Image processing system, image processing method, control program, recording medium

Abstract

To improve the versatility of the data to be extracted from images. [Solution] The image processing system (100) includes: a probability map generation unit (101) that generates a probability map showing the probability that each pixel in the target image is included in the area of ​​interest corresponding to the target image, using a trained estimation model from a target image including the object to be inspected; a binarization unit (102) that generates a binarized map obtained by binarizing the probability map at a predetermined threshold, in which the pixel values ​​of pixels not included in the area of ​​interest are set to zero; and an integrated image generation unit (103) that generates an integrated target image by integrating the target image and the binarized map by calculating the element product.

Description

[Technical Field]

[0001] The present disclosure relates to an image processing system, an image processing method, and the like. [Background Art]

[0002] A technique is known that extracts only a target object from an image by removing regions that are not the target image. For example, Non-Patent Document 1 discloses a technique for extracting a region corresponding to the brain from an image of the head captured using a magnetic resonance imaging apparatus. [Prior Art Documents] [Non-Patent Documents]

[0003] [Non-Patent Document 1] Stephen M. Smith, Fast Robust Automated Brain Extraction, Human Brain Mapping, 2002 [Summary of the Invention] [Problem to be Solved by the Invention]

[0004] In the technique disclosed in Non-Patent Document 1, after approximating the surface shape of the brain, which is the extraction target, with a polyhedron, the extraction process is performed, thereby removing unnecessary parts such as the skull from the image obtained by capturing the head, and extracting only the brain. In the technique disclosed in Non-Patent Document 1, when an object with a more complex surface shape such as a blood vessel is used as the extraction target, the processing load increases and the extraction process may become difficult.

[0005] The present disclosure provides an image processing system or the like with improved versatility regarding extraction targets from images. [Means for Solving the Problem]

[0006] An image processing system according to one aspect of the present disclosure includes: a probability map generation unit that generates a probability map showing the probability that each pixel in a target image, including an object to be inspected, is included in a region of interest corresponding to the object to be inspected, using a trained estimation model; a binarization unit that generates a binarized map obtained by binarizing the probability map at a predetermined threshold, wherein the pixel values ​​of pixels not included in the region of interest are set to zero; and an integrated image generation unit that generates an integrated target image by integrating the target image and the binarized map by calculating the element product.

[0007] An image processing method according to one aspect of the present disclosure is an image processing method performed by one or more computers, comprising: a probability map generation step in which the computer generates a probability map from a target image including an object to be inspected, using a trained estimation model, indicating the probability that each pixel in the target image is included in a region of interest corresponding to the object to be inspected; a binarization step in which the computer generates a binarized map obtained by binarizing the probability map at a predetermined threshold, wherein the pixel values ​​of pixels not included in the region of interest are set to zero; and an integrated image generation step in which the computer generates an integrated target image by integrating the target image and the binarized map by calculating the element product.

[0008] Each aspect of the image processing system described herein may be implemented by a computer, in which case a control program for the image processing system that enables the computer to implement the image processing system by operating the computer as each part (software element) of the image processing system, and a computer-readable recording medium on which the program is recorded, also fall within the scope of this disclosure. [Effects of the Invention]

[0009] According to one aspect of this disclosure, the versatility of the target to be extracted from the image can be improved. [Brief explanation of the drawing]

[0010] [Figure 1]This block diagram shows an example of a schematic configuration of an image processing system according to Embodiment 1 of this disclosure. [Figure 2] This is a functional block diagram showing an example of the configuration of an image processing system. [Figure 3] This diagram illustrates an example of processing in an image processing system. [Figure 4] This flowchart shows an example of the processing flow performed by the system. [Figure 5] This block diagram shows an example of the configuration of an image processing apparatus according to Embodiment 2 of this disclosure. [Figure 6] This diagram illustrates an example of processing in an image processing system. [Figure 7] This block diagram shows an example of the configuration of an image processing apparatus according to Embodiment 3 of the present disclosure. [Figure 8] This diagram illustrates an example of processing in an image processing system. [Figure 9] This is a block diagram showing an example of the schematic configuration of an image processing system according to Embodiment 4 of this disclosure. [Figure 10] This is a functional block diagram showing an example of the configuration of an image processing system. [Figure 11] This figure shows an example of processing results using the image processing system according to the embodiment of this disclosure. [Modes for carrying out the invention]

[0011] [Embodiment 1] Hereinafter, one embodiment of this disclosure will be described in detail with reference to Figures 1 to 4.

