Method for detecting lesion in endoscopic image and method and computing device for training artificial neural network model to perform the same

Abstract

To provide a method of training an artificial neural network model for detecting lesions in endoscopic images that is performed by a computing device including at least one processor.SOLUTION: A method includes: generating training data including labels for lesions on the basis of endoscopic images with characteristics of regions where the endoscopic images are captured and characteristics of the lesions; and training an artificial neural network model to detect the lesions in endoscopic images on the basis of the training data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a technique for training an artificial neural network model to detect lesions in endoscopic images and using the same to detect lesions in endoscopic images.

Background Art

[0002] An endoscope is a general term for a medical instrument that observes organs by inserting a scope into the body without performing surgery or autopsy. An endoscope inserts a scope into the human body, irradiates light, and visualizes the light reflected from the surface of the inner wall. The types of endoscopes are classified according to the purpose and body part, and can be roughly classified into a rigid endoscope in which the endoscope tube is formed of metal and a flexible endoscope typified by a gastrointestinal endoscope.

[0003] The inside of the digestive tract into which a flexible endoscope is inserted is very soft tissue and has an irregular shape. In addition, since the shape of the inside of the digestive tract varies from patient to patient, even experienced medical staff may not find the process of inserting the endoscope easy. In such a situation, medical staff have to concentrate on safely inserting the endoscope and searching for lesions, so when performing endoscopic procedures repeatedly, the fatigue level of medical staff increases significantly.

[0004] Therefore, for the convenience of medical staff, techniques for searching for lesions in endoscopic images have been studied. In particular, object recognition models that detect various features in images have also been actively applied to the field of endoscopic images. The object recognition artificial neural network model is trained to perform a region proposal operation that quickly searches for areas where an object may be present and a classification operation that classifies which object a specific object is.

[0005] On the one hand, due to the characteristics of the environment for photographing the inside of the narrow tubular digestive tract, endoscopic images contain various types of noise and the form of the image changes significantly depending on the movement of the scope. In addition, since the forms of lesions to be recognized are very diverse, it is difficult to improve the sensitivity and accuracy of the artificial neural network model for detecting them.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0007] The present disclosure was devised in response to the aforementioned background art, and relates to a method for detecting lesions in endoscopic images with learning data for an artificial neural network model considering the characteristics of the site where the endoscopic image is taken and the characteristics of the lesions, a method for training an artificial neural network model to perform the same, and a computing device.

[0008] However, the problems to be solved by the present disclosure are not limited to the problems mentioned above, and other problems not mentioned will be clearly understandable from the following description.

Means for Solving the Problems

[0009] According to an embodiment of the present disclosure for realizing the problems as described above, a method for training an artificial neural network model for detecting lesions in endoscopic images performed by a computing device is disclosed. The method includes generating learning data including labels for the lesions based on the endoscopic images according to the characteristics of the site where the endoscopic images are taken and the characteristics of the lesions, and training an artificial neural network model to detect the lesions in the endoscopic images based on the learning data.

[0010] As an alternative, the learning data may include lesion learning data including lesions and normal learning data not including lesions, and the ratio may be determined according to the clinical characteristics of the lesions.

[0011] As an alternative, the clinical characteristics of the lesions may include the frequency of occurrence of the lesions or the types of the lesions.

[0012] As an alternative, the lesion learning data may be generated by an enhancement technique that maintains the imaging characteristics of the lesions.

[0013] As an alternative, the learning data may include learning data, verification data, and evaluation data, and the learning data, the verification data, and the evaluation data may be configured based on the patient information from which the learning data is obtained.

[0014] As an alternative, the step of training the artificial neural network model may include a step of assigning weights so as to determine the inference result of the artificial neural network model reflecting the clinical characteristics of the lesions.

[0015] As an alternative, the step of assigning weights may include a step of assigning high weights to learning data including lesions with high reading difficulty.

[0016] As an alternative, the type of the loss function of the artificial neural network model may be determined according to the clinical characteristics of the lesions or the imaging characteristics of the lesions.

[0017] As an alternative, the loss function of the artificial neural network model may include a loss function of the Distance IoU (DIoU) structure.

[0018] As an alternative, the artificial neural network model may be trained to detect an image without lesions in endoscopic images based on the learning data.

