpolating the number of LORs. The stochastic interpolation of the number of LORs is performed based on, for example, the coincidence counting time width (Time Window), the dose of the radioactive isotope, the administration position of the radioactive isotope, and the like.

  In each of the above embodiments, the medical image processing apparatus and the medical image processing method in which the reconstructed image is used as the input and output of the machine learning model have been described as examples, but the present invention is not limited to this. The input and / or output of the machine learning model may be raw data such as collected data, simulated failure data, failure data and the like. Even with this configuration, the same effect as that of the above-described embodiment can be obtained. Further, the raw data may be data before simultaneous counting. In the case of this configuration, the number of data is large, but since the counter data of the failure detector is included, more accurate correction can be performed.

  In the medical image processing apparatus and the medical image processing method according to each of the above embodiments, the collected data and the PET image are collected and re-collected outside the PET apparatus 1 equipped with the medical image processing apparatus according to each of the embodiments. It may be configured.

  In each of the above embodiments, the case where the medical image processing apparatus is installed in the PET apparatus 1 has been described as an example, but the present invention is not limited to this. The medical image processing apparatus according to each embodiment may be installed in, for example, a nuclear medicine diagnosis apparatus other than a PET apparatus, an image processing server provided outside the nuclear medicine diagnosis apparatus, or the like. Further, the medical image processing apparatus according to each embodiment may be configured as a single independent apparatus. As long as the medical image processing apparatus has at least a storage circuit corresponding to the PET data memory 72 and the memory 75 and a processing circuit corresponding to the processing circuit 71 that implements the model generation function 715 and the correction processing function 716. Good. When accompanied by a user operation, the medical image processing apparatus further includes an input circuit and a display circuit corresponding to the input circuit 73 and the display circuit 74, respectively.

  According to at least one embodiment described above, it is possible to generate image data having an image quality corresponding to a normal state even in a state where the detector is out of order.

  The term "processor" used in the above description means, for example, a circuit such as a CPU, GPU, ASIC, programmable logic device (PLD), or the like. PLD includes SPLD, CPLD, and FPGA. The processor realizes the function by reading and executing the program stored in the memory circuit. The storage circuit in which the program is stored is a computer-readable non-transitory recording medium. Instead of storing the program in the memory circuit, the program may be directly incorporated in the circuit of the processor. In this case, the processor realizes the function by reading and executing the program incorporated in the circuit. Further, instead of executing the program, a function corresponding to the program may be realized by a combination of logic circuits. It should be noted that each processor of the present embodiment is not limited to the case where each processor is configured as a single circuit, and a plurality of independent circuits may be combined and configured as one processor to realize its function. Good. Furthermore, a plurality of constituent elements in FIG. 1 may be integrated into one processor to realize its function.

  Note that the processing circuit 71 may have a circuit configuration that realizes the same function as the machine learning model according to each embodiment in which the parameters are learned so that the failure image 801 is input and the corrected image 805 is output. Good. The circuit configuration is realized by, for example, an integrated circuit such as an ASIC or PLD.

  Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the invention described in the claims and the equivalents thereof as well as included in the scope and the gist of the invention.

1 ... PET device (medical image processing device),
10 ... Gantry,
11 ... Detector ring,
13 ... Signal processing circuit,
15 ... Simultaneous counting circuit,
50 ... sleeper,
51 ... base,
52 ... Support frame,
53 ... Top plate,
54 ... Bed driving device,
70 ... Console (medical image processing apparatus),
71 ... Processing circuit (medical image processing apparatus),
72 ... PET data memory,
73 ... Input circuit,
74 ... Display circuit,
75 ... memory,
110 ... Detector (PET detector, first PET detector, second PET detector),
113 ... Scintillator,
115 ... Photomultiplier tube,
711 ... Reconstruction function (acquisition unit),
712 ... Image processing function,
713 ... Imaging control function (specification unit),
714 ... Display control function,
715 ... Model generation function,
716 ... Correction processing function (processing unit, specifying unit, correction unit),
801, 802 ... Failure image (first medical data),
803 ... Failure detector number,
804, 805 ... Corrected image (second medical data),
810, 830, 850 ... Learning data set,
811, 831, 851 ... Learning data,
813, 833, 853 ... Normal image (second medical data for learning),
815, 835, 855 ... Detector number,
817, 837, 857 ... Simulated failure image (first medical data for learning),
820, 821, 823, 840, 860 ... Machine learning model.

Claims (15)

Of a plurality of PET detectors arranged in a ring shape, an acquisition unit that acquires first medical data that does not include data relating to the first PET detector,
A medical image processing apparatus comprising: a processing unit that generates second medical data including data related to the first PET detector from the acquired first medical data according to a machine learning model.   The medical image processing apparatus according to claim 1, wherein the machine learning model is learned for each of a plurality of PET detectors arranged in a ring shape. Further comprising a specifying unit for specifying a defective PET detector among the plurality of PET detectors,
The processor selects a machine learning model for the failed PET detector,
The medical image processing apparatus according to claim 2.   The processing unit, when a plurality of PET detectors are specified as the failed PET detectors, connects a plurality of machine learning models related to the specified plurality of PET detectors in series. Medical image processing device.   The medical image processing apparatus according to claim 1, wherein the acquired first medical data does not further include data based on a second PET detector different from the first PET detector.   6. The machine learning model is learned to output the second medical data based on the detector numbers of the first and second PET detectors together with the first medical data. The medical image processing apparatus according to item 1. Further comprising a specifying unit for specifying a defective PET detector among the plurality of PET detectors,
The processing unit generates the second medical data from the detector number of the identified PET detector together with the first medical data according to the machine learning model.
The medical image processing apparatus according to claim 5. A specifying unit for specifying a defective PET detector among the plurality of PET detectors;
The correction unit that corrects the second medical data output by the machine learning model based on position information of the first PET detector and the defective PET detector. Medical image processing device. The correction unit corrects the first medical data based on the position information of the first PET detector and the defective PET detector,
The processing unit generates the second medical data from the corrected first medical data according to the machine learning model,
The medical image processing apparatus according to claim 8. When a plurality of PET detectors are specified as the defective PET detectors, the correction unit is configured to detect the second PET detectors based on position information of the first PET detectors and the plurality of defective PET detectors. Correcting the medical data or the corrected second medical data,
The processing unit generates second medical data from the corrected second medical data according to the machine learning model,
The medical image processing apparatus according to claim 8.   The medical image processing apparatus according to claim 1, wherein the first medical data for learning is generated by excluding the data acquired by the first PET detector from the second medical data for learning. .   The medical image processing apparatus according to claim 1, wherein the first medical data for learning is acquired in a state in which data collection regarding the first PET detector is stopped.   The medical image processing apparatus according to claim 1, wherein the first and / or second medical data is list mode data, histogram data, or a reconstructed image.   The first medical data does not include data regarding the first PET detector, and includes data regarding PET detectors other than the first PET detector among the plurality of PET detectors, The second medical data includes data regarding the first PET detector, and includes data regarding PET detectors other than the first PET detector among the plurality of PET detectors. The medical image processing apparatus according to item 1.   Of the plurality of PET detectors arranged in a ring shape, first medical data that does not include data relating to the first PET detector is acquired, and the first PET is acquired from the acquired first medical data. A medical image processing method including generating second medical data including data on a detector according to a machine learning model.