JP2020076584A - Medical image processing device - Google Patents

Abstract

To accurately perform preprocessing of dynamic analysis so as to increase accuracy of dynamic analysis of a tracer given into a subject.SOLUTION: A medical image processing device according to an embodiment comprises a first acquisition unit and a first processing unit. The first acquisition unit acquires dynamic data (nuclear medicine data of a plurality of time phases). The first processing unit generates a time activity curve on the basis of the nuclear medicine data of the plurality of time phases by inputting the dynamic data (the nuclear medicine data of the plurality of time phases) to a learned model which generates the time activity curve indicating time variation in a radiation dose.SELECTED DRAWING: Figure 5

Description

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

  In medical image diagnosis of a subject, a dynamic analysis that analyzes the temporal movement of a tracer administered in the subject may be used based on a plurality of medical images that are continuous in time series.

  For example, when a contrast agent is used as a tracer, the contrast agent is based on a plurality of temporally consecutive medical images acquired by an X-ray CT (Computed Tomography) or a magnetic resonance imaging (MRI) apparatus. Dynamic analysis is performed. Further, in the nuclear medicine image diagnosis, based on a plurality of temporally consecutive nuclear medicine images acquired by a nuclear medicine diagnosis apparatus such as a SPECT (Single Photon Emission Computed Tomography) apparatus or a PET (Positron Emission Tomography) apparatus, the nuclear medicine image is detected. Kinetic analysis of the radiopharmaceutical administered in the sample is performed.

  In the dynamic analysis, a region of interest is set in a target organ or tumor, a time-density curve of pixel values in the region of interest is generated, and the time-density curve can be analyzed to obtain various kinds of information. However, the setting items in the preprocessing of the dynamic analysis include items that are greatly affected by the skill of the user, such as setting the region of interest for obtaining the time-density curve and setting the analysis processing range for the time-density curve. If these setting items vary, the results of the dynamic analysis also vary. Therefore, the result of the dynamic analysis may vary depending on the skill of the user.

JP, 2014-048158, A

  The problem to be solved by the present invention is to accurately perform pretreatment of dynamic analysis so as to improve the accuracy of dynamic analysis of a tracer administered in a subject.

  The medical image processing apparatus according to the embodiment has a first acquisition unit and a first processing unit. The 1st acquisition part acquires dynamic data (nuclear medicine data of a plurality of time phases). The first processing unit inputs the nuclear medicine data of multiple time phases to the learned model that generates the Time Activity Curve showing the time change of the radiation dose based on the nuclear medicine data of multiple time phases. Generate Time Activity Curve.

(A) is a block diagram showing a configuration example of a nuclear medicine diagnostic apparatus including a console as a medical image processing apparatus according to the first embodiment, and (b) is a sectional view taken along line X-X 'of (a). The block diagram which shows roughly the internal structural example of a nuclear medicine diagnostic apparatus. 3 is a schematic block diagram showing an example of functions implemented by a processor of the processing circuit according to the first embodiment. FIG. Explanatory drawing which shows an example of the data flow at the time of learning of the 1st learning method of a TAC generation function. Explanatory drawing which shows an example of the data flow at the time of operation of the 1st learning method of a TAC production | generation function. (A) is an explanatory view showing an example of a data flow at the time of learning of the 2nd learning method of a TAC generation function, (b) is an explanatory view showing an example of a data flow at the time of operation of the 2nd learning method. (A) is an explanatory view showing an example of a data flow at the time of learning of an addition image generation function, and (b) is an explanatory view showing an example of a data flow at the time of operation. (A) is an explanatory view showing an example of a data flow at the time of learning a ROI setting function, and (b) is an explanatory view showing an example of a data flow at the time of operation. Explanatory drawing which shows an example of the data flow at the time of the operation of a group of learned models. (A) is an explanatory view showing an example of a data flow at the time of learning of a processing range generation function, (b) is an explanatory view showing an example of a data flow at the time of operation. The block diagram which shows one structural example of the medical image processing apparatus which concerns on 2nd Embodiment.

  Hereinafter, an embodiment of a medical image processing apparatus will be described in detail with reference to the drawings.

  The medical image processing apparatus according to the present embodiment performs pre-processing for the dynamic analysis of the tracer administered in the subject, and has the time change of the tracer based on a plurality of medical images that are continuous in time series. It can generate data. As the medical image according to the present embodiment, a nuclear medicine image based on nuclear medicine data collected by a nuclear medicine diagnostic device such as a SPECT device or a PET device, an X-ray CT image acquired by an X-ray CT device, and an MRI device. It is possible to use an MR image or the like acquired in.

