JP2020144498A - Image diagnosis support device and image processing method - Google Patents

Abstract

To provide an image diagnosis support device and an image processing method which can present a basis on which a diagnostic result is determined with respect to an arbitrary image diagnostic model.SOLUTION: The image diagnosis support device comprises: a model reading unit which reads an image diagnostic model for outputting diagnostics result in response to an input medical image; a determination region extraction unit which extracts a plurality of determination regions from the medical image on the basis of a plurality of determination items; and an importance calculation unit which acquires the diagnostic results obtained by inputting the determination regions to the image diagnostic model while changing combinations of the determination regions and calculates importance of each of the determination regions on the basis of the acquired diagnostic results and combinations of the determination regions.SELECTED DRAWING: Figure 2

Description

The present invention relates to an image diagnosis in which a lesion site included in a medical image acquired by a medical image imaging device is judged to be good or bad by artificial intelligence or the like, and particularly an image displaying the importance of each image feature amount which is a basis for judging the diagnosis result. It relates to a diagnostic support device and an image processing method.

A medical image imaging device represented by an X-ray CT (Computed Tomography) device or the like is a device that images the morphology of a lesion site or the like, and the acquired medical image is used for image diagnosis by an image interpreting doctor. In recent years, an image diagnosis support device has been developed that makes an image diagnosis using artificial intelligence (AI) to judge the quality of a lesion site in order to reduce the burden on the image interpreting doctor due to the high performance of the medical image imaging device. ing. In many diagnostic imaging support devices, only the quality judgment of the lesion site is output in response to the input of a plurality of feature quantities extracted from the medical image.

Patent Document 1 discloses a diagnostic support device capable of quantitatively presenting not only the output of good / bad judgment but also the degree of contribution that is a factor leading to the good / bad judgment of the lesion site. Specifically, the input image features, the weight value of each path connecting the input layer, the intermediate layer, and the output layer of the hierarchical neural network constructed to judge the quality, the output value, and The contribution value is calculated for each type of input image feature amount based on the intermediate node value.

Japanese Patent No. 4480508

However, Patent Document 1 cannot support an existing diagnostic imaging model because it is necessary to construct a hierarchical neural network prior to calculating the degree of contribution, which is a factor leading to the judgment of quality. That is, it is not possible to show the basis for judging the diagnosis result for the diagnosis result output by an arbitrary diagnostic imaging model constructed by Random forest or SVM (Support Vector Machine).

Therefore, an object of the present invention is to provide an image diagnosis support device and an image processing method capable of presenting a basis for determining a diagnosis result in an arbitrary image diagnosis model.

In order to achieve the above object, the present invention has a model reading unit that reads a diagnostic imaging model that outputs a diagnostic result for an input medical image, and a plurality of determination regions from the medical image based on a plurality of determination items. The determination area extraction unit to be extracted and the diagnosis result obtained by inputting the determination area into the image diagnosis model are acquired while changing the combination of the determination areas, and each acquired diagnosis result and the combination of the determination area are obtained. The image diagnosis support device is characterized by comprising an importance calculation unit for calculating the importance for each determination region based on the above.

Further, the present invention includes a model reading step of reading an image diagnosis model that outputs a diagnosis result for an input medical image, and a judgment area extraction step of extracting a plurality of judgment areas from the medical image based on a plurality of judgment items. And, the diagnosis result obtained by inputting the judgment area into the image diagnosis model is acquired while changing the combination of the judgment areas, and the judgment area is based on each acquired diagnosis result and the combination of the judgment area. It is an image processing method characterized by having a computer execute an importance calculation step for calculating the importance of each.

According to the present invention, it is possible to provide an image diagnosis support device and an image processing method capable of presenting a judgment basis of a diagnosis result in an arbitrary image diagnosis model.