[0012] (Overview of Image Processing System 100) The image processing system 100 is a system that can efficiently remove unnecessary regions from a target image 111, which includes the object to be inspected, using an estimation model 112.

[0013] When the image processing system 100 according to the present disclosure is applied to the medical field, the image processing apparatus 1 may be an apparatus that targets internal tissues in a transmission image of a subject for examination. In this case, the subject may be, for example, a living organism such as a human, and the internal tissue of the subject may be, for example, a tissue such as an organ constituting the living organism. Further, when the image processing system 100 according to the present disclosure is applied to the engineering field, the image processing apparatus 1 may be an apparatus for inspecting an inspection target in a subject based on an image obtained by capturing the non-living subject. In this case, the inspection target may be, for example, an object constituting the inside or the like of the subject, and the image processing system 100 may be, for example, an apparatus for non-destructively inspecting the inspection target in the subject. In the present disclosure, a case where the image processing system 100 is introduced into a medical facility H will be described as an example.

[0014] (Configuration of Image Processing System 100) First, the configuration of the image processing system 100 will be described with reference to FIG. 1. FIG. 1 is a block diagram showing an example of a schematic configuration of the image processing system 100.

[0015] As shown in FIG. 1, the image processing system 100 may include an image management apparatus 2, an image processing apparatus 1, and a display apparatus 3. The image management apparatus 2, the image processing apparatus 1, and the display apparatus 3 may be directly connected, or may be communicatively connected via a local area network (LAN) as shown in FIG. 1. Alternatively, the image management apparatus 2, the image processing apparatus 1, and the display apparatus 3 may be communicatively connected via the Internet. The number of each of the image management apparatus 2, the image processing apparatus 1, and the display apparatus 3 included in the image processing system 100 may be one or plural.

[0016] [Image Management Apparatus 2] The image management apparatus 2 is a computer that stores and manages medical images. The image management apparatus 2 may comprise a control unit and a storage unit. In one example, the control unit may be a CPU (Central Processing Unit). A target image 111 obtained by imaging tissues inside the body of a subject may be stored in the storage unit in association with identification information of the subject, imaging date and time, and identification information indicating the doctor who captured the target image 111. The image management apparatus 2 may read the target image 111 from the storage unit in response to an instruction from the image processing apparatus 1, and output the target image 111 to the image processing apparatus 1. Furthermore, the target integrated image 115 generated by the image processing apparatus 1 may be stored in the storage unit.

[0017] [Image Processing Apparatus 1] The image processing apparatus 1 is a computer used by medical personnel such as doctors. A specific example of the configuration of the image processing apparatus 1 will be described later.

[0018] [Display Apparatus 3] The display apparatus 3 is a display capable of displaying various types of information on a screen. The display apparatus 3 may be a display (display unit) comprised in the image processing apparatus 1. Alternatively, it may be a display (display unit) comprised in the image management apparatus 2.

[0019] (Configuration of Image Processing Apparatus 1) Next, the configuration of the image processing apparatus 1 will be described with reference to FIG. 2. FIG. 2 is a functional block diagram showing an example of the configuration of an image processing system 100. The image processing apparatus 1 comprises a control unit 10 and a storage unit 11. In one example, the control unit 10 may be a CPU (Central Processing Unit). The control unit 10 reads a control program, which is software stored in the storage unit 11, develops the control program in a memory such as a RAM (Random Access Memory), and executes various functions.

[0020] The storage unit 11 may be composed of a non-volatile storage device such as a hard disk and flash memory. The storage unit 11 may store a target image 111, which is an image of tissue inside the subject's body. The storage unit 11 may also store an estimation model 112, which will be described later. The storage unit 11 may also store a probability map 113 generated by the image processing device 1 from the target image 111, a binarization map 114 generated by the image processing device 1, and a target integrated image 115 generated by the image processing device 1 by integrating the target image 111 and the binarization map 114 as described later.

[0021] The control unit 10 includes a probability map generation unit 101, a binarization unit 102, and an integrated image generation unit 103.

[0022] The probability map generation unit 101 generates a probability map 113 from the target image 111, which includes the object to be inspected, using a trained estimation model 112, which shows the probability that each pixel in the target image 111 is included in the area of ​​interest corresponding to the object to be inspected.

