Image diagnosis assisting device and image processing method

Abstract

The present invention provides an image diagnosis support device and image processing method that can obtain diagnostic results suitable for each device using an existing image diagnostic model. The image diagnosis support device includes: a model reading unit for reading an image diagnostic model, wherein the image diagnostic model outputs a diagnostic result for an input medical image, i.e., a diagnostic image; a storage unit for storing device data, wherein the device data is a plurality of medical images associated with diagnostic results stored by the device; and an adjustment unit for adjusting the diagnostic image to be input to the image diagnostic model or the image diagnostic model based on the device data.

Description

Technical Field

[0001] The present invention relates to an image diagnosis assisting device and an image processing method for assisting diagnosis of lesions contained in medical images obtained by a medical imaging device using artificial intelligence, and more particularly to a technique for improving diagnostic accuracy. Background Art

[0002] Medical imaging devices, such as X-ray CT (Computed Tomography) systems, capture images of lesions and other features, allowing doctors to diagnose lesions contained in these images. With the increasing performance of medical imaging devices, image diagnostic assistance devices that utilize artificial intelligence (AI) to diagnose lesions have been developed in recent years to reduce the burden on doctors. Many image diagnostic assistance devices infer diagnoses from medical images based on a large number of image feature quantities, representing characteristic values of the medical images. This makes it difficult for doctors to determine whether the inferred results are diagnostically useful.

[0003] Patent Document 1 discloses an information processing device that uses a machine-learned image diagnostic model to infer diagnostic names from medical images, and that is capable of presenting as reference information radiographic manifestations that have a significant impact on the inferred diagnostic name. Specifically, based on various image feature quantities of the medical image, the diagnostic name and radiographic manifestations that characterize the medical image are inferred, and radiographic manifestations that are inferred by influencing image feature quantities that are common to the image feature quantities that influence the inferred diagnostic name are presented as reference information.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2019-97805 Summary of the Invention

[0007] Problems to be solved by the invention

[0008] However, in Patent Document 1, image diagnosis is performed using an existing image diagnostic model, and therefore, the diagnostic results may not be suitable for each device. In other words, sufficient diagnostic accuracy may not be achieved due to differences in the nature of the patient, the type of imaging diagnostic device, and other device characteristics.

[0009] Therefore, an object of the present invention is to provide an image diagnosis support device and an image processing method that can obtain a diagnosis result suitable for each device using an existing image diagnosis model.

[0010] Technical solutions to solve problems

[0011] In order to achieve the above-mentioned purpose, the present invention provides an image diagnosis auxiliary device, characterized in that it includes: a model reading unit for reading an image diagnosis model, wherein the image diagnosis model outputs a diagnosis result for an input medical image, i.e., a diagnostic image; a storage unit for storing device data, wherein the device data is a plurality of medical images associated with the diagnosis results held by the device; and an adjustment unit, which adjusts the diagnostic image to be input into the image diagnosis model or the image diagnosis model based on the device data.

[0012] In addition, the present invention provides an image processing method, characterized in that a computer is caused to execute: a model reading step of reading an image diagnostic model, wherein the image diagnostic model outputs a diagnostic result for an input medical image, i.e., a diagnostic image; and an adjustment step of adjusting the diagnostic image to be input into the image diagnostic model or the image diagnostic model based on a plurality of medical images, i.e., device data, associated with the diagnostic results held by the device.

[0013] Effects of the Invention

[0014] According to the present invention, it is possible to provide an image diagnosis support device and an image processing method capable of obtaining a diagnosis result suitable for each device using an existing image diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a hardware configuration diagram of the image diagnosis support device of Example 1.

[0016] Figure 2 This is a functional block diagram of Example 1.

[0017] Figure 3 This is a diagram showing an example of the process flow of Example 1.

[0018] Figure 4 This is a diagram explaining calculation of the importance of each evaluation area.

[0019] Figure 5 This is a diagram showing an example of a display image showing diagnosis results and importance levels.

[0020] Figure 6 This is a diagram showing an example of the results of statistical processing of device data.

[0021] Figure 7 This is a diagram showing an example of an image of items selected for statistical processing.

[0022] Figure 8 This is a diagram showing an example of adjustment of a diagnostic image.