It is a hardware block diagram of the image diagnosis support apparatus of Example 1. FIG. It is a functional block diagram of Example 1. FIG. It is a figure which shows an example of the process flow of Example 1. FIG. It is a figure which shows an example of the determination item of Example 1. FIG. It is a figure which shows the area division for extracting the determination area of Example 1. FIG. It is a figure explaining the calculation of the importance degree of Example 1. It is a figure which shows an example of the display screen of Example 1. FIG. It is a figure which shows an example of the process flow of Example 2. FIG. It is a figure which shows an example of the operation screen of Example 2. FIG. It is a figure which shows an example of the process flow of Example 3. FIG. It is a figure which shows an example of the operation screen of Example 3. FIG.

Hereinafter, preferred examples of the image diagnosis support device and the image processing method according to the present invention will be described with reference to the accompanying drawings. In the following description and the accompanying drawings, the components having the same functional configuration are designated by the same reference numerals to omit duplicate description.

The hardware configuration of the diagnostic imaging support device 100 of this embodiment will be described with reference to FIG. The diagnostic imaging support device 100 is a so-called computer. Specifically, 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 transmit and receive signals by a bus 108. Possible to be connected and configured. Further, the diagnostic imaging support device 100 is connected to the medical image imaging device 110 and the medical image database 111 via a network adapter 105 and a network 109 so that signals can be transmitted and received. Here, "to be able to send and receive signals" is a state in which signals can be passed to each other or from one to the other, whether electrically or optically, wired or wireless.

The CPU 101 is a device that reads a system program or the like stored in the ROM 102 and controls the operation of each component. The CPU 101 loads the program stored in the storage unit 104 and the data necessary for executing the program into the RAM 103 and executes the program. The storage unit 104 is a device that stores a program executed by the CPU 101 and data necessary for executing the program. Specifically, a recording device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), an IC card, or the like. A device that reads and writes to a recording medium such as an SD card or DVD. Various data including data necessary for program execution are also transmitted and received from a network 109 such as a LAN (Local Area Network). The RAM 103 stores a program executed by the CPU 101, an intermediate progress of arithmetic processing, and the like.

The display unit 107 is a device for displaying the result of program execution, and specifically, a liquid crystal display or the like. The input unit 106 is an operation device in which the operator gives an operation instruction to the image diagnosis support device 100, and specifically, a keyboard, a mouse, or the like. The mouse may be another pointing device such as a trackpad or trackball. When the display unit 107 is a touch panel, the touch panel also functions as the input unit 106. The network adapter 105 is for connecting the diagnostic imaging support device 100 to a network 109 such as a LAN, a telephone line, or the Internet.

The medical image imaging device 110 is a device that acquires a medical image such as a tomographic image that images the morphology of a lesion site or the like, and specifically, an X-ray device, an X-ray CT (Computed Tomography) device, or an MRI (Magnetic) device. Resonance Imaging) equipment, ultrasonic diagnostic equipment, etc. A three-dimensional medical image is created by stacking a plurality of tomographic images. The medical image database 111 is a database system that stores medical images acquired by the medical image imaging device 110.

The functional block diagram of this embodiment will be described with reference to FIG. These functions may be configured by dedicated hardware using an ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or the like, or may be configured by software running on the CPU 101. .. In the following description, the case where each function is configured by software will be described. This embodiment includes a model reading unit 201, a determination area extraction unit 202, and an importance calculation unit 203. Each part will be described below.

The model reading unit 201 reads an image diagnosis model that outputs a diagnosis result for the input medical image 200. The diagnostic imaging model is an existing program constructed by Random forest, SVM (Support Vector Machine), hierarchical neural network, etc., and is an arbitrary program in which the processing algorithm is black-boxed. The medical image 200 input to the diagnostic imaging model may be any image, and may be a part of a two-dimensional medical image or a three-dimensional medical image. The diagnostic result output from the diagnostic imaging model is, for example, the result of determining whether the lesion site is good or bad, such as the probability that the lesion site is malignant.

The determination area extraction unit 202 extracts a plurality of determination areas from the medical image 200 based on the plurality of determination items. The determination items are items related to the diagnosis result, and examples of the determination items will be described later with reference to FIG. An example of the extracted determination region will be described later with reference to FIG.