  In the following example, a case where the medical image processing apparatus according to the present invention is included as a part of a nuclear medicine diagnosis apparatus will be described. Further, in the following example, an example of using a three-detector gamma ray detector rotating type SPECT device as a nuclear medicine diagnostic device will be described. The gamma ray detector rotating type SPECT apparatus may have one, two or four or more gamma ray detectors.

(First embodiment)
FIG. 1A is a block diagram showing a configuration example of a nuclear medicine diagnosis apparatus 1 including a console 30 as a medical image processing apparatus according to the first embodiment, and FIG. 1B is a XX line in FIG. It is sectional drawing which follows the line. Further, FIG. 2 is a block diagram schematically showing an internal configuration example of the nuclear medicine diagnosis apparatus 1.

  As shown in FIG. 1A, the nuclear medicine diagnostic device 1 has a gantry device 10, a bed device 20, and a console 30 as a medical image processing device.

  The gantry device 10 has a gantry body 11 that forms a cylindrical imaging space therein, and three gamma ray detectors 12a, 12b, and 12c. Further, the gantry device 10 further includes three collimators 13a, 13b and 13c detachably provided on the three gamma ray detectors 12a, 12b and 12c, and a rotary drive device 14, respectively.

  The gantry body 11 includes a pedestal, a fixed gantry configured by a housing fixed to the pedestal, and a rotary gantry including a rotating plate rotatably supported with respect to the fixed gantry. The fixed base and the rotary base are covered with a base cover, for example. An opening including an imaging region is provided in the central portion of the fixed mount, the rotary mount, and the mount cover that covers these.

  As shown in FIG. 1B, the nuclear medicine diagnostic apparatus 1 according to the present embodiment is a three-detector type gamma ray detector including three gamma ray detectors 12a, 12b and 12c arranged in a triangular shape. Type SPECT device.

  The three gamma ray detectors 12a, 12b and 12c are held on a rotary mount. The three γ-ray detectors 12a, 12b, and 12c rotate as a unit around the rotation axis by rotating the rotation base around the rotation axis (around the z axis) via the rotation drive device 14.

  The gamma ray detector 12a detects gamma rays emitted from RI (radioactive isotope) such as technesium administered to a subject (eg, patient). Since the gamma ray detectors 12b and 12c have the same configuration and operation as the gamma ray detector 12a, the description thereof will be omitted.

  The gamma ray detector 12a detects gamma rays emitted from RI (radioisotope) such as Tl-201 or Tc-99m that is administered to the subject (for example, patient) P. Since the gamma ray detectors 12b and 12c have the same configuration and operation as the gamma ray detector 12a, the description thereof will be omitted. The gamma ray detector 12a may be a scintillator type detector or a semiconductor type detector.

  When the gamma ray detector 12a is a scintillator type detector, the gamma ray detector 12a includes a collimator 13a for defining the incident angle of the gamma ray and a scintillator that emits a momentary flash when the gamma ray collimated by the collimator 13a is incident. And a light guide, a plurality of two-dimensionally arranged photomultiplier tubes for detecting the light emitted from the scintillator, and an electronic circuit for the scintillator. The scintillator is composed of, for example, thallium-activated sodium iodide NaI (Tl).

  The electronic circuit for the scintillator uses the output of multiple photomultiplier tubes each time the event of gamma ray incidence occurs, and the incident position information of the gamma rays within the detection plane composed of multiple photomultiplier tubes. (Position information), incident intensity information, and incident time information are generated and output to the processing circuit 35 of the console 30. This position information may be information of two-dimensional coordinates within the detection surface, or the detection surface may be virtually divided into a plurality of divided areas (primary cells) in advance (for example, divided into 128 × 128 pieces). Alternatively, the information may be information indicating which primary cell has been incident.

  On the other hand, when the gamma ray detector 12a is a semiconductor type detector, the gamma ray detector 12a includes a collimator 13a and a plurality of semiconductors for two-dimensionally arranged gamma ray detection for detecting the gamma rays collimated by the collimator 13a. It has an element (hereinafter referred to as a semiconductor element), an electronic circuit for a semiconductor, and the like. The semiconductor element is made of, for example, CdTe or CdZnTe (CZT).

  The electronic circuit for semiconductors generates incident position information, incident intensity information, and incident time information based on the output of the semiconductor element and outputs the information to the processing circuit 35 each time an event that a gamma ray is incident occurs. This position information is information indicating which semiconductor element of a plurality of semiconductor elements (for example, 128 × 128) has entered.

  That is, the gamma ray detector 12a outputs incident position information, incident intensity information, and incident time information for each event. Further, the position information is at least one of information indicating at which position of the primary cell the gamma ray is incident and information about two-dimensional coordinates in the detection plane.