[0023] Figure 9 This is a diagram showing an example of the processing flow of Example 2. DETAILED DESCRIPTION

[0024] Hereinafter, preferred embodiments of the image diagnosis support device and image processing method of the present invention will be described with reference to the accompanying drawings. In the following description and drawings, components having the same functional structure are denoted by the same reference numerals, and repeated descriptions are omitted.

[0025] Example 1

[0026] use Figure 1 The hardware structure of the image diagnosis support device 100 of Example 1 will be described. The image diagnosis support device 100 is a so-called computer. Specifically, it is composed of a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a storage unit 104, a network adapter 105, an input unit 106, and a display unit 107, which are connected via a bus 108 so as to be able to transmit and receive signals. Furthermore, the image diagnosis support device 100 is connected to a medical imaging device 110 and a medical image database 111 via the network adapter 105 and a network 109 so as to be able to transmit and receive signals. Here, "able to transmit and receive signals" means that signals can be transmitted to each other or from one device to the other, either electrically or optically, whether by wire or wirelessly.

[0027] CPU101 is a device that reads the system program stored in ROM102 and controls the operation of each structural element. CPU101 loads the program stored in the storage unit 104 and the data required to execute the program into RAM103 and executes it. The storage unit 104 is a device that stores the program executed by CPU101 and the data required to execute the program. Specifically, it is a device that reads and writes recording devices such as HDD (Hard Disk Drive), SSD (Solid State Drive), and recording media such as IC cards, SD cards, and DVDs. Various data including data required to execute the program are also sent and received from a network 109 such as a LAN (Local Area Network). RAM103 stores the program executed by CPU101 and the process of running the processing.

[0028] The display unit 107 is a device that displays program execution results, etc., and specifically, is a liquid crystal display, touch panel, etc. The input unit 106 is an operating device used by the operator to issue instructions to the image diagnosis support device 100, and specifically, is a keyboard, mouse, etc. A mouse may also be an indicating device other than a touchpad or trackball. Furthermore, if the display unit 107 is a touch panel, the touch panel also serves as the input unit 106. The network adapter 105 is a device used to connect the image diagnosis support device 100 to a network 109, such as a LAN, telephone line, or the Internet.

[0029] The medical imaging device 110 is a device that acquires medical images such as tomographic images that depict the morphology of a lesion, and more specifically, is a Roentgen device, an X-ray CT (Computed Tomography) device, an MRI (Magnetic Resonance Imaging) device, or an ultrasonic diagnostic device. It can also generate three-dimensional medical images by stacking multiple tomographic images. The medical image database 111 is a database system that stores medical images acquired by the medical imaging device 110.

[0030] use Figure 2 The functional block diagram of Example 1 is described. Furthermore, these functions can be implemented by dedicated hardware using an ASIC (Application Specific Integrated Circuit) or an FPGA (Field-Programmable Gate Array), or by software running on CPU 101. The following description will describe the case where each function is implemented by software. Example 1 includes a model reading unit 201, a judgment unit 202, and an adjustment unit 203. Each unit is described below.

[0031] The model reading unit 201 reads an image diagnostic model that outputs a diagnostic result for an input medical image from the storage unit 104 or from the Internet via the network adapter 105. The image diagnostic model is an existing program constructed using a random forest, SVM (Support Vector Machine), a hierarchical neural network, or the like, and is any program whose processing algorithm is black-boxed. The medical image input to the image diagnostic model can be any image, for example, a two-dimensional or three-dimensional medical image or a portion of a medical image. Furthermore, the diagnostic result output by the image diagnostic model may include, for example, a determination of whether a lesion is benign or malignant, or the probability that a lesion is malignant.

[0032] The judgment unit 202 determines whether the diagnostic results output from the image diagnostic model are sufficient as diagnostic accuracy. This judgment by the judgment unit 202 uses a threshold value determined for each device. If the diagnostic result is determined to be insufficient, the adjustment unit 203 takes action. The judgment unit 202 is not a required function, and the adjustment unit 203 can also operate without the judgment by the judgment unit 202.

[0033] The adjustment unit 203 adjusts the diagnostic image input to the image diagnosis model based on multiple medical images associated with the diagnostic results held by the device, i.e., the device data. In other words, in order to improve the accuracy of the diagnostic results output from the image diagnosis model, the size of the diagnostic image is enlarged or reduced, or any area contained in the diagnostic image is deleted. Figure 8 To be described later.

[0034] use Figure 3 An example of the process flow of Example 1 will be described.