The importance calculation unit 203 acquires the diagnosis result obtained by inputting the judgment area into the diagnostic imaging model while changing the combination of the judgment areas, and based on each acquired diagnosis result and the combination of the judgment areas, the judgment area Calculate the importance of each. The diagnostic results obtained while changing the combination of the determination areas will be described later with reference to FIG. The calculated importance of each determination area is illustrated in FIG. 7 and will be described later.

An example of the processing flow of this embodiment will be described with reference to FIG.

(S301)
The image diagnosis model is read into the model reading unit 201, and the medical image 200 to be diagnosed is input to the determination area extraction unit 202.

(S302)
The determination area extraction unit 202 extracts a plurality of determination areas from the input medical image 200. For the extraction of the determination area, for example, a medical practice guideline or the like in which the basis and procedure of medical care are described is used. In particular, the lung cancer clinical practice guideline that describes that the judgment item 401 such as the size, margin, shape, aeration, and brightness distribution of the tumor shown in FIG. Is preferable. Judgment item 401 in FIG. 4 was prepared based on the lung cancer screening screening guideline Lung-RADs prepared by the American College of Radiology (ACR).

Judgment item 401 can determine the degree of possibility of benign or malignant by findings. For example, if the tumor size is less than 8 mm or the margin is smooth, it is likely to be benign, and if the tumor size is 15 mm or more, the margin is a radial crown, or aeration is included. If so, there is a high possibility of malignancy. Although all of the judgment items 401 are greatly involved in the judgment of the quality of the lesion site, an error may be mixed in the process of quantification. Therefore, in this embodiment, a determination area is extracted from the medical image 200 based on the determination item, and the extracted determination area is input to the diagnostic imaging model. That is, without digitizing each determination item, the determination area including the information about each determination item is input to the diagnostic imaging model as an image representing the distribution information of the luminance value. By inputting the judgment area, which is an image containing information about each judgment item, into the diagnostic imaging model, it is possible to prevent the mixing of errors that occur in the process of digitization, so that the calculation accuracy can be improved in the subsequent processing.

An example of region division for extracting a determination region from the medical image 500 including the lung tumor 501 will be described with reference to FIG. The medical image 500 includes an aeration 502, a chest wall 503, a background 504, and a blood vessel 505, along with a lung tumor 501. An edge-enhanced image 506 can be obtained by applying a preprocessing filter such as Sobel or Previt to the medical image 500. Then, with respect to the edge-enhanced image 506, the divided image 507 is obtained by region division based on the luminance value, for example, boundary extraction based on the luminance gradient or grouping of pixels having similar luminance values.

The divided image 507 includes, for example, an aeration region 508 corresponding to the aeration 502, and since the aeration region 508 corresponds to the aeration described in the determination item 401, it is extracted as one of the determination regions. The divided image 507 also includes regions corresponding to the chest wall 503, blood vessels 505, etc., and these regions are anatomically divided regions that are not described in the determination item 401, and are therefore extracted as determination regions. Is also good. Further, a determination region may be further extracted based on the description of the determination item 401, for example, a region of less than 8 mm, which is a region of tumor size less than 8 mm, or a region of 8 to 15 mm, which is a region of tumor size of 8 mm to 15 mm. 510 and the like may be extracted.

(S303)
The importance calculation unit 203 calculates the importance for each determination area. Specifically, various combinations of the determination regions extracted in S302 are input to the diagnostic imaging model, diagnostic results for each combination of the determination regions are acquired, and each acquired diagnostic result and the combination of the determination regions are used. Based on this, the importance of each determination area is calculated. Hereinafter, this step will be described in more detail.

An example of the diagnosis result acquired for each combination of the determination areas will be described with reference to FIG. The combination of determination regions is represented by, for example, the sampling table 601. The sampling table 601 is a table composed of an item row in which n determination regions extracted in S302 are arranged and a sampling matrix, and whether or not each determination region is included in the image corresponding to each row is included in the sampling matrix. Indicated by the value of the matrix element. That is, if the value in each row is 1, the corresponding determination area is included in the image of the row, the determination area of the column having the value 0 is not included in the image, and the determination area not included is the brightness value 0. It is said to be or filled with black. For example, the image in the first line has a value of 1 in the area 1 and 0 in the values from the area 2 to the area n, so that the image includes only the area 1. When the region 1 is a lung tumor region, the image is a lung tumor region image 602 including only the lung tumor region. Further, since the value of the area 1 is 0 and the value of the area 2 to the area n is 1 in the image of the second line, the image does not include only the area 1.