  The imaging timing of the gamma ray detectors 12a, 12b and 12c is controlled by the processing circuit 35.

  Each of the collimators 13a, 13b, and 13c is made of a substance such as lead or tungsten that is difficult to transmit radiation, and is provided with a plurality of holes for controlling the direction in which photons fly. This hole has, for example, a polygonal shape such as a hexagon.

  The rotation drive device 14 transmits rotation means such as a motor for rotating the rotary mount at high speed around a predetermined rotation axis r, electronic components for controlling rotation of the rotary means, and rotation of the rotary means to the rotary mount. It has a transmission means such as a roller for rotating. The rotation driving device 14 is controlled by the processing circuit 35 via the data collection circuit 15 to rotate the rotary mount about a predetermined rotation axis r. For example, the processing circuit 35 rotates the gamma ray detectors 12a, 12b, and 12c around the subject P continuously or stepwise through the rotary mount, so that the SPECT projection data of the subject from a plurality of directions (hereinafter referred to as “SPECT projection data”). , Projection data) can be collected.

  The data acquisition circuit 15 is composed of, for example, a printed circuit board, and is controlled by the processing circuit 35 to control the gamma ray detectors 12a, 12b and 12c, the rotation driving device 14, and the top driving device 23, so that the subject P can be detected. The image capturing is executed.

  The data collection circuit 15 collects the outputs of the gamma ray detectors 12a, 12b, and 12c, for example, in the list mode, and outputs the collected projection data to the console 30. In the list mode, gamma ray detection position information, intensity information, information indicating the relative position between the gamma ray detectors 12a, 12b and 12c and the subject P (positions and angles of the gamma ray detectors 12a, 12b and 12c, etc.), and gamma rays. The detection time of is collected for each incident event of gamma rays.

  The couch device 20 includes a couchtop 21, a couch body 22 that moves the couchtop 21 in vertical and horizontal directions, and a couchtop drive device 23 that moves the couchtop 21.

  The top plate 21 is configured by a plate-shaped member having a longitudinal direction in the z-axis direction and a lateral direction in the x-axis direction. The subject is placed on the top plate 21. The subject P is placed on the top plate 21.

  The bed main body 22 is installed on the floor and supports the top plate 21 so as to be able to move up and down. For example, the bed main body 22 has two supporting members having a long shape and arranged in an X shape, and a connecting pin that pivotally supports the central portion of the two supporting members, and the upper end portions of the two supporting members. The top plate 21 is supported by the. In this case, the top plate 21 rises, for example, by narrowing the gap between the lower ends of the two support members, and descends by widening the gap between the lower ends.

  The top driving device 23 moves the top 21 under the control of the processing circuit 35 via the data collecting circuit 15. Specifically, the top driving device 23 has a motor as a drive source for moving the top 21 along the z-axis direction and the x-axis direction, an electronic component for controlling the motor, and the like. The top driving device 23 has a motor as a drive source for moving the top 21 up and down, an electronic component for controlling the motor, and the like.

  On the other hand, the console 30 of the nuclear medicine diagnostic apparatus 1 is composed of, for example, a general personal computer or workstation, and has an input interface 31, a display 32, a storage circuit 33, a network connection circuit 34, and a processing circuit 35. It should be noted that the console 30 may not be provided independently, and for example, a part of the components 31-35 of the console 30 may be provided dispersedly on the fixed base. The console 30 is an example of a medical image processing apparatus.

  The input interface 31 is composed of a general input device such as a trackball, a switch button, a mouse, a keyboard, and a numeric keypad, and outputs an operation input signal corresponding to a user's operation to the processing circuit 35. For example, the user can specify the imaging target site and the RI used in the examination via the input interface 31.

  The display 32 is configured by a general display output device such as a liquid crystal display or an OLED (Organic Light Emitting Diode) display.

  The memory circuit 33 has a configuration including a processor-readable recording medium such as a magnetic or optical recording medium or a semiconductor memory. The storage circuit 33 is controlled by the processing circuit 35 and stores the nuclear medicine data (nuclear medicine images, nuclear medicine image data, or raw data) of the subject P at multiple time phases. Part or all of the programs and data in these recording media may be configured to be downloaded by communication via an electronic network.

  The network connection circuit 34 is composed of, for example, a network card having a predetermined printed circuit board, and implements various information communication protocols according to the form of the network. The network connection circuit 34 connects the nuclear medicine diagnostic apparatus 1 and other devices according to these various protocols. For this connection, electrical connection or the like via an electronic network can be applied. Here, the electronic network refers to all information communication networks that use telecommunications technology, such as wireless / wired hospital backbone LAN (Local Area Network) and the Internet network, as well as telephone communication network, optical fiber communication network, and cable. It includes communication networks and satellite communication networks.

