Diagnostic imaging assistance device, diagnostic imaging assistance method, and diagnostic imaging assistance program

Abstract

Provided is a diagnostic imaging assistance device (10) that assists in determining the invasion depth of a lesion (90) in an endoscopic image. The diagnostic imaging assistance device (10) comprises: a data acquisition unit (12) that acquires three-dimensional data pertaining to a lesion (80) which has been extracted on the basis of an endoscopic image; a curve acquisition unit (14) that acquires, from the three-dimensional data, at least one cross-sectional curve (90L) of the lesion (90) which has a convex shape; a feature extraction unit (15) that extracts at least one prescribed feature amount from the cross-sectional curve (90L); and a computation unit (16) that calculates an invasion depth score of the lesion (90) on the basis of the feature amount.

Description

Imaging Diagnosis Support Device, Imaging Diagnosis Support Method, and Imaging Diagnosis Support Program

[0001] The present invention relates to an imaging diagnosis support device for assisting in determining the depth of a lesion in an endoscopic image, an imaging diagnosis support method for assisting in determining the depth of a lesion in an endoscopic image, and an imaging diagnosis support program for assisting in determining the depth of a lesion in an endoscopic image.

[0002] For malignant tumors without a risk of metastasis, resection using an endoscope, which is a minimally invasive treatment, is performed. Diagnosis of the depth of the tumor is important for determining its applicability.

[0003] WO 2022 / 190366 discloses a direction for accurately measuring the size of a lesion existing on the inner wall surface of a curved gastrointestinal tract using the three-dimensional shape of a biological tissue acquired by an endoscope.

[0004] WO 2022 / 190366[[ID=...]]

[0005] An embodiment of the present invention provides an imaging diagnosis support device for assisting in determining the depth of a lesion using the three-dimensional shape of a biological tissue extracted based on an endoscopic image, an imaging diagnosis support method for assisting in determining the depth of a lesion using the three-dimensional shape of a biological tissue extracted based on an endoscopic image, and an imaging diagnosis support program for assisting in determining the depth of a lesion using the three-dimensional shape of a biological tissue extracted based on an endoscopic image.

[0006] The imaging diagnosis support device according to an embodiment of the present invention includes a data acquisition unit that acquires three-dimensional data of a lesion region extracted based on an endoscopic image, a curve acquisition unit that acquires at least one cross-sectional curve of a convex lesion in a target region from the three-dimensional data, a feature extraction unit that extracts at least one predetermined feature amount from the cross-sectional curve, and a calculation unit that calculates a depth score of the lesion based on the feature amount.

[0007] An embodiment of the present invention provides an image diagnostic support method which involves acquiring three-dimensional data of a lesion region extracted based on an endoscopic image, obtaining at least one cross-sectional curve of a convex-shaped lesion from the three-dimensional data, extracting at least one predetermined feature quantity from the cross-sectional curve, and calculating a depth of invasion score of the lesion based on the feature quantity.

[0008] The image diagnostic support program in the embodiment of the present invention acquires three-dimensional data of a lesion region extracted based on an endoscopic image, obtains at least one cross-sectional curve of a convex-shaped lesion from the three-dimensional data, extracts at least one predetermined feature quantity from the cross-sectional curve, and causes the computer to perform a process of calculating a score for the depth of invasion of the lesion based on the feature quantity.

[0009] According to embodiments of the present invention, it is possible to provide an image diagnostic support device that assists in determining the depth of invasion of a lesion using the three-dimensional shape of biological tissue extracted based on endoscopic images, an image diagnostic support method that assists in determining the depth of invasion of a lesion using the three-dimensional shape of biological tissue extracted based on endoscopic images, and an image diagnostic support program that assists in determining the depth of invasion of a lesion using the three-dimensional shape of biological tissue extracted based on endoscopic images.

