Medical image processing system and processing method

Abstract

A medical image processing system according to the present invention comprises: a communication unit for receiving photographic images captured by a capsule endoscope; and a processing unit for generating a read image from the photographic images by using results of a lesion model trained through lesion classification, wherein the processing unit applies the photographic images to a pre-trained abnormal lesion classification model so as to classify images including a preset lesion, then re-classifies the classified images into lesion images and normal images, selects a representative image representing the lesion images from the classified lesion images, and generates the selected representative image as the read image.

Description

Medical image processing system and processing method

[0001] The present invention relates to a processing system and a processing method for medical images, and more particularly, to a processing system and a processing method for medical images provided from a capsule endoscope.

[0002] Capsule endoscopy is commonly used to diagnose various small bowel diseases. Diagnosing small bowel diseases involves examining lesions based on internal medical images acquired through a capsule endoscope inserted into the body.

[0003] The technology for conventional capsule endoscopes has already been disclosed in Korean Patent Publication No. 10-1969982 (Capsule endoscope device, magnetic controller, and capsule endoscope system, April 11, 2019). This disclosed invention acquires medical images through a capsule endoscope inserted into the human body, enabling diagnosis of lesions based on the medical images.

[0004] In diagnosing these lesions, abnormal lesions are detected in medical images and interpreted as images. Clinicians then diagnose the lesions based on these images. During the lesion diagnosis process, clinicians review a large number of video frames included in the images. Specifically, it is known that specialists can take anywhere from 30 minutes to over two hours to interpret the lesions from these video frames, depending on their skill level. Ultimately, lesion diagnosis requires both the time to generate the images and the time to interpret the lesions. Consequently, the long time required for lesion diagnosis has been a problem.

[0005] An object of the present invention is to provide a medical image processing system and processing method that reduces the time required to generate a reading image and reduces the number of video frames of the reading image provided to a clinician.

[0006] A medical image processing system according to the present invention includes a communication unit that receives a photographed image taken from a capsule endoscope, and a processing unit that generates an interpretation image from the photographed image using a lesion model result learned as a lesion classification method, wherein the processing unit applies the photographed image to a previously learned abnormal lesion classification model to classify an image including a preset lesion, and then reclassifies the classified image into a lesion image and a normal image, selects a representative image representing the lesion images from the classified lesion images, and generates the selected representative images as the interpretation images.

[0007] The above abnormal lesion classification model may include a YOLO (You Only Look Once) series model.

[0008] The above abnormal lesion classification model may include at least one of YOLO v4 and YOLO v8.

[0009] In labeling learning data of the above abnormal lesion classification model, an expert can identify lesions existing in the learning data and perform labeling by applying a preset labeling method to the text for the lesion and the lesions existing in the learning data.

[0010] In the above lesion marking, a square or polygon surrounding the lesion can be marked.

[0011] The above-mentioned captured image includes at least one of a hemorrhagic, inflammatory, vascular, and polyp lesion image, and the above-mentioned abnormal lesion classification model can binary classify the captured image into the lesion image and the normal image regardless of the type of the lesion during the above-mentioned reclassification process.

[0012] The above abnormal lesion classification model can apply a bounding box to the lesion included in the above lesion image.

[0013] The above abnormal lesion classification model statistically calculates a threshold for each lesion including at least one of the hemorrhagic, inflammatory, vascular, and polyp lesions, and when the captured image is provided, the lesion image and the normal image can be binary classified based on the threshold for each lesion.

[0014] In the selection of the above representative images, a pre-learned video frame processing model is applied, and the video frame processing model can extract features of color and texture of the lesion images and compare the similarity between the images.

[0015] In the above similarity analysis, similarity can be analyzed based on the Bhattacharyya Distance.

[0016] In the above similarity analysis, the lesion images for which mutual comparison is performed are converted to HSV, and the number of bins of color and saturation values ​​can be adjusted to produce a histogram for color values ​​(H-Histogram) and a histogram for saturation values ​​(S-Histogram).

[0017] In selecting the representative image, a comparison between a reference image and subsequent images is sequentially performed, and when a subsequent image having a different degree of similarity to the reference image appears, the subsequent image having a different degree of similarity is used as a reference image, and a comparison between the reference image and subsequent images is sequentially performed, and the reference images can be selected as the representative images.