The size of the sampling matrix is determined according to the number n of the determination regions and the number m of the combinations of the determination regions, and the sampling matrix is a matrix of m rows and n columns. The number m of the combinations of the determination regions is preferably 2 ⁿ , which is the total number of combinations of n determination regions, in order to improve the accuracy of calculating the importance, but this is not always the case. For example, m = n may be set to represent a combination in which only each of the n determination regions is included.

When an image for each combination of judgment areas represented by a sampling matrix or the like is input to the diagnostic imaging model, the diagnosis result for each combination is output. In FIG. 6, the lung tumor region image 602, the aerated region image 603, the chest wall region image 604, and the blood vessel region image 605, which include only each of the lung tumor region, the aerated region, the chest wall region, and the blood vessel region, are used as diagnostic imaging models. An example is a diagnostic result 606 that is output by being input. In FIG. 6, 0.95 is output as the probability of being malignant with respect to the lung tumor region image 602, 0.84 in the aerated region image 603, 0.42 in the chest wall region image 604, and 0.42 in the vascular region image 605. 0.41 is output.

From the results illustrated in FIG. 6, it can be judged that the lung tumor region and the aerobic region are relatively more important in obtaining the diagnosis result from the diagnostic imaging model than the chest wall region and the blood vessel region. In this embodiment, the importance levels w1, w2, ..., Wn of the n determination regions are calculated by the following equations.

w = (A ^T A) ^-1 A ^T y ... (Equation 1)
Here, A is a sampling matrix of m rows and n columns, w = [w1, w2, ..., Wn] is the importance of each judgment area, and y = [y1, y2, ..., ym] is a combination of judgment areas. It is a diagnosis result for each. The matrix element of A is 0 or 1, and the element of y is a numerical value of 0 or more and 1 or less.

(S304)
The importance of each determination area calculated in S303 is displayed on the display unit 107. The importance display screen 700, which is a display example of this step, will be described with reference to FIG. 7. The importance display screen 700 includes a diagnosis result display unit 701, an importance display unit 702, and an importance image display unit 703. The diagnosis result display unit 701 displays, for example, a probability of 98.5% that the lesion site is malignant as a diagnosis result for the input medical image.

On the importance display unit 702, each importance calculated for each determination area is displayed in association with the determination area. In addition, each determination area may be classified and displayed in a group having a positive influence or a negative influence on the diagnosis result according to the value of each importance. In FIG. 7, "margin" and "aeration" are shown in the positively affected group, "tumor size <8 mm" is shown in the negatively affected group, and "tumor size 8 to 15 mm" is shown in the unaffected group. "Background", "blood vessel" and "chest wall" are displayed. The operator can confirm that "margin" is the first basis for judgment of the diagnosis result and "aeration" is also included in the basis for judgment by the classified display as illustrated in FIG. 7.

The importance image display unit 703 displays an importance image that is colored with a color corresponding to the importance of each determination area. The color bar 704 shows the correspondence between importance and color, for example, black for negative values, white for values near zero, and diagonal lines for positive and relatively large values. Be associated. An importance image is generated by coloring each determination region of the divided image 507 according to the correspondence relationship in the color bar 704, for example. In FIG. 7, the less than 8 mm region 705 is displayed in black, the marginal region 706 and the aerated region 707 are displayed in diagonal lines, and the 8 to 15 mm region 708 and the background region 709, the blood vessel region 710, and the chest wall region 711 are displayed in white. By displaying the importance image illustrated in FIG. 7, the operator can grasp at a glance the determination area that is the basis for determining the diagnosis result.

Through the processing flow described above, it is possible to present the basis for judgment on the diagnosis result output from an arbitrary diagnostic imaging model. By presenting the basis for determining the diagnosis result, the operator can understand the reason for the diagnosis result, and the case where the diagnosis result is not accepted by the operator can be reduced.