[0010] Figure 1 is a diagram showing the configuration of an endoscope system including an image diagnostic support device of the first embodiment. Figure 2 is a flowchart of the image diagnostic support method of the first embodiment. Figure 3 is a 3D data plan view. Figure 4 is a 3D data cross-sectional view. Figure 5 is a 3D data perspective view. Figure 6 is a plan view showing the cutting line for obtaining the cross-sectional curve of a lesion. Figure 7 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 8 is a cross-sectional view of a lesion with deep penetration. Figure 9 is a cross-sectional view of a lesion with shallow penetration. Figure 10 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 11 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 12 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 13 is a plan view showing the cutting line for obtaining the cross-sectional curve of a lesion. Figure 14 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 15 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 16 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 17 is a diagram illustrating the method of extracting features from the cross-sectional curve. Figure 18 is a diagram showing the configuration of an endoscope system including an image diagnostic support device according to the second embodiment. Figure 19 is a flowchart of the image diagnostic support method according to the second embodiment. Figure 20 is a flowchart of the P-type Fourier transform process.

[0011] <First Embodiment> The image diagnostic support device (hereinafter referred to as "support device") 10 of this embodiment shown in Figure 1, together with the endoscope device 2, the image processing device 4, and the monitor 5, constitutes an endoscope system 1.

[0012] The endoscopic device 2 includes an endoscope 3 and an air insufflation device 6. The endoscope 3 acquires video of the wall of the stomach or large intestine, which has been expanded by air insufflation from the air insufflation device 6, for example.

[0013] The image processing device 4 outputs video from the endoscope 3 to the monitor 5. The user performs the examination while viewing the endoscopic image displayed on the monitor 5. When a lesion is displayed on the monitor 5, the user operates the release switch (not shown) on the endoscope 3. The image processing device 4 captures the endoscopic image at the moment the release switch is operated and saves the captured endoscopic image.

[0014] As will be described later, the support device 10 comprises a CPU 11, a data acquisition unit 12, a region setting unit 13, a curve acquisition unit 14, a feature extraction unit 15, and a calculation unit 16. The CPU 11 is a control unit that controls the overall operation of the support device 10.

[0015] At least one of the multiple configurations of the support device 10 may consist of internal circuits of a software-operated semiconductor element, or it may consist of dedicated hardware circuits, or it may include both internal circuits of a semiconductor element and dedicated hardware circuits. All of the multiple configurations of the support device 10 may be functional units of a software-operated processor.

[0016] The support device 10 calculates a lesion depth score P using three-dimensional data of the lesion region extracted based on endoscopic images. In other words, the support device 10 assists the user in determining the depth of lesion invasion.

[0017] <Image Diagnosis Support Method> The image diagnosis support method using the image diagnosis support device 10 will be explained according to the flowchart in Figure 2.

[0018] <Step S10> Data Acquisition The endoscope 3 acquires, for example, a video of the inside of the stomach or large intestine. The image processing device 4 performs image processing, such as noise reduction, on each frame of the video from the endoscope 3 and outputs the endoscopic image to the monitor 5. The user performs the examination while watching the video displayed on the monitor 5. For example, when an image containing a lesion is displayed on the monitor 5, the user operates the release switch on the endoscope 3. The image processing device 4 captures and saves a still image at the moment the release switch is operated.

[0019] The image processing device 4 acquires three-dimensional shape data of the lesion area from the endoscopic image using known methods. For example, if the endoscope 3 is equipped with a stereo camera, the image processing device 4 acquires the three-dimensional shape from images taken by the two cameras using the principle of triangulation. Alternatively, the endoscope 3 may project a measurement pattern using laser light and acquire the three-dimensional shape based on the image of the projected measurement pattern. Alternatively, the three-dimensional shape may be acquired from the features of multiple frames of a video taken with a monocular camera.

[0020] Furthermore, the acquisition of 3D shape data may be performed by a device other than the image processing device 4. For example, the 3D shape data may be acquired by a server 20 that is connected to the endoscope system 1 via an internet connection or the like.

[0021] The data acquisition unit (data acquisition circuit) 12 of the support device 10 acquires three-dimensional data of the lesion region extracted based on the endoscopic image.