[0018] In selecting the representative images above, the representative images can be selected so that the number of images between a pair of representative images does not exceed n.

[0019] In selecting the representative image above, comparison between the reference image and subsequent images is performed sequentially, subsequent images n or more after the reference image are selected as reference images, and the reference images can be selected as the representative images.

[0020] Meanwhile, a method for processing a medical image according to the present invention includes a step of receiving a photographed image taken from a capsule endoscope, a step of applying the photographed image to a pre-learned abnormal lesion classification model to classify an image including a preset lesion, and a step of reclassifying the classified image into a lesion image and a normal image, a step of selecting a representative image representing the lesion images from the classified lesion images, and a step of generating the selected representative images as an interpretation image.

[0021] The medical image processing system and processing method according to the present invention have the effect of reducing the time for interpretation of medical images, thereby increasing the interpretation efficiency of clinicians, and shortening the waiting time of patients.

[0022] The technical effects of the present invention are not limited to the effects mentioned above, and other technical effects not mentioned will be clearly understood by those skilled in the art from the description below.

[0023] Figure 1 is a schematic diagram showing a medical image processing system according to the present embodiment.

[0024] Figure 2 is a conceptual diagram showing a diagnostic assistance algorithm installed in a medical image processing system according to the present embodiment.

[0025] Figure 3 is a conceptual diagram showing an abnormal lesion classification model of a diagnostic assistance algorithm according to this embodiment.

[0026] Figure 4 is a conceptual diagram showing a labeling method of an abnormal lesion classification model according to the present embodiment.

[0027] Figure 5 is a conceptual diagram showing a learning method of an abnormal lesion classification model according to this embodiment.

[0028] Figure 6 is a conceptual diagram illustrating a similarity evaluation method for extracting a representative frame from a video frame processing model of a diagnostic assistance algorithm according to the present embodiment.

[0029] Figure 7 is a conceptual diagram illustrating the concept of forcibly extracting a representative frame in a video frame processing model of a diagnostic assistance algorithm according to the present embodiment.

[0030] Figure 8 is a conceptual diagram showing an example of extracting a representative frame from a video frame processing model of a diagnostic assistance algorithm according to the present embodiment.

[0031] Figure 9 is a conceptual diagram illustrating an abnormal lesion classification model of a diagnostic assistance algorithm according to another embodiment.

[0032] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. However, the present embodiments are not limited to the embodiments disclosed below and may be implemented in various forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention. The shapes of elements in the drawings may be exaggerated for clearer explanation, and elements indicated by the same reference numerals in the drawings represent the same elements.

[0033] Medical Image Processing System

[0034] Fig. 1 is a schematic diagram showing a medical image processing system according to the present embodiment, and Fig. 2 is a conceptual diagram showing a diagnostic assistance algorithm installed in the medical image processing system according to the present embodiment.

[0035] As illustrated in FIGS. 1 and 2, the medical image processing system (1000, hereinafter referred to as the processing system) according to the present embodiment receives a photographed image of the small intestine provided from a capsule endoscope (10). The processing system (1000) selects a lesion image including an abnormal lesion from the photographed image. The processing system (1000) then reduces the number of selected lesion images to generate an optimal interpretation image.

[0036] For example, the processing system (1000) may include a communication unit (100), a storage unit (200), a processing unit (300), and a display unit (400).

[0037] The communication unit (100) receives the captured image provided from the capsule endoscope (10). At this time, the communication unit (100) can receive the captured image through communication between the capsule endoscope (10) and the human body. However, the communication method of the communication unit (100) can be implemented in various ways and is not limited to human body communication.

[0038] The storage unit (200) stores a diagnostic assistance algorithm (210) that generates a reading image based on a captured image. The diagnostic assistance algorithm (210) may include an abnormal lesion classification model (211) and a video frame processing model (212). The abnormal lesion classification model (211) and the video frame processing model (212) may be deep learning models learned through prior learning.

[0039] The processing unit (300) generates a reading image based on the captured image. The processing unit (300) sequentially applies the captured image to the abnormal lesion classification model (211) and the video frame processing model (212). Accordingly, the processing unit (300) generates a reading image based on the diagnostic assistance algorithm (210).

[0040] The display unit (400) can process the reading image and output the final generated reading image. Accordingly, a clinician can perform a lesion diagnosis based on the reading image output from the display unit (400).