[0023] Figure 3 illustrates an example of processing in the image processing system 100. The target image 111 is an image containing the object to be examined. The object to be examined may be, for example, tissue inside the body of the subject. Specific examples of tissue inside the body of the subject include organs such as the bile duct or pancreatic duct. As illustrated in Figure 3, the target image 111 may include a first tissue 1111 which is the object to be examined, as well as a second tissue 1112 which is not the object to be examined. Both the first tissue 1111 and the second tissue 1112 schematically illustrate hypothetical organs. The target image 111 may be a three-dimensional image or a two-dimensional image, as shown in Figure 3.

[0024] The target image 111 may be a medical image obtained by imaging a subject using either a magnetic resonance imaging (MRI) system or an X-ray computed tomography (X-ray) system. Alternatively, the target image 111 may be a three-dimensional medical image obtained by imaging a subject using magnetic resonance imaging (MRI) system, in other words, magnetic resonance cholangiopancreatography (MRCP).

[0025] The estimation model 112 is a machine learning model that estimates a probability map 113, described later, from a target image 111 that includes the object to be inspected, as illustrated in Figure 3. The estimation model 112 may be, for example, a deep learning model. Specific examples of deep learning models used as the estimation model 112 include convolutional neural network (CNN) models such as U-Net, or Transformer models. The configuration for training the estimation model 112 may be, for example, an external configuration to the image processing system 100, or it may be a configuration provided by the image processing system 100a, which is an example of the image processing system 100, as shown in the learning control unit 104 described later in Embodiment 2. Details of the configuration for training the estimation model 112 will be described later, using the learning control unit 104 in Embodiment 2 as an example.

[0026] The region of interest is the region in the target image 111 that corresponds to the object being examined. The region of interest may, for example, be the region corresponding to the first tissue 1111 in the target image 111.

[0027] The probability map 113, as illustrated in Figure 3, is data generated from the target image 111 containing the object to be examined using a trained estimation model 112, and represents the probability that each pixel in the target image 111 is included in the area of ​​interest corresponding to the object to be examined. The probability map 113 may be, for example, continuous data.

[0028] The binarization unit 102 generates a binarized map 114 which is obtained by binarizing the probability map 113 with a predetermined threshold, and where the pixel values ​​of pixels not included in the region of interest are set to zero. The binarized map 114 is data generated from the probability map 113, as illustrated in Figure 3. The binarized map 114 is data obtained by binarizing the probability map 113 with a predetermined threshold. The predetermined threshold may be a real number within the range of 0 to 1. The predetermined threshold may be, for example, the maximum value of pixels not included in the region of interest in the probability map 113. Here, the binarized map 114 is data where the pixel values ​​of pixels not included in the region of interest in the probability map 113 are set to zero. In other words, the binarized map 114 may be, for example, a binary mask in which the values ​​of pixels in the probability map 113 that exceed the predetermined threshold are set to 1, and the values ​​of pixels that fall below the predetermined threshold in the probability map 113 are set to 0.

[0029] The integrated image generation unit 103 generates a target integrated image 115 by integrating the target image 111 and the binarization map 114 by calculating the element-wise product. The target integrated image 115 is an image obtained by integrating the target image 111 and the binarization map 114 by calculating the element-wise product, as illustrated in Figure 3. Here, the integrated image generation unit 103 may also generate a target integrated image 115 by integrating the two by calculating the element-wise product of the pixel values ​​of each pixel contained in the target image 111 and the binarization map 114, and setting the pixel values ​​of pixels contained in regions other than the region of interest in the target image 111 to 0. In other words, the target integrated image 115 may be an image in which the pixel values ​​of pixels contained in regions other than the region corresponding to the first tissue 1111 are set to 0, as illustrated in Figure 3. The target integrated image 115 may be a 3D image, as shown in Figure 3, or it may be a 2D image.

[0030] (Processing performed by the image processing system 100) Next, the processing performed by the image processing system 100 (i.e., the image processing device 1) will be explained using Figure 4. Figure 4 is a flowchart showing an example of the processing flow performed by the image processing system 100.

[0031] First, the probability map generation unit 101 generates a probability map 113 from the target image 111, which includes the object to be inspected, using a trained estimation model 112, showing the probability that each pixel in the target image 111 is included in the region of interest corresponding to the object to be inspected (Step S1: Probability map generation step).