[0035] (S301)

[0036] The model reading unit 201 reads the image diagnosis model from the Internet or the like via the storage unit 104 or the network adapter 105 .

[0037] (S302)

[0038] The diagnostic image, which is a medical image to be diagnosed, is input into the image diagnosis model and is read from a RIS (Radiology Information System) or the like stored in the equipment.

[0039] (S303)

[0040] For example, the image diagnostic model outputs the probability that a tumor contained in the diagnostic image is malignant as the diagnostic result for the diagnostic image input in S302. Furthermore, the importance of each evaluation region may be output along with the diagnostic result. Evaluation regions are areas extracted from the diagnostic image that contribute to the diagnostic result, and the importance of evaluation regions is a value indicating the degree to which each of the multiple evaluation regions influences the diagnostic result. By displaying the importance of each evaluation region, the operator can identify the evaluation region that has an impact, i.e., the evaluation region that forms the basis for the diagnosis, based on the diagnostic result.

[0041] use Figure 4 The following describes an example of calculating the importance of each evaluation region. The evaluation regions, representing the diagnostic results, are extracted based on treatment guidelines, which describe the basis and sequence of treatment. For example, in determining whether a lung tumor is good or bad, the lung tumor region, air-filled region, chest wall region, and vascular region are extracted as evaluation regions based on lung cancer treatment guidelines.

[0042] The extracted evaluation regions are input into the image diagnosis model as images representing luminance distribution information, and the importance of each evaluation region is calculated. Specifically, the importance of each evaluation region is calculated based on the combination of the evaluation regions and the diagnostic results obtained by inputting various combinations of evaluation regions into the image diagnosis model. The combinations of evaluation regions are represented, for example, by a sampling table 401.

[0043] The sampling table 401 is a table consisting of rows of items arranged in the order of the extracted n evaluation regions and a sampling matrix. The values of the matrix elements of the sampling matrix indicate whether each evaluation region is included in the image corresponding to each row. That is, if the value in each row is 1, the corresponding evaluation region is included in the image of that row, and the evaluation region in the column with a value of 0 is not included in the image of that row. The evaluation region not included in the image of that row has a brightness value of 0 or is filled with black. For example, the value of region 1 is 1, and the values of regions 2 to n are 0. Therefore, the image of the first row becomes an image containing only region 1. In the case where region 1 is a lung tumor region, the image becomes a lung tumor region image 402 containing only the lung tumor region. In addition, the value of region 1 is 0, and the values of regions 2 to n are 1. Therefore, the image of the second row becomes an image that does not contain region 1.

[0044] The size of the sampling matrix is determined by the number of evaluation areas n and the number of combinations of evaluation areas m. That is, the sampling matrix becomes a matrix with m rows and n columns. In order to improve the calculation accuracy of the importance, the number of combinations of evaluation areas m is preferably the number of all combinations of n evaluation areas, that is, 2 n For example, m=n may be set to indicate a combination including n evaluation areas.

[0045] When an image of each combination of evaluation regions represented by a sampling matrix or the like is input to the image diagnosis model, a diagnosis result for each combination is output. Figure 4 In the example, a lung tumor region image 402, an air region image 403, a chest wall region image 404, and a blood vessel region image 405, each of which is contained in the lung tumor region, the air region, the chest wall region, and the blood vessel region, are input into the image diagnosis model to output a diagnosis result 406. Figure 4 The diagnostic result 406 outputs 0.95 as the importance of each evaluation area in the lung tumor area image 402, 0.84 in the air-containing area image 403, 0.42 in the chest wall area image 404, and 0.41 in the blood vessel area image 405. Figure 4 From the results shown in the example, the operator can confirm that the lung tumor region and the air-filled region are more important than the chest wall region and the blood vessel region as the basis for the diagnosis result.

[0046] use Figure 5, an example of the screen displayed in S303 is described. Figure 5 The screen includes a diagnostic image display unit 501, an evaluation region display unit 502, an importance display unit 503, and a diagnostic result display unit 504. The diagnostic image display unit 501 displays an image of the tumor region, which is the region containing the tumor in the diagnostic image. The evaluation region display unit 502 displays the image displayed in the diagnostic image display unit 501, divided into multiple evaluation regions. The importance display unit 503 displays an image in which the importance calculated for each evaluation region is visualized. The diagnostic result display unit 504 displays the diagnostic result, indicating a 97% probability that the tumor is malignant, and also displays information such as the size of the tumor, the characteristics of the tumor margin, and the presence of air.