[0041] <Abnormal Lesion Classification Model YOLO v4>

[0042] Fig. 3 is a conceptual diagram showing an abnormal lesion classification model of a diagnostic assistance algorithm according to the present embodiment, and Fig. 4 is a conceptual diagram showing a labeling method of the abnormal lesion classification model according to the present embodiment.

[0043] As illustrated in FIGS. 3 and 4, the abnormal lesion classification model (211) according to the present embodiment binary classifies the captured image. That is, the abnormal lesion classification model (211) classifies the captured image into a lesion image and a normal image. In addition, the abnormal lesion classification model (211) can output a lesion image by applying a bounding box to the area where the lesion is located in the lesion image.

[0044] The abnormal lesion classification model (211) can be applied to various object detection models. For example, the abnormal lesion classification model (211) can be applied to a model based on YOLO (You Only Look Once). Accordingly, when labeling image data applied to the abnormal lesion classification model (211), various labeling methods can be performed depending on the model.

[0045] For example, in image data labeling, an expert identifies lesions present in the image data. The expert can then label the lesions with text.

[0046] For example, in image data labeling, an expert identifies lesions in the image data. The expert then labels the lesions with text and directly marks the lesions in the image data. At this time, the expert can mark the location of the lesion by drawing a rectangle around it.

[0047] For example, in image data labeling, an expert identifies lesions in the image data. The expert then labels the lesions with text and directly marks the lesions in the image data. At this point, the expert can draw polygons based on the lesions in the image data to mark their locations.

[0048] Meanwhile, to facilitate understanding of the invention, an embodiment in which the abnormal lesion classification model (211) is prepared as a YOLO model will be described below. In labeling this YOLO model, an expert may label text and draw a rectangle around the lesion.

[0049] For example, the YOLO model can be implemented as a YOLO v4 model. The YOLO v4 model adopts YOLO v3, CSPDarknet53, SPP, PAN, and BoF ALC BoS architectures. Conventional YOLO models have been vulnerable to detecting small objects. However, the YOLO v4 model utilizes a large input resolution, demonstrating high performance even in the detection of small lesions.

[0050] Accordingly, the abnormal lesion classification model (211) to which the YOLO model is applied classifies images containing preset lesions when a captured image is input, and performs binary classification regardless of the lesion type to classify the images into lesion images and normal images. At this time, the preset lesions may be images containing hemorrhagic, inflammatory, vascular, and polyp lesions.

[0051] Typically, the YOLO model can classify images into four classes, determining whether the images contain hemorrhagic, inflammatory, vascular, or polypoid lesions. However, as the number of classes to be classified increases, the performance of the deep learning model inevitably deteriorates.

[0052] Accordingly, the abnormal lesion classification model (211) does not classify the captured image into four classes when it is input. The abnormal lesion classification model (211) classifies the image including the lesion regardless of the lesion type and binary classifies it into two classes, that is, the lesion image and the normal image. That is, the abnormal lesion classification model (211) classifies the captured image including at least one of hemorrhagic, inflammatory, vascular, and polyp lesions, and reclassifies the classified captured image into the lesion image and the normal image. At this time, the standard for distinguishing the lesion image and the normal image can be determined based on the four thresholds after statistically calculating the thresholds for each of the hemorrhagic, inflammatory, vascular, and polyp lesions.

[0053] <Learning and Evaluation of Abnormal Lesion Classification Model>

[0054] Figure 5 is a conceptual diagram illustrating a learning method of an abnormal lesion classification model according to this embodiment.

[0055] As illustrated in FIG. 5, in learning the abnormal lesion classification model (211) according to the present embodiment, model learning can be performed and verification and performance testing can be performed.

[0056] For example, in the training of the abnormal lesion classification model (211), 5,462,422 small intestine images were extracted from clinical cases of 10,386 patients. The small intestine images were then classified into development data and test data.

[0057] A total of 162,160 images were used for development. This included 109,902 normal images and 52,258 lesion images. The lesion images consisted of 10,000 hemorrhagic lesion images, 28,818 inflammatory lesion images, 9,103 vascular lesion images, and 4,337 polypoid lesion images. The development data was divided into a training dataset and a development / validation dataset. Eighty percent of the development data was used as the training dataset, and 20% as the development / validation dataset.