[0022] <Step S20> Setting the Area of ​​Interest The area setting unit (area setting circuit) 13 sets the area of ​​interest 80, which includes the convex-shaped lesion 90, in relation to the 3D data, based on the user's operation of the area setting unit 13. The area of ​​interest 80 shown in the 3D data plan view of Figure 3 is rectangular, but the area of ​​interest 80 may also be circular or the like. The area of ​​interest 80 is set to have a margin of a predetermined length or more outside the outer edge of the lesion 90. The area setting unit 13 may also automatically set the area of ​​interest 80 based on the color of the endoscopic image or the 3D data.

[0023] <Step S20> The cross-sectional curve acquisition unit (curve acquisition circuit) 14 acquires the cross-sectional curve 90L of the outer surface of the lesion 90 (see Figure 7, etc.) from the 3D data.

[0024] Furthermore, as shown in the 3D data cross-sectional view in Figure 4, if the shape of the area of ​​interest 80 excluding the lesion is a concave curved surface, it is preferable for the curve acquisition unit 14 to convert the area of ​​interest 80 to a plane (reference plane) as shown in the 3D data perspective view in Figure 5, before acquiring the cross-sectional curve 90L. The same processing can also be performed if the shape of the area of ​​interest 80 excluding the lesion is a convex curved surface.

[0025] The curve acquisition unit acquires the cross-sectional curve 90L shown in Figure 7, along a straight line SL that passes through the vertex T of the convex portion of the lesion 90, for example, as shown in Figure 6. The straight line SL may also pass through the centroid G of the convex portion when the lesion 90 is viewed from above. Preferably, the straight line SL is aligned with the long axis of the lesion 90.

[0026] <Step S40> Feature Extraction The feature extraction unit 15 (feature extraction circuit) 15 extracts predetermined feature quantities from the cross-sectional curve 90L along the straight line SL. In the example shown in Figure 7, the feature quantity is the inclination angle θ (θ1, θ2) of the outer edge of the cross-sectional curve 90L with respect to the reference line. For example, the average value of the two inclination angles θ1 and θ2 is the feature quantity.

[0027] The reason for extracting the inclination angle θ of the outer edge of the cross-sectional curve 90L as a feature is that there is a correlation between the inclination angle θ and the depth of lesion invasion.

[0028] In other words, as shown in Figure 8, in lesions with deep invasion, the tumor lesion 90 grows in an expansive manner, like an inflated balloon, beneath the muscularis mucosa 95. Therefore, the cross-sectional curve 90L of the lesion 90 has a small slope angle θ and is smooth with few irregularities.

[0029] In contrast, in the case of a shallow lesion growing within the muscularis mucosa 95, as shown in Figure 9, the cross-sectional curve 90L of the lesion 90 has a large slope angle θ and relatively large irregularities.

[0030] As a feature, the slope angle θ of the outer edge region of the lesion 90 may be extracted. For example, as shown in Figure 10, the feature extraction unit 15 divides the cross-sectional curve 90L into multiple equal curves. The feature extraction unit 15 calculates the slope of each of the divided curves by performing a linear approximation. The region closest to the outer edge is the outer edge region, with the position where the absolute value of the slope (angle) of the linear approximation line is largest as the boundary, and the region sandwiched between the outer edge regions at both ends is the central region.

[0031] Furthermore, as shown in Figure 11, each of the equally divided curves is linearly approximated to find the angle θ of each line, and the "average absolute value of the first to second," "average absolute value of the first to third," and "average absolute value of the first to Nth" are calculated from the left and right sides, respectively, and when the average absolute value of the angle θave is used as the vertical axis, the result is as shown in Figure 12.

[0032] Instead of using the average angle θave, the point cloud can be sequentially added from the left or right side as described above, and the absolute angle of the first-order approximation line can be calculated each time. The feature extraction unit 15 may also extract features by taking the maximum of the calculated left and right absolute angle values ​​|θ| as angles |θmax1| and |θmax2|. In this case, the left and right positions X1 and X2 where angles |θmax1| and |θmax2| are obtained become the boundary between the outer region and the central region.