  The processing circuit 35 reads the program stored in the storage circuit 33 and executes the program to accurately perform the preprocessing of the dynamic analysis so as to improve the accuracy of the dynamic analysis of the tracer administered to the subject P. A processor that executes a process.

  In the pre-processing of the dynamic analysis, a time activity curve (hereinafter, referred to as TAC) showing a time change of the radiation dose of the region of interest is generated from nuclear medicine data (hereinafter, appropriately referred to as dynamic data) of a plurality of time phases of the subject P. At least include processing. In the following description, in addition to the TAC generation process, a region of interest setting process used for TAC generation, a generation process of an addition image used for setting the region of interest, and a dynamic process for TAC are performed as pre-processing for dynamic analysis. An example in the case of including a process of generating at least one of an analysis processing start point of analysis, an analysis processing range, and a background calculation processing range will be shown. After the execution of these pre-processing, the analysis processing is executed. The analysis process may be executed after the preprocess in the console 30 as an example of the medical image processing apparatus that executes the preprocess, or may be executed in another device.

  FIG. 3 is a schematic block diagram showing an example of functions implemented by the processor of the processing circuit 35 according to the first embodiment.

  As shown in FIG. 3, the processor of the processing circuit 35 realizes a dynamic data acquisition function 41, a TAC generation function 42, an addition image generation function 43, a ROI setting function 44, a TAC acquisition function 45, and a processing range generation function 46. .. Further, each of these functions is stored in the storage circuit 33 in the form of a program.

  The dynamic data acquisition function 41 acquires dynamic data (nuclear medicine data of a plurality of time phases of the subject P). During operation of the learned model, dynamic data (nuclear medicine data of multiple time phases) is acquired via the data acquisition circuit 15. The dynamic data (nuclear medicine data of multiple time phases) may include dynamic data (nuclear medicine data of multiple time phases) created by rearranging dynamic tomographic images in slice units. The dynamic data acquisition function 41 is an example of a first acquisition unit.

  The TAC generation function 42 inputs the nuclear medicine data of a plurality of time phases to a learned model that generates a TAC (Time Activity Curve showing the time change of the radiation dose) based on the nuclear medicine data of a plurality of time phases. As a result, TAC is generated. The TAC generation function 42 is an example of a first processing unit.

  A plurality of trained models are prepared according to the collection conditions of nuclear medicine data of a plurality of time phases. The collection condition includes a collection site, a collection method, a radiopharmaceutical, and a physique of the subject. Further, the collection method includes a collection time and a collection expansion rate.

  Further, the TAC generation function 42 may generate the TAC further based on the information on the region of interest set in the nuclear medicine data. In addition, the TAC generation function 42 may generate a TAC corresponding to each of the plurality of regions of interest when the information of the plurality of regions of interest set in the nuclear medicine data is available.

  The addition image generation function 43 is based on the nuclear medicine data of the plurality of time phases, and generates the addition image in which the nuclear medicine data of at least a part of the plurality of time phases is added to the first learned model. , Addition images are generated by inputting nuclear medicine data of multiple time phases. The console 30 as the medical image processing apparatus may not include the addition image generation function 43.

  The ROI setting function 44 is set in the nuclear medicine data by inputting the addition image to the second learned model that generates information about the region of interest set in the nuclear medicine data based on the added image. Generate information about the region of interest to be processed. The console 30 as the medical image processing apparatus may not include the ROI setting function 44.

  The TAC acquisition function 45 acquires the TAC generated by the TAC generation function 42. The TAC acquisition function 45 is an example of a second acquisition unit.

  The processing range generation function 46 inputs the TAC to the third learned model that generates at least one of the analysis processing start point, the analysis processing range, and the background calculation processing range in the TAC based on the TAC. Thus, at least one of the analysis processing start point, the analysis processing range, and the background calculation processing range is generated. The processing range generation function 46 is an example of a second processing unit.

  FIG. 4 is an explanatory diagram showing an example of a data flow during learning by the first learning method of the TAC generation function 42.

  The TAC generation function 42 receives a large number of training data and performs learning to sequentially update the parameter data 422.

  In the first learning method, the training data are dynamic data (nuclear medicine data of a plurality of time phases) 511, 512, 513, ... Constituting the dynamic data group 51 as training input data, and each dynamic data ( , Corresponding to multiple temporal phases of nuclear medicine data) and a teacher data group 61 composed of ideal TACs 611, 612, 613 ,. The ideal TACs 611, 612, 613, ... Use those actually generated from the corresponding dynamic data 511, 512, 513 ,.