[0032] Next, the binarization unit 102 generates a binarized map 114 which is obtained by binarizing the probability map 113 at a predetermined threshold, and in which the pixel values ​​of pixels not included in the region of interest are set to zero (step S2: binarization step).

[0033] Next, the integrated image generation unit 103 generates a target integrated image 115 by integrating the target image 111 and the binarization map 114 by calculating the elemental product (Step S3: Integrated image generation step).

[0034] With the above configuration, the surface shape of the object to be inspected and extracted from the image is irrelevant. Therefore, the above configuration improves the versatility of the object to be extracted from the image.

[0035] [Embodiment 2] Other embodiments of this disclosure will be described below with reference to Figures 5 and 6. For the sake of clarity, components having the same function as those described in the above embodiments will be denoted by the same reference numerals, and their descriptions will not be repeated.

[0036] (Configuration of image processing system 100a) Figure 5 is a block diagram showing an example of the configuration of an image processing system 100a according to Embodiment 2 of this disclosure. The image processing system 100a differs from the image processing system 100 in that it includes a learning control unit 104 for training the estimation model 112.

[0037] The learning control unit 104 may perform backpropagation on the value of the loss function based on the comparison between (1) and (2) below, and update the parameters of the estimated model 112.

[0038] (1) The pixel value of each pixel in the combined sample image 118, which is obtained by integrating the sample image 116 containing the subject to be examined and the sample probability map 117 generated by the estimation model 112 from the sample image 116, which shows the probability that each image contained in the sample image 116 is contained in the region corresponding to the subject to be examined, by calculating the element product.

[0039] (2) The pixel values ​​of each pixel in the correct image 119, where the pixel values ​​of pixels in areas other than the area corresponding to the object to be examined in the sample image 116 are set to zero.

[0040] Figure 6 illustrates an example of processing in the image processing system 100a. The sample image 116 may be an image corresponding to the training target image 111 which includes the subject to be examined. The sample probability map 117 may be data generated by the estimation model 112 from the sample image 116, as illustrated in Figure 6, and may be data indicating the probability that each image contained in the sample image 116 is included in the region corresponding to the subject to be examined. In other words, the sample probability map 117 may be equivalent to the probability map 113 when the target image 111 is the sample image 116. The sample integrated image 118 may be an image obtained by integrating the sample image 116 and the sample probability map 117 by calculating the element-wise product, as illustrated in Figure 6. Note that the sample image 116 from which the element-wise product is taken may be normalized beforehand, for example, by multiplying each pixel by 1 / 2 and adding 1.

[0041] The correct answer image 119 may be an image in which the pixel values ​​of pixels in areas other than the region corresponding to the object being examined in the sample image 116 have been set to zero. As illustrated in Figure 6, if the sample image 116 includes a first tissue 1111 which is the object being examined, as well as a second tissue 1112 which is not the object being examined, the correct answer image 119 may be an image in which the pixel values ​​of pixels in areas other than the region corresponding to the first tissue 1111 have been set to zero. The correct answer image 119 may be created using, for example, an open-source medical image processing tool such as Horos.

[0042] The loss function may be a quantification of the difference between the pixel value of each pixel in the sample-integrated image 118 and the pixel value of each pixel in the ground truth image 119. The loss function may be, for example, the mean squared error, the mean absolute error, etc.

[0043] For example, the mean squared error (MSE) is expressed by the following formula:

number

number

number

[0044] For example, if the loss function is the mean squared error (MSE), the learning control unit 104 may update the parameters of the estimation model 112 so that the mean squared error (MSE) is minimized.

[0045] With the above configuration, the estimation model can be trained to improve its versatility in determining what to extract from images.

[0046] [Embodiment 3] Other embodiments of this disclosure will be described below with reference to Figures 7 and 8. For the sake of clarity, components having the same function as those described in the above embodiments will be denoted by the same reference numerals, and their descriptions will not be repeated.

[0047] (Configuration of image processing system 100b) Figure 7 is a block diagram showing an example of the configuration of an image processing system 100b according to Embodiment 3 of the present disclosure. The image processing system 100b differs from the image processing system 100 in that it includes a projection image generation unit 105 that generates a projection image 120 obtained by projecting a target integrated image 115, which is a three-dimensional image, into two dimensions.