[0033] In other words, the feature extraction unit 15 divides the cross-sectional curve 90L into sections, obtains the slopes of multiple dividing lines (multiple differential lines), and then divides it into the outer edge region and the central region based on the position where the absolute value of the slope obtained by averaging the slopes of the dividing lines, either individually or as a multiple value, is maximized. To put it another way, the feature extraction unit 15 divides it into the outer edge region and the central region based on the position where the slope angle of the virtual line connecting the endpoints of the cross-sectional curve and points on the cross-sectional curve is maximized.

[0034] The feature extraction unit 15 may perform differential processing on the cross-sectional curve 90L and extract the maximum value of the absolute value of the differential line as a feature quantity.

[0035] The feature extraction unit 15 may also extract feature quantities using multiple cross-sectional curves 90L along each of the multiple straight lines SL. For example, as shown in Figure 13, nine cross-sectional curves 90L are extracted along nine straight lines SL1-SL9. The straight lines SL1-SL9 are arranged rotationally symmetrically at equal intervals around the centroid G.

[0036] Multiple inclination angles θ are extracted as features from each of the multiple cross-sectional curves 90L1-SL9. For example, the average value of the multiple inclination angles θ is extracted as a feature. Among the multiple cross-sectional curves 90L1-SL9, inclination angles that are excessively large or excessively small compared to the other angles may be excluded from the calculation of the average value.

[0037] The feature is not limited to the inclination angle θ. The feature may also be the aspect ratio (height H / width W) of the cross-sectional curve 90L, as shown in Figure 14. Alternatively, the feature may be the ratio (R2 / R1) of the first radius of curvature R1 to the second radius of curvature R2, as shown in Figure 15. The first radius of curvature R1 is the minimum radius of curvature of the first approximate quadratic curve using data from the left and right outer edge regions of the cross-sectional curve. The second radius of curvature R2 is the minimum radius of curvature of the second approximate quadratic curve using data from the central region of the cross-sectional curve.

[0038] The feature may also be the difference area S shown in Figure 10C. The difference area S is the sum of the difference areas between the second approximate quadratic curve and the cross-sectional curve in the central region. This sum is normalized by dividing it by the length in the direction of the reference plane. The feature may also be the frequency characteristics of the second approximate quadratic curve in the central region, shown in Figure 17. The frequency characteristics indicate the level of unevenness extracted by performing a Fourier transform on the second approximate quadratic curve.

[0039] <Step S50> The calculation unit (calculation circuit) 16 calculates the lesion depth score P based on the feature quantities.

[0040] For example, a relationship between the inclination angle θ and the depth of invasion score P is pre-stored in the calculation unit 16. The calculation unit 16 calculates the depth of invasion score P of the lesion by inputting the inclination angle θ extracted by the feature extraction unit into the relationship. The score P is in the range of 0-1, obtained, for example, by binary logistic regression analysis. The closer the score P is to 0, the less deeply invaded the lesion is, and the closer the score P is to 1, the more deeply invaded the lesion is, indicating deep invasive cancer.

[0041] Incidentally, the support device 10 may have a determination unit (determination circuit) 17 that determines the depth of the lesion. The determination unit 17 determines the depth of the lesion by comparing the depth score P with a predetermined threshold value. For example, when the score P is 0.6 or more, the determination unit 17 determines that the depth is deep; when the score P is 0.4 or less, the determination unit 17 determines that the depth is shallow; and when the score P is more than 0.4 and less than 0.6, the determination unit 17 may determine that the depth is "unknown".

[0042] <Step S60> The display support device 10 displays the depth score and determination result of the lesion together with the endoscopic image on the monitor 5.

[0043] The support device 10 supports the determination of the depth of the lesion 90 using the three-dimensional shape of the biological tissue extracted based on the endoscopic image during the endoscopic examination. Incidentally, the support device 10 may support the determination of the depth of the lesion 90 when the user creates an examination report after the endoscopic examination.