[0058] The training dataset consisted of 87,922 normal images and 41,807 lesion images. The lesion images consisted of 8,000 hemorrhagic lesion images, 23,055 inflammatory lesion images, 7,282 vascular lesion images, and 3,470 polypoid lesion images. Therefore, a total of 129,729 images were used in training the abnormal lesion classification model.

[0059] The development and validation dataset consisted of 21,980 normal images and 10,451 lesion images. The lesion images consisted of 2,000 hemorrhagic lesion images, 5,763 inflammatory lesion images, 1,821 vascular lesion images, and 867 polypoid lesion images. Therefore, the development and validation dataset, totaling 32,431 images, was used to validate the learning of the abnormal lesion classification model.

[0060] After learning and verification, the abnormal lesion classification model (211) was subjected to performance testing.

[0061] A total of 5,300,262 images were used for the test data. This included 4,829,022 normal images and 471,240 lesion images. The lesion images consisted of 471,240 hemorrhagic lesion images, 150,237 inflammatory lesion images, 17,242 vascular lesion images, and 27,204 polypoid lesion images. The test data consisted of images identical to those used in actual clinical trials.

[0062] Afterwards, the performance evaluation and inference time test of the final learned abnormal lesion classification model (211) were conducted, and it was confirmed that the abnormal lesion classification model (211) to which the YOLO model was applied showed high performance in binary classification of the captured image.

[0063] <Video Frame Processing Model>

[0064] Fig. 6 is a conceptual diagram illustrating a similarity evaluation method for extracting a representative frame from a video frame processing model of a diagnostic assistance algorithm according to the present embodiment, and Fig. 7 is a conceptual diagram illustrating a concept for forcibly extracting a representative frame from a video frame processing model of a diagnostic assistance algorithm according to the present embodiment. Fig. 8 is a conceptual diagram illustrating an example for extracting a representative frame from a video frame processing model of a diagnostic assistance algorithm according to the present embodiment.

[0065] As illustrated in FIGS. 6 to 8, the video frame processing model (212) according to the present embodiment selects a representative image, i.e., a representative frame, of a lesion image provided from the abnormal lesion classification model (211). At this time, the selected representative frame can be output from the display unit (400).

[0066] The video frame processing model (212) compares the similarity of multiple frames of a lesion image. Then, the video frame processing model (212) selects a representative frame representing similar frames. At this time, the video frame processing model (212) extracts features regarding the color and texture of the frames and compares the similarity of the two frames.

[0067] At this time, the video frame processing model (212) analyzes the similarity between two frames through the Bhattacharyya Distance.

[0068] For example, the range of values ​​through the Bhattachaya distance can be 0 to 1. At this time, the video frame processing model (212) can determine the similarity between frames by judging that the closer the value of the Bhattachaya distance is to 0 or 0, the higher the similarity.

[0069] For example, as shown in Fig. 6, it can be assumed that multiple frames are composed of frames 1 through 10. Accordingly, the video frame processing model (212) sequentially compares the reference frame and subsequent frames based on the first frame. At this time, the Bhattachaya distance value from the first frame to the seventh frame may be 0 or close to 0. In addition, the Bhattachaya distance between the first frame and the eighth frame may be 1 or close to 1.

[0070] Accordingly, the video frame processing model (212) determines that frames 1 through 7 are similar frames. Then, the video frame processing model (212) selects a representative frame of the similar frames. Afterwards, the video frame processing model (212) sequentially performs a comparison between the reference frame (the 8th frame) and subsequent frames based on the 8th frame, which is determined to be a different frame from the 1st frame. Then, the video frame processing model (212) selects a representative frame up to the similar frames.

[0071] Meanwhile, the video frame processing model (212) can select a reference frame as the representative frame during the process of selecting a representative frame. In other words, the frame that is the initial similarity comparison target can be selected as the representative frame. For example, referring to FIG. 6, the first frame and the eighth frame can each be selected as representative frames.

[0072] Meanwhile, in calculating the Bhattachaya distance, the video frame processing model (212) performs normalization of the two frames to be compared. In addition, the video frame processing model (212) can calculate the Bhattachaya distance through histogram analysis.

[0073] As an example, the video frame processing model (212) will be described for calculating the Bhattachaya distance for the first frame and the second frame. At this time, the frame may be an RGB image of 320X320 size. Accordingly, the video frame processing model (212) adjusts the size of each image to 1 / 2. Then, the video frame processing model (212) performs HSV conversion on the adjusted size image. That is, the video frame processing model converts red, green, and blue values ​​into hue, saturation, and value values.