[0048] Figure 8 illustrates an example of processing in the image processing system 100b. The target image 111 and the target integrated image 115 may be, for example, three-dimensional images as shown in Figure 8. The projection image generation unit 105 may generate a projection image 120, which is an image obtained by projecting values ​​based on both the maximum and minimum pixel values ​​on the projection line onto the target integrated image 115.

[0049] The projected image 120 may be, for example, an image obtained by projecting a 3D image onto a 2D plane. Here, the projected image 120 may be, for example, a MIP (Maximum Intensity Projection) image, a MinIP (Minimum Intensity Projection) image, etc.

[0050] For example, the projection image generation unit 105 may generate a projection image 120 which is a MIP image obtained by projecting the maximum value among the pixel values ​​on the projection line from the target integrated image 115. Alternatively, for example, the projection image generation unit 105 may generate a projection image 120 which is a MinIP image obtained by projecting the minimum value among the pixel values ​​on the projection line from the target integrated image 115. Furthermore, the image management device 2 may store the projection image 120 generated by the image processing device 1 in its storage unit.

[0051] Here, the evaluation function used to evaluate the estimated model 112 may be, for example, the Peak Signal-to-Noise Ratio (PSNR) or MIP-PSNR.

[0052] For example, PSNR can be expressed using the mean squared error MSE described above, as follows:

number

number

[0053] MIP-PSNR is the j-th angle (j=1,2,…,N) relative to the head-to-tail direction.

number

number

number

number

[0054] The estimated model 112 may be verified, for example, by a configuration for verifying the estimated model 112. The configuration for verifying the estimated model 112 may be, for example, a configuration outside the image processing system 100, or it may be a configuration further provided by image processing system 100b, which is an example of image processing system 100. For example, if the evaluation function is MIP-PSNR, the configuration for verifying the estimated model 112 may be a combination of the estimated model 112 and the binarization threshold that maximizes the MIP-PSNR.

[0055] According to the above configuration, the object to be inspected in 3D can be projected into 2D.

[0056] [Embodiment 4] Other embodiments of this disclosure will be described below with reference to Figures 9 and 10. For the sake of clarity, components having the same function as those described in the above embodiments will be denoted by the same reference numerals, and their descriptions will not be repeated.

[0057] (Configuration of image processing system 100c) Figure 9 is a block diagram showing an example of the schematic configuration of an image processing system 100c according to Embodiment 4 of this disclosure. Figure 9 shows an example in which the image processing system 100c is introduced in medical facilities H1 to Hn. In the image processing system 100c, there may be one or more medical facilities. Also, the number of image processing devices 1 provided in the image processing system 100c may be one or more.

[0058] As shown in Figure 9, in the image processing system 100c, each medical facility H1 to Hn may be equipped with an image management device 2, a terminal device 4, and a display device 3. In each of the medical facilities H1 to Hn, the image management device 2, the terminal device 4, and the display device 3 may be directly connected, or they may be connected via a LAN for communication, as shown in Figure 9. Furthermore, the image management device 2, the terminal device 4, and the display device 3 and the image processing device 1 may be connected via the Internet for communication. That is, the image processing device 1 may be a device that applies cloud computing, or it may be a device installed within a specific medical facility or company.

[0059] Figure 10 is a diagram illustrating an example of processing in the image processing system 100c. In the image processing system 100c, the image processing device 1 acquires target images 111 that have been taken and managed at medical facilities H1 to Hn, as illustrated in Figure 10. In this case, the image processing device 1 may acquire the target images 111 along with a facility ID unique to each of the medical facilities H1 to Hn and a subject ID unique to the subject.

[0060] Then, as illustrated in Figure 10, the image processing device 1 transmits the integrated target image 115, obtained by analyzing the acquired target image 111, to each medical facility H1 to Hn, which provided the target image 111. This allows the image processing system 100c to be installed collectively at medical facilities H1 to Hn.

[0061] [Examples of implementation using software] The functions of the image processing devices 1, 1a, and 1b (hereinafter referred to as "devices") included in the image processing systems 100, 100a, 100b, and 100c are programs that cause a computer to function as the device, and these programs can be realized by programs that cause a computer to function as each control block of the device (particularly each part included in the control units 10, 10a, and 10b).

[0062] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., memory) as hardware for executing the program. By executing the program using this control device and storage device, the functions described in each of the embodiments are realized.