[0044] Incidentally, at least some functional parts of the image processing device 4 and the support device 10 may be provided in a server 20 connected via an Internet line or the like to the image processing device 4 or the support device 10. For example, the image of the endoscope 3 may be transmitted from the image processing device 4 to the server, diagnostic support processing may be performed on the server, and the diagnostic result may be transmitted to the image processing device 4 via the line.

[0045] As described above, the image diagnosis support method of the embodiment acquires three-dimensional data of a lesion area extracted based on an endoscopic image, acquires at least one cross-sectional curve of a convex-shaped lesion from the three-dimensional data, extracts at least one predetermined feature amount from the cross-sectional curve, and calculates the depth score of the lesion based on the feature amount.

[0046] The image diagnosis support program of the embodiment causes a computer to execute a process of acquiring three-dimensional data of a lesion area extracted based on an endoscopic image, acquiring at least one cross-sectional curve of a convex-shaped lesion from the three-dimensional data, extracting at least one predetermined feature amount from the cross-sectional curve, and calculating the depth score of the lesion based on the feature amount.

[0047] A non - temporary computer - readable storage medium 30 may store a program for causing a computer to execute an image diagnosis support program.

[0048] <Modification Example of the First Embodiment> The image diagnosis support devices of the modification examples and other embodiments described below are similar to the image diagnosis support device and have the same effects as the image diagnosis support device. Therefore, components having the same functions as those of the image diagnosis support device are denoted by the same reference numerals, and the description thereof is omitted.

[0049] <Modification Example 1 of the First Embodiment> The feature extraction unit 15 of the image diagnosis support device 10A of this modification example extracts a plurality of feature amounts, and the calculation unit 16 calculates a depth score P using the plurality of feature amounts.

[0050] The plurality of feature amounts are selected from the inclination angle θ of the outer edge of the cross - sectional curve 90L, the aspect ratio (H / W) of the cross - sectional curve 90L, the ratio between the first curvature radius and the second curvature radius, the differential area, and the frequency characteristics of the cross - sectional curve 90L, which have been described above.

[0051] For example, the calculation unit 16 calculates the depth score P of the lesion 90 using the inclination angle θ and the aspect ratio (H / W). A calculation formula for calculating the depth score P with the inclination angle θ and the aspect ratio as variables is stored in the calculation unit 16. The calculation unit 16 may calculate the depth score P using multivariate analysis, for example, binary logistic regression analysis, of three or more feature amounts.

[0052] The image diagnosis support device 10A can calculate a depth score P with higher accuracy than the image diagnosis support device 10.

[0053] <Second Embodiment> Since the support device 10 of the first embodiment calculates the depth score P based on partial features of the cross - sectional curve 90L, there is a possibility of misjudgment.

[0054] As shown in FIG. 18, the support device 10B of this embodiment includes a screening unit 18 that performs screening processing. As shown in the flowchart of FIG. 19, step S35 (screening) is performed after step S30 (cross - sectional curve acquisition) and before step S40 (feature amount extraction).

[0055] For example, the screening unit 18 performs screening using the P-type Fourier operation. According to the P-type Fourier operation, lesions 90 can be selected based on the overall shape of the cross-sectional curve 90L, which is an open curve.

[0056] The P-type Fourier operation will be briefly explained following the flowchart in Figure 20.

[0057] <Step S110> The section curve data acquired in step S30 is input to the calculation unit 16.

[0058] <Step S120> Sectional curve division Similar to the feature extraction unit 15 already explained, the sectional curve 90L is divided into multiple equal curves.

[0059] <Step S130> The multiple straight lines connecting the start and end points of the multiple curves are transformed into a complex coordinate system.

[0060] <Step S140> Multiple linear data points, converted to complex numbers through discrete Fourier transform, are subjected to a discrete Fourier transform to calculate a P-type Fourier descriptor.