[0074] And the video frame processing model (212) adjusts the number of bins of color and saturation values ​​to perform 2D histogram analysis. At this time, the video frame processing model (212) adjusts the number of bins of color values ​​from 180 to 60. And the video frame processing model (212) adjusts the saturation value from 256 to 32. Thereafter, the video frame processing model (212) produces a histogram for color values ​​(H-Histogram) and a histogram for saturation values ​​(S-Histogram).

[0075] And the video frame processing model (212) can calculate the value of the Bhattachaya distance based on the histogram for the color value and the histogram for the saturation value for each of the first frame and the second frame.

[0076] Meanwhile, the video frame processing model (212) prevents excessive skipping of frames when selecting representative frames. When analyzing frame similarity using the Bhattachaya distance, numerous frames between representative frames may be skipped. However, in the medical field, if frames are skipped simply based on similarity analysis, unexpected problems may arise. Therefore, the video frame processing model (212) can forcibly extract representative frames at a preset frame interval N.

[0077] For example, the video frame processing model (212) can select a representative frame so that the number of frames does not exceed 10. Referring to the first frame to the eighteenth frame of FIG. 8, the values ​​through the Bhattachaya distance from the first frame to the twelfth frame may be 0 or close to 0. And from the 13th frame, the value of the Bhattachaya distance may be 1 or close to 1. And the 14th frame may have a Bhattachaya distance value with the 13th frame that is 1 or close to 1. At this time, if the video frame processing model (212) uses only the value of the Bhattachaya distance, the first frame, the 13th frame, and the 14th frame are selected as representative frames.

[0078] However, as described above, the video frame processing model (212) can be set so that the selection interval of the representative frame does not exceed 10 frames. Accordingly, even if the similarity between the first frame and the 12th frame is determined to be high, the video frame processing model (212) selects the first frame, the 11th frame, the 13th frame, and the 14th frame as representative frames. That is, the video frame processing model (212) selects the first frame, which is the initial reference. Then, the video frame processing model (212) selects the 11th frame, which is more than 10 frames from the first frame, as the representative frame, and selects the 13th frame, which has a different similarity from the 11th frame, as the representative frame. Then, the video frame processing model (212) can select the 14th frame, which has a different similarity from the 13th frame, as the representative frame.

[0079] Thereafter, the video frame processing model (212) causes the selected representative frames to be output from the display unit (400). Accordingly, the clinician can shorten the time required for lesion diagnosis through the representative frames.

[0080] <Performance Evaluation Results of the Medical Image Processing System>

[0081] In order to evaluate the performance of the processing system (1000) according to the present invention, a performance evaluation was performed on Equal to or Higher than NVIDIA GeForce RTX 2080 and Window 10. The processing system (1000) can shorten the reading time of a clinician, which previously took 30 minutes to 2 hours or more, to less than 10 minutes. In addition, the speed of generating a reading image was found to take approximately 40 minutes based on 129,600 frames of a clinical case. In addition, the processing system (1000) showed high performance in terms of performance, with a sensitivity of 93.0%, a specificity of 89.0%, and an accuracy of 90.0%. In addition, the processing system (1000) showed a compression ratio of the number of video frames of 80% or more.

[0082] Meanwhile, this embodiment describes that the abnormal lesion classification model is prepared using YOLO v4. However, in another embodiment, the abnormal lesion classification model can be prepared to include YOLO v8.

[0083] <Abnormal Lesion Classification Model YOLO v8>

[0084] Figure 9 is a conceptual diagram illustrating an abnormal lesion classification model of a diagnostic assistance algorithm according to another embodiment.

[0085] As shown in Fig. 9, the YOLO model according to the present embodiment can be prepared as a YOLO v8 model.

[0086] For example, the YOLO v8 model uses a modified CSPDarknet53 backbone. In particular, the CSPLayer used in YOLO v5 has been replaced with the C2f module in the YOLO v8 model. Therefore, the YOLO v8 model has the advantage of accelerating computational speed by having the Partial Pyramid Pooling Fast (SPPF) layer pool image features into a fixed-size map. At this time, each convolution of the YOLO v8 model applies Batch Normalization (BN) and SiLU Activation, and the Head part is divided into Process Objectness, Classification, and Regression tasks, and each task is performed individually.