In the processing flow shown in FIG. 3, the importance of each judgment area, which is the basis for judgment of the diagnosis result output for a single medical image, was calculated and presented. In this embodiment, a plurality of medical images are input to the diagnostic imaging model, and the importance of each judgment area output for each medical image is used to be important for a medical image group which is a set of a plurality of medical images. The global importance, which is the degree, may be calculated. Global importance can be an evaluation index for the diagnostic imaging model used for diagnostic imaging. For example, if the determination area having high global importance corresponds to the determination item emphasized in the clinical practice guidelines such as Lung-RADs, it can be determined that the image diagnosis model used for the image diagnosis has high reliability. Further, if the judgment area having high global importance does not correspond to the judgment item in which importance is given, it can be judged that the content needs to be updated due to insufficient accuracy of the diagnostic imaging model.

The calculation of global importance will be described. The global importance is calculated by weighting and adding the importance of each judgment area output for a plurality of medical images. The weighting factor used for the weighted addition is given a larger value for a more characteristic medical image. The characteristics of each medical image appear in the variation in importance for each judgment area. For example, the importance of the three determination regions of the first image and the second image is [0.2, 0.1, 0.1] and [0.3, 0.1, -0.3], respectively. In this case, the difference between the importance is small in the first image and the variation is small, whereas the variation in the importance is large in the second image. That is, while the three determination regions are at substantially the same level in the first image, the second image contains highly characteristic determination regions, so it is preferable to increase the weighting coefficient of the second image. The weighting coefficient is set according to the variation in importance for each determination region, specifically, the difference between the maximum value and the minimum value, the standard deviation, and the variance. If no characteristic is found between the medical images, the global importance may be calculated by a simple average which is a weighted addition with a weighting coefficient of 1.

In the first embodiment, it has been described that the importance of all the determination areas extracted by the determination area extraction unit 202 is displayed. When the number of extracted judgment areas is large, it takes time to confirm the importance of each judgment area. In this embodiment, it will be described that the time required for confirmation can be shortened by enabling the operator to select the determination area and displaying only the importance of the selected determination area. Since the overall configuration of this embodiment is the same as that of the first embodiment, the description thereof will be omitted.

An example of the processing flow of this embodiment will be described with reference to FIG. Since S301 and S302 are the same processes as in the first embodiment, description thereof will be omitted, and S803 and subsequent steps will be described.

(S803)
A desired determination region is selected from the plurality of determination regions extracted in S302. The determination area selection screen 900 as illustrated in FIG. 9 is used for selecting the determination area in this step. The determination area selection screen 900 includes an image display unit 901, a determination area selection unit 902, a diagnosis result display unit 903, and a basis display unit 904. The input medical image is displayed on the image display unit 901.

The determination area selection unit 902 displays buttons that can be selected corresponding to each determination area extracted in S302, such as a button 905 with a size of less than 8 mm, a button 906 with a size of 8 to 15 mm, an aeration button 907, and an edge button 908. Will be done. The operator selects the determination area corresponding to each button by pressing the button displayed on the determination area selection unit 902. FIG. 9 shows a state in which the size less than 8 mm button 905, the size 8 to 15 mm button 906, the aeration button 907, and the edge button 908 are pressed.

The diagnosis result display unit 903 displays, for example, a 98% probability that the lesion site is malignant as a diagnosis result for the input medical image. The rationale display unit 904 displays the importance of the determination area selected in this step. The basis display unit 904 will be described later in S805.

(S804)
The importance calculation unit 203 calculates the importance for each determination area. The calculation of importance is the same as in Example 1.

(S805)
The importance of each determination area calculated in S804 is displayed on the basis display unit 904 in the determination area selection screen 900 of FIG. The importance of each determination area displayed on the basis display unit 904 is limited to the importance of the determination area selected in S803, that is, the determination area corresponding to the button pressed by the determination area selection unit 902. The rationale display unit 904 displays an importance image together with a numerical value of the importance of the selected determination area. The colored region of the importance image is limited to the determination region selected in S803. FIG. 9 illustrates an importance image in which the less than 8 mm region 909 is colored black, the marginal region 912 and the aerated region 911 are shaded, and the 8 to 15 mm region 910 is colored white.