[0047] (S304)

[0048] Based on the diagnostic results output in S303, the determination unit 202 determines whether the diagnostic accuracy of the image diagnostic model is sufficient. If the diagnostic accuracy is insufficient, the process returns to S303 via S305. If it is sufficient, the process proceeds to S306. The diagnostic accuracy is determined using a threshold value determined for each device. For example, if the diagnosis result, i.e., the probability of a tumor being malignant, is lower than a threshold value, the diagnostic accuracy is determined to be insufficient. The threshold value determined for each device, for example, is set to a value that represents the probability that the diagnostic result is correct.

[0049] (S305)

[0050] The adjustment unit 203 adjusts the diagnostic image based on the device data. Once the diagnostic image is adjusted, the process returns to S303, and the image diagnosis model outputs a diagnosis result for the adjusted diagnostic image. The diagnostic image is adjusted based on, for example, the results of statistical processing of the device data.

[0051] use Figure 6 , which illustrates an example of the results of statistical processing of device data. Figure 6 6 shows a device histogram 601 and a teaching histogram 602. The device histogram 601 is a histogram generated using each medical image of the device data. Figure 6 602 is a histogram with the size of the tumor contained in each medical image as the horizontal axis. Similarly to the device histogram 601, the horizontal axis represents the size of the tumor contained in each medical image. Furthermore, the teaching data is a separate set of medical images from the device data, and the diagnostic results are associated with each medical image. Furthermore, statistical processing of device data is not limited to the creation of a histogram with tumor size as the horizontal axis; statistical processing can also be performed using items selected by the operator.

[0052] use Figure 7, an example of a screen for selecting items for statistical processing of device data is described. Figure 7 The screen includes: Use Figure 5 The diagnostic image display unit 501, evaluation area display unit 502, importance display unit 503, diagnostic result display unit 504, item selection unit 701, and statistical result display unit 702 are shown. Item selection unit 701 is a checkbox for the operator to select items for statistical processing. The items corresponding to the selected checkboxes are used for statistical processing. By selecting the desired items from item selection unit 701, the operator can improve diagnostic accuracy based on their perspective.

[0053] The statistical result display unit 702 displays the results of statistical processing using the items selected in the item selection unit 701. The statistical result display unit 702 displays not only the results of statistical processing of the device data, but also the results of statistical processing of the teaching data. By comparing the two data, it is clear whether the properties of the device are consistent with the image diagnosis model. For example, by comparing Figure 6 By comparing device histogram 601 and training histogram 602, it is discovered that there is a difference in the distribution of tumor sizes between the device data and the training data, and that the device properties do not conform to the image diagnostic model. To align the device properties with the image diagnostic model, adjustment unit 203 may also adjust the diagnostic image to reduce the difference between the device data and the training data. For example, the size of the diagnostic image may be adjusted based on the difference between the mode value of device histogram 601 and the mode value of training histogram 602.

[0054] use Figure 8 , an example of adjusting a diagnostic image is described. Figure 8 801 and the second adjustment example 802 are shown in FIG. First adjustment example 801 is an example of reducing the size of a diagnostic image based on a comparison between the device histogram 601 and the training histogram 602. Specifically, if the mode of the device histogram 601 is 30 mm, and the mode of the training histogram 602 is 20 mm, then the image is reduced by multiplying one side by 2 / 3. Second adjustment example 802 is an example of removing high-brightness areas near a tumor. Specifically, the high-brightness area below the tumor, i.e., the chest wall area, is filled with black to remove the high-brightness area.

[0055] (S306)

[0056] The adjustment unit 203 stores the adjustment parameters for the diagnostic image in S305, i.e., the adjustment parameters, in the storage unit 104. For example, if one side of the diagnostic image is multiplied by 2 / 3 to reduce the size of the diagnostic image, 2 / 3 is stored as the adjustment parameter. Furthermore, if the process proceeds to S306 without passing through S305, the adjustment parameters may not be stored. Furthermore, by the time S306 is reached, a diagnostic result suitable for the device has been obtained in S303.

[0057] By adjusting the diagnostic image so that the properties of the device conform to the image diagnostic model through the processing flow described above, it is possible to obtain a diagnostic result suitable for each device using the existing image diagnostic model. Figure 3 In the processing flow of , the judgment process of S304 is not necessary. In the case where the judgment process of S304 is not required, the image diagnosis model only needs to output the diagnosis result for the diagnosis image subjected to the adjustment process of S305 in S303.