[0087] In this way, the medical image processing system and processing method according to the present invention have the effect of reducing the time for interpretation of medical images, thereby increasing the interpretation efficiency of clinicians and shortening the waiting time of patients.

[0088] The embodiments of the present invention described above and illustrated in the drawings should not be construed as limiting the technical concept of the present invention. The scope of protection of the present invention is limited only by the matters set forth in the claims, and those skilled in the art will be able to make various improvements and modifications to the technical concept of the present invention. Accordingly, such improvements and modifications, as long as they are obvious to those skilled in the art, will fall within the scope of protection of the present invention.

Claims

1. A communication unit that receives an image captured from a capsule endoscope; and It includes a processing unit that generates a reading image from the above-mentioned photographed image using the lesion model result learned as a lesion classification method, The above processing unit, Applying the above-mentioned captured image to a pre-learned abnormal lesion classification model to classify the image containing the preset lesion, and then reclassifying the classified image into a lesion image and a normal image. Select a representative image representing the lesion images from the above classified lesion images, A medical image processing system characterized in that it generates the selected representative images as the reading images.

2. In paragraph 1, The above abnormal lesion classification model is, A medical image processing system characterized by including a YOLO (You Only Look Once) series model.

3. In paragraph 1, The above abnormal lesion classification model is, A medical image processing system characterized by including at least one model among YOLO v4 and YOLO v8.

4. In paragraph 3, In labeling the learning data of the above abnormal lesion classification model, The expert identifies lesions present in the training data, A medical image processing system characterized in that labeling is performed by applying a preset labeling method to existing lesions in text and learning data for the above lesions.

5. In paragraph 4, In the above lesion sign, A medical image processing system characterized by marking a square or polygon surrounding the above lesion.

6. In paragraph 1, The above filmed video is, Contains at least one of the following images: hemorrhagic, inflammatory, vascular, and polypoid lesions; The above abnormal lesion classification model is, A medical image processing system characterized in that, in the above reclassification process, the captured image is binary classified into the lesion image and the normal image regardless of the type of the lesion.

7. In paragraph 6, The above abnormal lesion classification model is, A medical image processing system characterized by applying a bounding box to a lesion included in the above lesion image.

8. In paragraph 6, The above abnormal lesion classification model is, Statistically calculating a threshold for each lesion including at least one of the above hemorrhagic, inflammatory, vascular and polyp lesions, A medical image processing system characterized in that, when the above-mentioned captured images are provided, the lesion images and the normal images are binary classified based on thresholds for each of the lesions.

9. In paragraph 1, In the selection of the above representative videos, A pre-trained video frame processing model is applied, The above video frame processing model is, A medical image processing system characterized by extracting features of color and texture of the above lesion images and comparing the similarity between the images.

10. In paragraph 9, In the above similarity analysis, A medical image processing system characterized by analyzing similarity based on the Bhattacharyya Distance.

11. In paragraph 9, In the above similarity analysis, The lesion images for which mutual comparison is performed are converted to HSV, A medical image processing system characterized in that it produces a histogram for color values ​​(H-Histogram) and a histogram for saturation values ​​(S-Histogram) by adjusting the number of bins for color and saturation values.

12. In paragraph 1, In selecting the representative video above, Comparison of the reference image and subsequent images is performed sequentially, When a subsequent image with a different similarity to the reference image appears, a comparison between the reference image and the subsequent images is sequentially performed using the subsequent image with a different similarity as the reference image, A medical image processing system characterized by selecting the above reference images as the representative images.

13. In paragraph 12, In selecting the representative video above, A medical image processing system characterized in that the representative images are selected so that the number of images between a pair of representative images does not exceed n.

14. In paragraph 1, In selecting the representative video above, Comparison of the reference image and subsequent images is performed sequentially, Select n subsequent images from the above reference image as reference images, A medical image processing system characterized by selecting the above reference images as the representative images.

15. A step of receiving an image captured from a capsule endoscope; A step of applying the above-mentioned captured image to a pre-learned abnormal lesion classification model to classify an image including a preset lesion, and then reclassifying the classified image into a lesion image and a normal image; A step of selecting a representative image representing the lesion images from the classified lesion images; and A method for processing medical images, comprising a step of generating the selected representative images into reading images.

Images