  The dynamic data group 51 includes thinning out the first and last predetermined time phases of the dynamic data (nuclear medicine data of multiple time phases) 511, 512, 513, ..., or adding dummy data, or each data The training data may be increased by increasing the number of dynamic data groups 51 by, for example, smoothing.

  The TAC generation function 42 has ideal TACs 611 and 612 for the result of processing the dynamic data 511, 512, 513, ... , 613, ..., So-called learning is performed to update the parameter data 422. Generally, when the rate of change of the parameter data 422 converges within the threshold value, the learning is determined to be completed. Hereinafter, the parameter data 422 after learning is particularly referred to as learned parameter data 422t.

  FIG. 5 is an explanatory diagram showing an example of a data flow during operation of the first learning method of the TAC generation function 42. During operation of the first learning method, the TAC generation function 42 receives the dynamic data (nuclear medicine data of multiple time phases) 71 of the subject P, which is the target of TAC generation, and uses the learned model 420 to acquire the TAC 81. To generate.

  The neural network 421 and the learned parameter data 422t form the learned model 420. The neural network 421 is stored in the storage circuit 33 in the form of a program. The learned parameter data 422t may be stored in the storage circuit 33, or may be stored in a storage medium connected to the console 30 via the network. When the learned model 420 (the neural network 421 and the learned parameter data 422t) is stored in the storage circuit 33, the TAC generation function 42 as an example of the first processing unit realized by the processor of the processing circuit 35 is The learned model 420 is read out from the memory circuit 33 and executed to generate the TAC 81 based on the dynamic data (nuclear medicine data of multiple time phases) 71.

  Each of the learned models according to the present embodiment including the learned model 420 may be constructed by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) and an FPGA (Field Programmable Gate Array).

  In addition, it is preferable that the teacher data used for each learning of the learned model according to the present embodiment, including the learned model 420, be created by a plurality of users. This is because the teacher data may reflect the habit of the creator of the teacher data, and the data may be biased.

  In addition, a model that has been trained according to the collection conditions (collection site, collection method (collection time, collection expansion rate, etc.), radiopharmaceutical, and physique of the subject) so as to improve the accuracy of TAC 81 generation by the TAC generation function 42. 420 is prepared. In this case, the dynamic data (nuclear medicine data of multiple time phases) forming the dynamic data group 51 is classified according to the collection condition, and the learned model 420 is constructed for each classification. Note that in this case, the collection condition of the dynamic data group 51 and the collection condition of the input data 71 during operation of the learned model 420 constructed based on the collection condition should match.

  To obtain the TAC 81 more accurately by preparing a plurality of trained models 420 according to the collection condition and using the trained model 420 according to the collection condition of the dynamic data (multi-phase nuclear medicine data) 71. You can

  FIG. 6A is an explanatory diagram showing an example of a data flow during learning of the second learning method of the TAC generation function 42A, and FIG. 6B shows an example of a data flow during operation of the second learning method. FIG.

  The second learning method is different from the first learning method in that the region of interest information group 52 and the added image group 53 are used as the training input data in addition to the dynamic data group 51. The ROI information group 52 includes information on ROIs (hereinafter referred to as ROI information) 521, 522, 523 set in each of the dynamic data (nuclear medicine data of multiple time phases) 511, 512, 513, .... ,. The ROI information includes information on the position of the region of interest set in the dynamic data (nuclear medicine data of multiple time phases). Further, the addition image group 53 is configured by addition images 531, 532, 533, ... Used for setting the ROI included in the ROI information 521, 522, 523 ,.

  The TAC generation function 42A obtains learned parameter data 422At by learning using the neural network 421A. The neural network 421A and the learned parameter data 422At form a learned model 420A.

  During operation of the second learning method, the TAC generation function 42A inputs the dynamic data (nuclear medicine data of multiple time phases) 71 of the subject P, which is the target of TAC generation, ROI information 72, and the added image 73. Then, the TAC 81 is generated using the learned model 420A.

  Further, the TAC generation function 42 may use the third learning method that does not use the added image group 53 as the training input data. When the second learning method or the third learning method is in operation, the ROI information 72 is used to generate the TAC 81 with higher accuracy than the case where the learned model 420 generated by the first learning method is used. can do.