According to the processing flow described above, the operator can select a desired judgment area from the plurality of judgment areas extracted by the judgment area extraction unit 202, and the importance limited to the selected judgment area is displayed. Therefore, the operator can confirm the judgment basis of the diagnosis result in a short time.

In Example 1, it has been described that the importance of the determination area extracted by the determination area extraction unit 202 is displayed. If the extracted determination area does not include the determination area of interest to the operator, the diagnosis result of the diagnostic imaging model may not be accepted by the operator. In this embodiment, it will be described that the diagnostic result can be easily accepted by the operator by enabling the operator to add the judgment area and displaying the importance of the added judgment area. Since the overall configuration of this embodiment is the same as that of the first embodiment, the description thereof will be omitted.

An example of the processing flow of this embodiment will be described with reference to FIG. Since S301 and S302 are the same processes as in the first embodiment, the description thereof will be omitted, and S1003 and subsequent steps will be described.

(S1003)
When the plurality of determination areas extracted in S302 do not include the determination area of interest to the operator, a new determination area is added in this step. For the addition of the determination area in this step, the determination area addition screen 1100 as illustrated in FIG. 11 is used. The determination area addition screen 1100 has an image display unit 1101, a determination area addition unit 1102, a diagnosis result display unit 1103, and a basis display unit 1104. The input medical image is displayed on the image display unit 1101.

The determination area addition unit 1102 includes registered buttons 1105 and 1106, which are buttons indicating that a new determination area has been added and registered, new addition buttons 1107, which are buttons for adding a new determination area, and the like. Is displayed. The addition of the new determination area is started by the operator pressing the new addition button 1107, and the trace of the desired area on the medical image displayed on the image display unit 1101 and the name of the traced area are added. Completed by grant. Region tracing is performed using, for example, a mouse. Area extraction by image processing may be executed in the traced area. The registered button 1105 is a button exemplifying that the traced tumor 1108 is given the name "shape" and is registered, and the registered button 1106 is the traced aeration 1109 with the name "aeration". It is a button exemplifying that it has been given and registered.

The diagnosis result display unit 1103 displays, for example, a 98% probability that the lesion site is malignant as a diagnosis result for the input medical image. The grounds display unit 1104 displays the importance of the determination area added in this step. The basis display unit 1104 will be described later in S1105.

(S1004)
The importance calculation unit 203 calculates the importance for each determination area added in S1003. The calculation of importance is the same as in Example 1.

(S1005)
The importance of each determination area calculated in S1104 is displayed on the basis display unit 1104 in the determination area addition screen 1100 of FIG. The rationale display unit 1104 displays the importance image together with the numerical value of the importance of the added determination area. The colored region of the importance image is limited to the determination region added in S1103. FIG. 11 illustrates an importance image in which the “shape” region 1110 and the “aeration” region 1111 are colored with diagonal lines.

According to the processing flow described above, the operator can add a desired determination area in addition to the plurality of determination areas extracted by the determination area extraction unit 202, and the importance of the added determination area is displayed. As a result, the operator can confirm the judgment basis of the diagnosis result including the judgment area of interest, so that it becomes easier to accept the diagnosis result.

The plurality of examples of the present invention have been described above. The present invention is not limited to these examples, and includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not limited to the one including all the configurations described. In addition, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment. Further, it is possible to add the configuration of another embodiment to the configuration of one embodiment. Further, it is possible to add / delete / replace a part of the configuration of each embodiment with another configuration.