[0058] Example 2

[0059] In Example 1, adjustment of the diagnostic image input to the image diagnostic model based on device data is described. In order to improve diagnostic accuracy, not only the diagnostic image input to the image diagnostic model is adjusted, but the image diagnostic model can also be adjusted through relearning. In Example 2, generation of relearning data for adjusting the image diagnostic model based on device data and adjustment of the image diagnostic model using the relearning data are described. In addition, the overall structure of Example 2 is the same as that of Example 1, so the description is omitted. In addition, the functional block diagram of Example 2 is the same as that of Example 1, but because the operation of the adjustment unit 203 is different, the adjustment unit 203 is described.

[0060] The adjustment unit 203 of the second embodiment generates relearning data for adjusting the image diagnostic model based on multiple medical images associated with diagnostic results stored in the device, i.e., device data. The image diagnostic model is adjusted by relearning using the generated relearning data to improve the accuracy of the diagnostic results.

[0061] use Figure 9 , an example of the process flow of Example 2 is described. Among them, S301 to S303 are the same processes as Example 1, so the description is simplified.

[0062] (S301)

[0063] The model reading unit 201 reads the image diagnosis model.

[0064] (S302)

[0065] The diagnostic image is input to the image diagnosis model.

[0066] (S303)

[0067] The image diagnosis model outputs a diagnosis result for the diagnostic image.

[0068] (S904)

[0069] The determination unit 202 determines whether the diagnostic accuracy of the image diagnostic model is sufficient based on the diagnostic result output in S303. If the diagnostic accuracy is insufficient, the process returns to S303 via S905. If it is sufficient, the process proceeds to S906.

[0070] (S905)

[0071] The adjustment unit 203 generates relearning data based on the device data and adjusts the image diagnosis model using the generated relearning data. When the image diagnosis model is adjusted, the process returns to S303, and the adjusted image diagnosis model outputs a diagnosis result for the diagnostic image.

[0072] Several examples of relearning data are described below. To obtain diagnostic results tailored to the device, it is desirable that the relearning data reflect the properties of the device data. Therefore, the device data can also be used directly as relearning data.

[0073] Furthermore, if the number of images in the device data is significantly smaller than the number of images in the teaching data used to generate the image diagnostic model, insufficient learning may prevent sufficient diagnostic accuracy. In such cases, by adding teaching data to the device data, relearning data can be generated that does not cause a learning deficit.

[0074] However, if the properties of the added teaching data are significantly different from the properties of the device data, it will be difficult to obtain a diagnosis result suitable for the device. Therefore, it is also possible to extract data with properties close to the device data from the teaching data and add the extracted data to the device data to generate relearning data. For example, it is also possible to use Figure 6 The probability function calculated using the illustrated device histogram 601 extracts data to be added to the device data from the training data. The probability function is calculated by fitting the device histogram 601. Furthermore, the distribution of data extracted from the training data using the probability function becomes similar to that of the device histogram 601, resulting in data that closely resembles the properties of the device data. Therefore, the image diagnostic model adjusted using the relearned data generated by relearning can output diagnostic results that are appropriate for the device.

[0075] (S906)

[0076] The adjustment unit 203 stores the image diagnostic model adjusted in S905, i.e., the relearned image diagnostic model, in the storage unit 104. Furthermore, if the process proceeds to S906 without passing through S905, the relearned image diagnostic model need not be stored. Furthermore, by the time S906 is reached, a diagnosis result appropriate for the device is obtained in S303.

[0077] By adjusting the image diagnosis model to match the properties of the device through the processing flow described above, it is possible to use the existing image diagnosis model to obtain a diagnosis result suitable for each device. Figure 9 In the process flow of the present invention, the judgment process of S904 is not essential. In the case where the judgment process of S904 is not performed, it is sufficient for the image diagnosis model adjusted in S905 to output the diagnosis result of the diagnosis image in S303.

[0078] The above describes a number of embodiments of the present invention. The present invention is not limited to these embodiments and includes various variations. For example, the above embodiments, which are described in detail to facilitate understanding of the present invention, are not limited to the entire structure described. In addition, a portion of the structure of a particular embodiment can be replaced with the structure of another embodiment. Furthermore, the structure of another embodiment can be added to the structure of a particular embodiment. In addition, a portion of the structure of each embodiment can be added, deleted, or replaced with another structure.