  Further, in the second learning method and the third learning method, a plurality of ROIs may be set for each of the ROI information 521, 522, 523, ... As the training input data. In this case, during operation, the ROI information 72 includes information on a plurality of ROIs set for one dynamic data (nuclear medicine data of a plurality of time phases) 71, and the TAC generation function 42 (42A) respectively TAC corresponding to the ROI of In this case, for example, the user may select, via the input interface 31, one TAC that is considered optimal for the dynamic data (multi-phase nuclear medicine data) 71 from the plurality of generated TACs.

  FIG. 7A is an explanatory diagram showing an example of a data flow during learning of the addition image generation function 43, and FIG. 7B is an explanatory diagram showing an example of a data flow during operation.

  The addition image generation function 43 receives a large number of training data and performs learning to sequentially update the parameter data 432. The training data are dynamic data (nuclear medicine data of multiple time phases) 511, 512, 513, ..., Which form a dynamic data group 51 as training input data, and each dynamic data (nuclear medicine data of multiple time phases). ), The teacher data group 53 composed of ideal added images 531, 532, 533 ,. The teacher data group 53 may include setting information of an addition range for generating an addition image. The addition range is defined by the time phase section of the dynamic data group 51, for example. The ideal added images 531, 532, 533, ... Use those actually generated from the corresponding dynamic data (nuclear medicine data of multiple time phases) 511, 512, 513 ,. ..

  The addition image generation function 43 is an addition image in which ideal results are obtained by processing dynamic data (nuclear medicine data of multiple time phases) 511, 512, 513, ... With the neural network 431 every time training data is given. The parameter data 432 is updated so as to approach 531, 532, 533, .... The parameter data 432 after learning is particularly referred to as learned parameter data 432t. The neural network 431 and the learned parameter data 432t form a first learned model 430.

  At the time of operation, the addition image generation function 43 receives the dynamic data (nuclear medicine data of multiple time phases) 71 as the generation target of the addition image with respect to the first learned model 430 read from the storage circuit 33, The addition image 73 is generated using the first learned model 430.

  FIG. 8A is an explanatory diagram showing an example of a data flow during learning of the ROI setting function 44, and FIG. 8B is an explanatory diagram showing an example of a data flow during operation.

  The ROI setting function 44 receives a large amount of training data and performs learning to sequentially update the parameter data 442. The training data includes added images 531, 532, 533, ... Which constitute the added image group 53 as training input data, and ideal ROI information 521, 522, 523, ... Corresponding to each added image. And a teacher data group 52 constituted by. As the ideal ROI information 521, 522, 523, ..., those actually set for the corresponding addition images 531, 532, 533 ,. It is more accurate to set the ROI for the added image than for the dynamic data (nuclear medicine data of multiple time phases) directly.

  The ROI setting function 44 processes the added images 531, 532, 533, ... With the neural network 441 every time the training data is given, and the ideal result is ROI information 521, 522, 523 ,. The parameter data 442 is updated so as to approach. The parameter data 442 after learning is particularly referred to as learned parameter data 442t. The neural network 441 and the learned parameter data 442t form the second learned model 440.

  At the time of operation, the ROI setting function 44 inputs the added image 73, which is the target for generating ROI information, to the second learned model 440 read from the storage circuit 33, and uses the second learned model 440. The ROI information 72 is generated.

  Subsequently, the addition image generation process by the first learned model 430 (see FIG. 7B), the ROI setting process by the second learned model 440 (see FIG. 8B), and the learned model 420A are performed. A case where the TAC generation process (see FIG. 6B) is used as a series of processes will be described.

  FIG. 9 is an explanatory diagram showing an example of a data flow during operation of a group of learned models 420B.

  As shown in FIG. 9, even if the addition image generation process by the first learned model 430, the ROI setting process by the second learned model 440, and the TAC generation process by the learned model 420A are operated as a series of processes. Good. Also in this case, in appearance, similar to the learned model 420, the dynamic data (nuclear medicine data of multiple time phases) 71 is input to the lump of the learned model 420B and the TAC 81 is output.

  Further, while operating the addition image generation processing by the first learned model 430 and the ROI setting processing by the second learned model 440 as a series of processing, the learned model 420 or the learned model 420A, or the third learning The learned model 420 generated by the method may be operated as another process.

  FIG. 10A is an explanatory diagram showing an example of a data flow during learning of the processing range generation function 46, and FIG. 10B is an explanatory diagram showing an example of a data flow during operation.

  The processing range generation function 46 receives a large number of training data and performs learning to sequentially update the parameter data 462. The training data is an ideal process in which TACs 541, 542, 543, ... Which constitute the TAC group 54 as training input data, and ideal cursor positions C1, C2, C3 corresponding to each TAC are set. , And the teacher data group 62 composed of the range setting information 621, 622, 623 ,. As the ideal processing range setting information 621, 622, 623, ..., What is actually set for each of the TACs 541, 542, 543 ,.