[0063] The above program may be recorded on one or more computer-readable recording media, not temporary ones. These recording media may or may not be provided by the above device. In the latter case, the program may be supplied to the above device via any wired or wireless transmission medium.

[0064] Furthermore, some or all of the functions of each of the above control blocks can also be implemented by logic circuits. For example, an integrated circuit in which logic circuits functioning as each of the above control blocks are formed is also included in the scope of this disclosure. In addition, it is also possible to implement the functions of each of the above control blocks by, for example, a quantum computer.

[0065] Furthermore, each process described in the above embodiments may be performed by AI (Artificial Intelligence). In this case, the AI ​​may operate on the control device described above, or it may operate on other devices (for example, an edge computer or a cloud server).

[0066] This disclosure is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of this disclosure.

[0067] 〔summary〕 The image processing system according to Embodiment 1 of the present disclosure includes: a probability map generation unit that generates a probability map showing the probability that each pixel in the target image is included in a region of interest corresponding to the target image, using a trained estimation model from a target image including the object to be inspected; a binarization unit that generates a binarized map obtained by binarizing the probability map at a predetermined threshold, wherein the pixel values ​​of pixels not included in the region of interest are set to zero; and an integrated image generation unit that generates an integrated target image by integrating the target image and the binarized map by calculating the element product.

[0068] The image processing system according to Embodiment 2 of the present disclosure further comprises a learning control unit for training the estimation model, wherein the learning control unit may perform backpropagation on the value of a loss function based on a comparison of (1) the pixel value of each pixel in a sample integration image obtained by integrating a sample image including the object to be inspected and a sample probability map generated by the estimation model from the sample image, which shows the probability that each image included in the sample image is included in the region corresponding to the object to be inspected, by calculating the element product of these two, and (2) the pixel value of each pixel in a ground truth image where the pixel values ​​of pixels included in regions other than the region corresponding to the object to be inspected in the sample image are set to zero, and update the parameters of the estimation model.

[0069] In the image processing system according to aspect 3 of this disclosure, in aspect 2, the loss function may be the mean squared error or the mean absolute error.

[0070] The image processing system according to aspect 4 of the present disclosure is, in any of aspects 1 to 3 above, the object of examination is tissue inside the body of the subject, and the object image may be a medical image obtained by imaging the subject using either a magnetic resonance imaging apparatus or an X-ray computed tomography apparatus.

[0071] In the image processing system according to aspect 5 of this disclosure, the target image in aspect 4 may be a three-dimensional medical image obtained by imaging the subject using cholangiopancreatography with a magnetic resonance imaging device.

[0072] The image processing system according to aspect 6 of the present disclosure, in any of aspects 1 to 5 above, further comprises a projection image generation unit that generates an image by projecting values ​​based on both the maximum and minimum pixel values ​​on a projection line onto the target integrated image, wherein the target image and the target integrated image are three-dimensional images.

[0073] An image processing method according to aspect 7 of the present disclosure is an image processing method performed by one or more computers, comprising: a probability map generation step in which the computer generates a probability map from a target image including an object to be inspected, using a trained estimation model, showing the probability that each pixel in the target image is included in a region of interest corresponding to the object to be inspected; a binarization step in which the computer generates a binarized map obtained by binarizing the probability map at a predetermined threshold, wherein the pixel values ​​of pixels not included in the region of interest are set to zero; and an integrated image generation step in which the computer generates an integrated target image by integrating the target image and the binarized map by calculating the element product.

[0074] The control program according to aspect 8 of the present disclosure is a control program for causing a computer to function as an image processing system as described in any of aspects 1 to 6 above, and is a control program for causing the computer to function as the probability map generation unit, the binarization unit, and the integrated image generation unit.

[0075] The recording medium according to aspect 9 of this disclosure is a computer-readable recording medium on which the control program described in aspect 8 is recorded. [Examples]

[0076] One embodiment of this disclosure is described below.

[0077] In this embodiment, the image processing system 100 was used to estimate the region corresponding to the examination target in the integrated target image 115 from the target image 111. The target image 111 was an image acquired by MRCP. The examination targets were the biliary duct system and the pancreatic duct system. This embodiment was performed for both the case where the estimation model 112 is U-Net (hereinafter also referred to as "U-Net case") and the case where the estimation model 112 is Transformer (hereinafter also referred to as "Transformer case"). In this embodiment, the predetermined threshold when the binarization unit 102 binarizes the probability map 113 was 0.7 in both cases. The ground truth image 119 in this embodiment was generated by removing signals of unnecessary organs from the MRCP image by a technician or physician using a medical image processing tool, and then performing MIP processing 18 to 20 times at equal intervals while rotating 180° with respect to the long axis of the body. This embodiment was also performed for two cases, CASE 1 and CASE 2.