[0019] As an alternative, the loss function of the artificial neural network model can be determined based on the loss function for detecting the lesion and the loss function for detecting a video that does not include the lesion.

[0020] Disclosed is a method for detecting a lesion in an endoscopic video, which is performed by a computing device including at least one processor according to an embodiment of the present disclosure for realizing the above-described problems. The method includes detecting, in an endoscopic video, a lesion or a video not including a lesion, using an artificial neural network model learned with learning data generated based on at least one of characteristics of a site where the endoscopic video is taken, clinical characteristics of the lesion, and imaging characteristics of the lesion.

[0021] Disclosed is a computing device for training an artificial neural network model for detecting a lesion in an endoscopic video according to an embodiment of the present disclosure for realizing the above-described problems. The device includes a processor including at least one core, and a memory including program code executable by the processor. The processor generates learning data including a label for the lesion based on the endoscopic video according to characteristics of a site where the endoscopic video is taken and characteristics of the lesion, and trains an artificial neural network model to detect a lesion in the endoscopic video based on the learning data.

Advantages of the Invention

[0022] By providing learning data for an artificial neural network model that detects a lesion in consideration of clinical characteristics of the lesion, the present disclosure can achieve a high level of sensitivity and accuracy required in the medical field, as compared with an artificial neural network model learned simply based on morphological characteristics.

[0023] In addition, the present disclosure can improve the inference accuracy by detecting endoscopic images without lesions in consideration of the actual endoscopic procedure situation where the rate of obtaining endoscopic images without lesions is high.

Brief Description of the Drawings

[0024]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Modes for Carrying Out the Invention

[0025] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those having ordinary knowledge in the technical field of the present disclosure (hereinafter referred to as those skilled in the art) can easily implement them. The embodiments presented in the present disclosure are provided so that those skilled in the art can use or implement the content of the present disclosure. Therefore, various modifications to the embodiments of the present disclosure will be apparent to those skilled in the art. That is, the present disclosure can be embodied in various different forms and is not limited to the following embodiments.

[0026] Throughout the specification of the present disclosure, the same or similar reference numerals refer to the same or similar components. Also, for the purpose of clearly explaining the present disclosure, the reference numerals of the parts not related to the description of the present disclosure can be omitted from the drawings.

[0027] The term "or" as used in this disclosure is intended to mean inclusive "or" rather than exclusive "or". That is, in this disclosure, unless otherwise specified or the meaning is not clear from the context, "x uses a or b" should be understood to mean one of the natural inclusive substitutions. For example, in this disclosure, unless otherwise specified or the meaning is not clear from the context, "x uses a or b" can be interpreted as either x uses a, x uses b, or x uses both a and b.

[0028] The term "at least one of A or B" as used in this disclosure should be interpreted to indicate all of A, B, and the combination of A and B.

[0029] The term "and / or" as used in this disclosure should be understood to include while indicating all possible combinations of one or more of the related concepts listed.

[0030] The terms "comprising" and / or "including" as used in this disclosure should be understood to mean that a particular feature and / or component is present. However, the terms "comprising" and / or "including" should be understood not to exclude the presence or addition of one or more other features, other components, and / or combinations thereof.

[0031] In this disclosure, unless otherwise specified or indicating a singular form and the context is not clear, the singular should generally be interpreted to include "one or more".

[0032] The term "the Nth (N is a natural number)" used in the present disclosure can be understood as an expression used to distinguish the components of the present disclosure from each other according to a predetermined criterion such as a functional perspective, a structural perspective, or convenience of description. For example, in the present disclosure, components that perform different functional roles can be distinguished as the first component or the second component. However, components that are substantially the same within the technical idea of the present disclosure but must be divided for convenience of description can also be distinguished as the first component or the second component.

[0033] The term "model" used in the present disclosure can be understood as a system embodied using mathematical concepts and language to solve a specific problem, a set of software units for solving a specific problem, or an abstract model of a processing process for solving a specific problem. For example, a neural network "model" can represent an entire system embodied as a neural network having problem-solving ability through learning. Here, a neural network can have problem-solving ability by optimizing parameters that connect nodes or neurons through learning. A neural network "model" can include a single neural network or a set of neural networks combined with a plurality of neural networks.