  The cursor position C1 indicates the analysis processing start point of the dynamic analysis executed after the preprocessing according to the present embodiment. Further, the range between the cursor position C2 and the cursor position C3 defines a range used for background calculation (background calculation processing range). Further, the range between the cursor positions C1 and C3 defines the analysis processing range of the dynamic analysis.

  The number of TAC groups 54 is determined by thinning out the first and last predetermined time phases of TACs 541, 542, 543, ..., Adding dummy data, or smoothing a curve. You may increase training data by increasing.

  The processing range generation function 46 processes the TACs 541, 542, 543, ... With the neural network 461 every time the training data is provided, and the ideal processing range setting information 621, 622, 623 ,. , The parameter data 462 is updated so as to approach. The parameter data 462 after learning is particularly referred to as learned parameter data 462t. The neural network 461 and the learned parameter data 462t form a third learned model 460.

  In operation, the processing range generation function 46 inputs the TAC 81, which is the generation target of the processing range setting information, to the third learned model 460 read from the storage circuit 33, and uses the third learned model 460. To generate the processing range setting information 91. At this time, the outputs of the learned models 420, 420A, and 420B may be used as the TAC 81, and the TAC generation process and the process range setting process may be operated as a series of processes.

  According to the processing circuit 35 of the nuclear medicine diagnosis apparatus 1 according to the present embodiment, by using the learned model 420, the TAC generation processing as the preprocessing of the dynamic analysis can be performed accurately without depending on the skill of the user. it can. Further, by using the first, second, and third learned models 430, 440, and 460, the addition image generation process and the ROI setting process as the preprocessing of the dynamic analysis can be performed accurately without depending on the skill of the user. , And processing range setting processing can be performed respectively. Further, by using the TAC 81 and the processing range setting information 91 obtained by this pre-processing, the accuracy of the dynamic analysis processing as the post-processing can be greatly improved.

  Note that each of the learned models according to the present embodiment including the learned model 420 may feed back a set of input data and output data at the time of operation to the learning as training data.

  Further, the TAC 81, which is the product of the learned models 420 (420A, 420B), 430, 440, and 460, the added image 73, the ROI information 72, and the processing range setting information 91, respectively, are for receiving correction by the user. It may be displayed on the display 32 together with the dialogue window to accept the correction by the user via the input interface 31. Further, when the correction is accepted, the data after the correction may be fed back to the learning as teacher data.

(Second embodiment)
FIG. 11 is a block diagram showing a configuration example of the medical image processing apparatus 30N according to the second embodiment. The medical image processing apparatus 30N shown in the second embodiment is provided independently of modalities such as a nuclear medicine diagnosis apparatus 101, an image server 102, an X-ray CT apparatus 103, and an MRI apparatus 104 connected via a network 100. The difference is the console 30 shown in the first embodiment. Since other configurations and operations are substantially the same as those of the console 30 shown in FIG. 1, the same configurations are denoted by the same reference numerals and description thereof will be omitted.

  The medical image processing apparatus 30N is independent of modalities such as the nuclear medicine diagnostic apparatus 101, the image server 102, the X-ray CT apparatus 103, and the MRI apparatus 104, and can transmit and receive data to and from each of these modalities via the network 100. Connected.

  The network connection circuit 34N connects the medical image processing apparatus 30N, the nuclear medicine diagnosis apparatus 101, the image server 102, the X-ray CT apparatus 103, and the MRI apparatus 104. For this connection, electrical connection or the like via an electronic network can be applied. Here, the electronic network refers to all information communication networks that use telecommunications technology, such as wireless / wired hospital backbone LAN (Local Area Network) and the Internet network, as well as telephone communication network, optical fiber communication network, and cable. It includes communication networks and satellite communication networks.

  The dynamic data acquisition function 41N is used when operating the learned model through the network 100, if necessary, from the nuclear medicine diagnosis apparatus 101, the image server 102, the X-ray CT apparatus 103, or the MRI apparatus 104 at a plurality of times. Obtain medical data (dynamic data) of the phase.

  For example, when acquiring dynamic data from the nuclear medicine diagnostic apparatus 101, the dynamic data is nuclear medicine data of a plurality of phases, and the trained models 420, 430, 440, and 460 are constructed in the same manner as in the first embodiment. To be done. Further, for example, when acquiring from the X-ray CT apparatus 103, the dynamic data is a plurality of X-ray CT images or X-ray CT image data that are continuous in time series, and the learned model 420 is the time concentration curve of the contrast agent. The TDC is generated, and the learned model 460 generates the processing range setting information based on the TDC. Other operations of the processing circuit 35N are the same as those of the processing circuit 35 according to the first embodiment.