[0078] Figure 11 shows an example of processing results using the image processing system 100 according to this embodiment. Note that all images in Figure 11 are MIP images.

[0079] In this embodiment, as shown in Figure 11, in both the U-Net and Transformer cases, the removal of regions not included in the region of interest and the retention of regions included in the region of interest were generally achieved. In particular, in the case of the Transformer, there were almost no missing or residual regions, and the result was closest to the ground truth image. Also, as shown in Figure 11, in both the U-Net and Transformer cases, there were no clear boundaries, and the probability distribution changed smoothly. Looking at the probability map in the case of the Transformer, the probability was higher around the bile duct and pancreatic duct compared to the other probability maps. [Explanation of symbols]

[0080] 1 Image Processing Device 2. Image management device 3 Display device 4 Terminal devices 10 Control Unit 11 Storage section 100, 100a, 100b, 100c Image Processing System 101 Probability Map Generation Unit 102 Binarization section 103 Integrated Image Generation Unit 104 Learning Control Unit 105 Projection Image Generation Unit 111 Target Images 112 Estimated Models 113 Probability Map 114 Binarized Map 115 Target Integrated Image 116 specimen images 117 Sample Probability Map 118 Integrated sample images 119 Correct image 120 Projection Images 1111 The First Organization 1112 The Second Organization

Claims

1. A probability map generation unit generates a probability map from a target image containing the object to be inspected, using a trained estimation model, which shows the probability that each pixel in the target image is included in the area of ​​interest corresponding to the object to be inspected. A binarization unit that generates a binarized map obtained by binarizing the probability map at a predetermined threshold, wherein the pixel values ​​of pixels not included in the region of interest are set to zero. An integrated image generation unit generates a target integrated image by integrating the target image and the binarized map by calculating the elemental product, An image processing system equipped with the following features.

2. The system further comprises a learning control unit for training the estimation model, The learning control unit, (1) The pixel value of each pixel in the integrated sample image obtained by integrating the sample image containing the subject to be examined and the sample probability map generated by the estimation model from the sample image, which shows the probability that each image contained in the sample image is included in the region corresponding to the subject to be examined, by calculating the element product of the two, (2) The pixel values ​​of each pixel in the correct image, where the pixel values ​​of pixels in the region other than the region corresponding to the object to be examined in the sample image are set to zero, Backpropagation is performed on the value of the loss function based on the comparison, and the parameters of the estimation model are updated. The image processing system according to claim 1.

3. The loss function is either the mean squared error or the mean absolute error. The image processing system according to claim 2.

4. The subject of the examination is the tissue inside the subject's body. The aforementioned target image is These are medical images obtained by imaging the subject using either a magnetic resonance imaging apparatus or an X-ray computed tomography apparatus. The image processing system according to claim 1.

5. The aforementioned target image is This is a three-dimensional medical image of the subject obtained by cholangiopancreatography using a magnetic resonance imaging device. The image processing system according to claim 4.

6. The aforementioned target image and the aforementioned integrated target image are three-dimensional images. The system further includes a projection image generation unit that generates an image by projecting values ​​based on both the maximum and minimum pixel values ​​on the projection line onto the aforementioned integrated target image. The image processing system according to claim 1.

7. An image processing method performed by one or more computers, A probability map generation step in which a computer generates a probability map showing the probability that each pixel in the target image, including the object to be inspected, is included in the region of interest corresponding to the object to be inspected, using a pre-trained estimation model. A binarization step in which a computer generates a binarized map in which the probability map is binarized at a predetermined threshold, and the pixel values ​​of pixels not included in the region of interest are set to zero, A computer generates a target integrated image by calculating the elemental product of the target image and the binarized map, and Image processing methods including [specific details omitted].

8. A control program for causing a computer to function as an image processing system according to claim 1, wherein the computer functions as the probability map generation unit, the binarization unit, and the integrated image generation unit.

9. A computer-readable recording medium that stores the control program described in claim 8.

Images