[0061] <Step S150> SVM Processing: Prior to this, principal component analysis is performed using cross-sectional curves of deep-invasion lesions and non-deep-invasion lesions as training data, and weight coefficients corresponding to multiple principal components are obtained. The P-type Fourier descriptor calculated in step S140 is processed by a Support Vector Machine (SVM) and screening is performed.

[0062] For example, it is preferable to use, for instance, eight principal components out of the sixteen principal components that have high shape explanatory power. The meaning of each principal component can be understood by performing a known inverse transform operation. Therefore, a screening process that ensures explainability is performed by using the P-type Fourier operation.

[0063] Furthermore, the screening method is not limited to the P-type Fourier operation. For example, lesions that the user judges to be clearly shallow in depth may be excluded. Alternatively, screening may be performed using the method shown in the modification of the third embodiment described later.

[0064] According to the support device 10B of this embodiment, there is less risk of misdiagnosis compared to the support device 10, etc.

[0065] <Modification of the second embodiment> In the support device 10B, the P-type Fourier operation was used for the screening process. However, the depth of invasion score of the lesion 90 may be calculated by the P-type Fourier operation, or the depth of invasion may be directly determined using machine learning such as SVM.

[0066] In other words, the support device 10C of this modified example comprises a data acquisition unit 12 that acquires three-dimensional data of a lesion 90 extracted based on an endoscopic image, a curve acquisition unit 14 that acquires at least one cross-sectional curve 90L of a convex-shaped lesion 90 from the three-dimensional data, and a calculation unit 16 that performs a P-type Fourier calculation on the cross-sectional curve 90L to calculate a depth score or determine the depth of invasion of the lesion 90.

[0067] <Third Embodiment> The image diagnostic support device 10D of this embodiment assists in determining the depth of invasion of lesions formed on the walls of internal spaces, for example, tumors on the inner surface of the stomach or large intestine.

[0068] Because the space inside the body expands due to the air supply from the air supply device 6, for example, the curvature of the concave surface of the region of interest 80 shown in Figure 4 increases.

[0069] Lesions that penetrate deeply, such as cancer, are often hardened. Therefore, even if the internal shape changes, the change in the shape of the lesion is small.

[0070] The calculation unit 16 determines the depth of the lesion based on the characteristic quantities of the lesion 90 associated with the change in the spatial expansion state. The characteristic quantities are the inclination angle θ of the cross-sectional curve 90L, the aspect ratio (height H / width W), etc., as previously explained.

[0071] The calculation unit 16 pre-stores, for example, a relationship between the change in the slope angle θ before and after spatial expansion (Δθ) and the depth score P. It goes without saying that the calculation unit 16 may also calculate the depth score P using changes in multiple features.

[0072] Note that "before spatial expansion" does not mean "before air is supplied." For example, one could start with a state of expansion due to air supply for observation, then slightly degas the space, changing the expanded state while acquiring changes in the feature quantities.

[0073] <Modification of the Third Embodiment> In this modified embodiment, the image diagnostic support device 10E uses the changes in the characteristic quantities of the lesion 90 due to changes in the spatial expansion state caused by air insufflation / degassing in the screening process (S35) shown in Figure 19. Alternatively, in the screening process, a P-type Fourier descriptor may be used to analyze the difference in characteristic quantities before and after air insufflation, and a depth score P may be calculated based on the magnitude of the difference.

[0074] The range of values ​​described above is not limited to that range and can be increased or decreased as appropriate. Furthermore, the present invention is not limited to the embodiments described above, and various changes and modifications can be made as long as they do not alter the essence of the present invention.

[0075] This application is filed on the basis of a priority claim to Japanese Patent Application No. 2024-160421, filed in Japan on September 17, 2024, and the above disclosures are incorporated by reference in the specification and claims of this application.