[0079] Description of Reference Numerals

[0080] 100: Image diagnosis support device, 101: CPU, 102: ROM, 103: RAM, 104: Storage unit, 105: Network adapter, 106: Input unit, 107: Display unit, 108: Bus, 109: Network, 110: Medical imaging device, 111: Medical image database, 201: Model reading unit, 202: Determination unit, 203: Adjustment unit, 401: Sampling table, 402: Lung tumor area Image, 403: Gas-containing area image, 404: Chest wall area image, 405: Blood vessel area image, 406: Diagnosis result, 501: Diagnosis image display unit, 502: Evaluation area display unit, 503: Importance display unit, 504: Diagnosis result display unit, 601: Equipment histogram, 602: Teaching histogram, 701: Item selection unit, 702: Statistical result display unit, 801: First adjustment example, 802: Second adjustment example.

Claims

1. An image diagnosis assisting device, characterized in that: include: a model reading unit for reading an image diagnostic model, wherein the image diagnostic model outputs a diagnosis result for an input medical image, i.e., a diagnostic image; a storage unit for storing device data, wherein the device data is a plurality of medical images associated with diagnosis results held by the device; and an adjustment unit that adjusts a diagnostic image to be input to the image diagnosis model or the image diagnosis model based on the device data, The adjustment unit generates a device histogram that is a histogram of sizes of tumors included in each medical image of the device data, and adjusts the size of the diagnostic image based on the device histogram.

2. The image diagnosis assisting device according to claim 1, wherein: The image diagnosis model is generated by learning a plurality of medical images, namely, teaching data, which are different from the device data. The adjustment unit generates a teaching histogram that is a histogram of the sizes of tumors contained in each medical image of the teaching data, and adjusts the size of the diagnostic image based on a device mode value that is a mode value of the device histogram and a teaching mode value that is a mode value of the teaching histogram.

3. The image diagnosis assisting device according to claim 2, wherein: The adjustment unit multiplies a value obtained by dividing the teaching mode value by the device mode value by the size of the diagnostic image.

4. The image diagnosis assisting device according to claim 1, wherein: When a high-brightness region exists near a tumor in each medical image of the device data, the adjustment unit adjusts the diagnostic image by removing the high-brightness region.

5. The image diagnosis assisting device according to claim 1, wherein: The apparatus further comprises a judgment unit for judging whether adjustment by the adjustment unit is necessary based on the diagnosis result of the image diagnosis model. The adjuster adjusts the diagnostic image to be input to the image diagnostic model or the image diagnostic model when the determiner determines that the adjustment is necessary.

6. The image diagnosis assisting device according to claim 1, wherein: A screen for selecting items to be used for statistical processing of the device data is displayed.

7. An image diagnosis assisting device, characterized in that: include: a model reading unit for reading an image diagnostic model, wherein the image diagnostic model outputs a diagnosis result for an input medical image, i.e., a diagnostic image; a storage unit for storing device data, wherein the device data is a plurality of medical images associated with diagnosis results held by the device; and an adjustment unit that adjusts a diagnostic image to be input to the image diagnosis model or the image diagnosis model based on the device data, The adjustment unit generates an apparatus histogram, which is a histogram of the size of a tumor included in each medical image of the apparatus data, and adjusts the image diagnosis model by performing relearning using relearning data, which is data generated based on the apparatus histogram.

8. The image diagnosis assisting device according to claim 7, wherein: The image diagnosis model is generated by learning a plurality of medical images, namely, teaching data, which are different from the device data. The adjustment unit extracts the relearning data from the teaching data based on the device histogram.

9. The image diagnosis assisting device according to claim 8, wherein: The adjustment unit calculates a probability function for extracting a medical image from the teaching data based on the device histogram, and extracts the relearning data using the probability function.

10. An image processing method, characterized in that: Causes the computer to execute: a model reading step of reading an image diagnostic model, wherein the image diagnostic model outputs a diagnostic result for an input medical image, i.e., a diagnostic image; and an adjustment step of adjusting the diagnostic image to be input to the image diagnostic model or the image diagnostic model based on a plurality of medical images associated with the diagnostic results held by the device, i.e., device data; In the adjusting step, a device histogram, which is a histogram of the sizes of tumors included in each medical image of the device data, is generated, and the size of the diagnostic image is adjusted based on the device histogram.

Images