  The processing circuit 35N of the medical image processing apparatus 30N according to the second embodiment also has the same effect as the processing circuit 35 of the console 30 according to the first embodiment. Further, according to the medical image processing apparatus 30N according to the second embodiment, in addition to nuclear medicine data, medical data including X-ray CT data and MR data can be used, and time can be calculated based on medical data of multiple time phases. Data with changes (TAC, TDC) can be generated.

  Further, the medical image processing apparatus 30N according to the present embodiment is for generating data having a change with time of the tracer. For example, if the tracer is a radiopharmaceutical and the medical data is nuclear medicine data, the dynamic data of the heart ( It is possible to generate TAC based on multiple phases of nuclear medicine data). Using this TAC, the dynamic analysis for estimating the cerebral blood flow is performed after the preprocessing executed by the medical image processing apparatus 30N according to the present embodiment.

  According to at least one embodiment described above, the pretreatment of the dynamic analysis can be appropriately performed so as to improve the accuracy of the dynamic analysis of the tracer administered in the subject.

  In the above embodiment, the word “processor” means, for example, a dedicated or general-purpose CPU (Central Processing Unit), GPU (Graphics Processing Unit), or an application-specific integrated circuit (ASIC), A circuit such as a programmable logic device (for example, Simple Programmable Logic Device (SPLD), Complex Programmable Logic Device (CPLD), and FPGA) is meant. The processor realizes various functions by reading and executing a program stored in a storage medium.

  Further, in the above embodiment, an example in which a single processor of the processing circuit realizes each function has been described, but a processing circuit is configured by combining a plurality of independent processors, and each processor realizes each function. Good. When a plurality of processors are provided, the storage medium for storing the program may be provided individually for each processor, or one storage medium collectively stores the programs corresponding to the functions of all the processors. Good.

  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. The embodiments and their modifications are included in the scope of the invention and the scope thereof, and are included in the invention described in the claims and the scope of equivalents thereof.

1 Nuclear Medicine Diagnostic Device 30 Console 30N Medical Image Processing Device 41, 41N Dynamic Data Acquisition Function 42, 42A TAC Generation Function 43 Addition Image Generation Function 44 ROI Setting Function 45 TAC Acquisition Function 46 Processing Range Generation Function 51 Dynamic Data Group 52 Region of Interest Information group 53 Addition image group 54 TAC group 71 Dynamic data 72 ROI information 73 Addition image 91 Processing range setting information 420, 420A, 420B Learned model 430 First learned model 440 Second learned model 460 Third learning Model

Claims (8)

A first acquisition unit for acquiring nuclear medicine data of a plurality of phases,
Based on the nuclear medicine data of the plurality of time phases, to the learned model that generates a Time Activity Curve showing the time change of the radiation dose, the Time Activity Curve by inputting the nuclear medicine data of the plurality of time phases A first processing unit for generating,
A medical image processing apparatus equipped with. The trained model is
Based on the nuclear medicine data of the plurality of time phases, further comprising a first learned model for generating an addition image in which the nuclear medicine data of at least a part of the plurality of time phases are added,
The medical image processing apparatus according to claim 1. The trained model is
Further comprising a second trained model that generates information about a region of interest set in the nuclear medicine data based on the added image,
The medical image processing apparatus according to claim 2. The trained model is
Further, based on the information of the region of interest set in the nuclear medicine data, to generate the Time Activity Curve,
The medical image processing apparatus according to claim 1. The trained model is
Further based on the information of the plurality of regions of interest set in the nuclear medicine data, to generate the Time Activity Curve corresponding to each of the plurality of regions of interest,
The medical image processing apparatus according to claim 4. The trained model is
A plurality is prepared according to the collection conditions of the nuclear medicine data of the plurality of time phases,
The medical image processing apparatus according to claim 1. A second acquisition unit for acquiring the Time Activity Curve output by the learned model,
Based on the Time Activity Curve, the Time Activity Curve is set to a third learned model that generates at least one of an analysis processing start point, an analysis processing range, and a background calculation processing range in the Time Activity Curve. A second processing unit that generates at least one of the analysis processing start point, the analysis processing range, and the background calculation processing range by inputting;
The medical image processing apparatus according to claim 1, further comprising: A first acquisition unit that acquires medical data of a plurality of phases,
A first model for generating the data having the time change by inputting the medical data of the plurality of time phases to a trained model that generates the data having the time change based on the medical data of the plurality of time phases A processing unit,
A medical image processing apparatus equipped with.

Images