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 Image imaging device, 111: Medical image database, 200: Medical image, 201: Model reading unit, 202: Judgment area extraction unit, 203: Importance calculation unit, 204: Importance, 401: Judgment item, 500: Medical image, 501: Lung tumor, 502: Aeration, 503: Chest wall, 504: Background, 505: Blood vessel, 506: Edge-enhanced image, 507: Split image, 508: Aeration region, 509: Less than 8 mm region, 510: 8 to 15 mm Area, 601: Sampling table, 602: Lung tumor area image, 603: Aeration area image, 604: Chest wall area image, 605: Vascular area image, 606: Diagnosis result, 700: Importance display screen, 701: Diagnosis result display Unit, 702: Importance display unit, 703: Importance image display unit, 704: Color bar, 705: Less than 8 mm area, 706: Marginal area, 707: Aeration area, 708: 8 to 15 mm area, 709: Background Area, 710: Vascular area, 711: Chest wall area, 900: Judgment area selection screen, 901: Image display unit, 902: Judgment area selection unit, 903: Diagnosis result display unit, 904: Groundwork display unit, 905: Size less than 8 mm Button, 906: size 8 to 15 mm button, 907: aerated button, 908: margin button, 909: less than 8 mm area, 910: 8 to 15 mm area, 911: aerated area, 912: margin area, 1100: judgment Area addition screen, 1101: Image display unit 1102: Judgment area addition unit 1103: Diagnosis result display unit 1104: Rationale display unit 1105: Registered button 1106: Registered button 1107: New addition button 1108: Tumor, 1109: Aeration, 1110: "Shape" region, 1111: "Aeration" region

Claims (12)

A model reading unit that reads the diagnostic imaging model that outputs the diagnostic results for the input medical image,
A judgment area extraction unit that extracts a plurality of judgment areas from the medical image based on a plurality of judgment items,
The diagnostic results obtained by inputting the determination area into the diagnostic imaging model are acquired while changing the combination of the determination areas, and each of the determination areas is obtained based on the acquired diagnosis results and the combination of the determination areas. An image diagnosis support device including an importance calculation unit for calculating importance. The diagnostic imaging support device according to claim 1.
The importance calculation unit is an image diagnosis support device characterized in that a combination of the determination regions is set using a sampling matrix and the importance is calculated based on the sampling matrix. The diagnostic imaging support device according to claim 2.
When the sampling matrix is A and the diagnosis result is y, the importance calculation unit sets w indicating the importance.
w = (A ^T A) ^-1 A ^T y
An image diagnosis support device characterized by being calculated by. The diagnostic imaging support device according to claim 2.
The image diagnosis support device, characterized in that the sampling matrix represents all combinations of the determination regions. The diagnostic imaging support device according to claim 1.
The image diagnosis support device, wherein the determination area is an anatomically divided area. The diagnostic imaging support device according to claim 5.
The determination area is an image diagnosis support device characterized by containing aeration. The diagnostic imaging support device according to claim 1.
An image diagnosis support device further comprising a display unit for displaying an importance image generated by coloring each of the determination areas with a color corresponding to the importance. The diagnostic imaging support device according to claim 1.
The importance calculation unit calculates the importance of each of the determination regions for a plurality of medical images, and uses the importance of each of the medical images to determine the global importance, which is the importance of the plurality of medical images. An image diagnosis support device characterized by calculating. The diagnostic imaging support device according to claim 8.
The importance calculation unit calculates the variation in the importance of each medical image for each determination region, and weights and adds the importance for each medical image using the weighting coefficient calculated based on the variation. An image diagnosis support device characterized by calculating the global importance. The diagnostic imaging support device according to claim 1.
A selection unit that selects at least one of a plurality of determination areas,
An image diagnosis support device further comprising a display unit for displaying the importance corresponding to a determination area selected by the selection unit. The diagnostic imaging support device according to claim 1.
An additional part that adds a new judgment area to the plurality of judgment areas,
An image diagnosis support device further comprising a display unit that displays the importance corresponding to the determination area added by the additional unit. A model reading step to read the diagnostic imaging model that outputs the diagnostic result for the input medical image,
A judgment area extraction step for extracting a plurality of judgment areas from the medical image based on a plurality of judgment items, and
The diagnosis result obtained by inputting the determination area into the image diagnosis model is acquired while changing the combination of the determination areas, and each of the determination areas is based on the acquired diagnosis result and the combination of the determination areas. An image processing method characterized by having a computer execute an importance calculation step for calculating importance.

Images