[0076] 1... Endoscopy system 2... Endoscopy device 3... Endoscope 4... Image processing device 5... Monitor 6... Air insufflation device 10, 10A-10E... Image diagnostic support device (support device) 11... CPU 12... Data acquisition unit 13... Region setting unit 14... Curve acquisition unit 15... Feature extraction unit 16... Calculation unit 17... Judgment unit 18... Screening unit 20... Server 30... Storage medium 80... Area of ​​interest 90... Lesion 95... Muscularis mucosa

Claims

A data acquisition unit that acquires three-dimensional data of the lesion region extracted based on endoscopic images, A curve acquisition unit that acquires at least one cross-sectional curve of a convex-shaped lesion in the region of interest from the aforementioned three-dimensional data, A feature extraction unit that extracts at least one predetermined feature quantity from the cross-sectional curve, An image diagnostic support device comprising: a calculation unit that calculates a depth of invasion score of the lesion based on the aforementioned feature quantities. 　 The image diagnostic support apparatus according to claim 1, characterized in that the curve acquisition unit acquires the cross-sectional curve that passes through the vertex of the convex portion within the region of interest, or the centroid of the convex portion within the region of interest. 　 The image diagnostic support apparatus according to claim 1, characterized in that the curve acquisition unit converts the region of interest into a plane before acquiring the cross-sectional curve. 　 The image diagnostic support device according to claim 1, characterized in that the feature quantities are the inclination angle of the cross-sectional curve, the aspect ratio of the cross-sectional curve, the ratio of the first radius of curvature of a first approximate quadratic curve using data from the outer edge region of the cross-sectional curve to the second radius of curvature of a second approximate quadratic curve using data from the central region, the difference area between the second approximate quadratic curve and the cross-sectional curve, and the frequency characteristics of the cross-sectional curve in the central region. 　 The image diagnostic support apparatus according to claim 4, characterized in that the feature extraction unit divides the image into the outer edge region and the central region based on the position where the inclination angle of the virtual line connecting the endpoint of the cross-sectional curve and a point on the cross-sectional curve is maximum. 　 The image diagnostic support apparatus according to claim 4, characterized in that the feature extraction unit divides the cross-sectional curve into sections, obtains the slopes of multiple dividing lines, and then divides the area into the outer edge region and the central region based on the position where the absolute value of the slope obtained by averaging the slopes of the dividing lines, either individually or as a multiple method, is maximized. 　 The image diagnostic support apparatus according to claim 1, characterized in that the calculation unit uses a plurality of feature quantities. 　 The curve acquisition unit acquires multiple cross-sectional curves, The image diagnostic support apparatus according to claim 1, characterized in that the calculation unit performs calculations using at least one of the feature quantities of each of the plurality of cross-sectional curves. 　 The aforementioned penetration depth score is compared with a predetermined threshold, The image diagnostic support device according to claim 1, further comprising a determination unit for determining the depth of invasion of the lesion. 　 The image diagnostic support apparatus according to claim 1, further comprising a screening unit that performs a screening process before extraction by the feature extraction unit. 　 The image diagnostic support apparatus according to claim 10, characterized in that the screening unit performs the screening process using a P-type Fourier operation. 　 The aforementioned lesion is formed on the wall surface of the space inside the body, The aforementioned space is expanded by air supply, The image diagnostic support device according to claim 1, wherein the calculation unit comprises calculating the depth of invasion score based on the change in the characteristic quantity of the lesion accompanying the change in the expansion state of the space. 　 We obtain three-dimensional data of the lesion area extracted based on endoscopic images. From the aforementioned three-dimensional data, obtain at least one cross-sectional curve of the convex-shaped lesion. At least one predetermined feature quantity is extracted from the cross-sectional curve, A method for supporting image diagnosis, characterized by calculating a depth of invasion score for the lesion based on the aforementioned features. 　 We obtain three-dimensional data of the lesion area extracted based on endoscopic images. From the aforementioned three-dimensional data, obtain at least one cross-sectional curve of the convex-shaped lesion. At least one predetermined feature quantity is extracted from the cross-sectional curve, An image diagnostic support program characterized by causing a computer to perform a process of calculating a depth of invasion score of the lesion based on the aforementioned features.

Images