[0034] The above explanations of the terms are for helping the understanding of the present disclosure. Therefore, it should be noted that when the above terms are not explicitly described as matters limiting the content of the present disclosure, the content of the present disclosure is not used in the sense of limiting the technical idea.

[0035] FIG. 1 is a block configuration diagram of a computing device according to an embodiment of the present disclosure.

[0036] According to an embodiment of the present disclosure, the computing device 100 may be a hardware device or a part of a hardware device that performs comprehensive data processing and operations, and may also be a software-based computing environment connected via a communication network. For example, the computing device 100 may perform functions of intensive data processing, and may be a server that is a subject sharing resources, or may be a client that shares resources through interaction with the server. Further, the computing device 100 may be a cloud system that enables a plurality of servers and clients to interact with each other to comprehensively process data. The above description is only an example related to the type of the computing device 100, and the type of the computing device 100 can be configured in various ways within the scope understandable by those skilled in the art based on the content of the present disclosure.

[0037] The computing device 100 can be wirelessly connected to an endoscope device that acquires various information including medical images inside the body. That is, the computing device 100 can receive necessary information for performing operations described later from the endoscope device via, for example, a network unit, and provide the generated information to the endoscope device. Exemplarily, the computing device 100 can receive endoscope images for generating learning data from the endoscope device. Further, the computing device 100 can receive input data for inference of a learned artificial neural network model from the endoscope device and provide the inference result to the endoscope device. On the other hand, the computing device 100 can also perform the above-described operations for a server in a hospital including a number of endoscope devices. On the other hand, the computing device 100 can be implemented depending on the internal configuration of the endoscope device. In this case, the computing device 100 can play a role corresponding to the control unit of the endoscope device.

[0038] Referring to FIG. 1, a computing device 100 according to an embodiment of the present disclosure may include a processor 110, a memory 120, and a network unit 130. However, since FIG. 1 is merely an example, the computing device 100 may include other configurations for implementing a computer environment. Also, only some of the above-disclosed configurations may be included in the computing device 100.

[0039] The processor 110 according to an embodiment of the present disclosure can be understood as a component unit including hardware and / or software for performing computing operations. For example, the processor 110 can read a computer program and execute data processing for machine learning. The processor 110 can process operation processes such as processing of input data for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. The processor 110 for performing such data processing can include a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application specific integrated circuit (ASIC), or a field programmable gate array (FPGA), etc. Since the types of the processor 110 described above are merely examples, the types of the processor 110 can be variously configured within the scope understandable by those skilled in the art based on the content of the present disclosure.

[0040] The processor 110 can execute a series of operations to train an artificial neural network model for detecting lesions included in endoscopic images. Here, the processor 110 can generate training data reflecting the clinical characteristics of the endoscopic images. Specifically, the processor 110 can generate training data including labels for the lesions based on the endoscopic images according to the characteristics of the site where the endoscopic images are taken and the characteristics of the lesions. And endoscopic images without lesions can be included in the training data. Then, the processor 110 can train the artificial neural network model to detect lesions in the endoscopic images based on the training data.

[0041] Therefore, by providing training data considering the type, occurrence frequency, risk level, morphology, etc. of the lesions, the present disclosure can achieve a high level of sensitivity and accuracy required in the medical field compared to artificial neural network models trained simply based on morphological characteristics.

[0042] In addition to simply detecting lesions using static steel images, to ensure the speed and reliability to the extent that it can be used in the actual endoscopic operation process, the processor 110 can design and train the structure of the artificial neural network model. Specifically, the processor 110 can train the artificial neural network model to execute operations for detecting lesions and operations for detecting images without lesions. In the actual endoscopic operation situation, since the proportion of endoscopic images without lesions is high, the inference accuracy can be improved by detecting not only endoscopic images with lesions but also endoscopic images without lesions.

[0043] As described above, the processor 110 can use the trained artificial neural network model to detect lesions or images without lesions in the endoscopic images.

[0044] The memory 120 according to an embodiment of the present disclosure can be understood as a component unit including hardware and / or software for storing and managing data processed by the computing device 100. That is, the memory 120 can store any form of data generated or determined by the processor 110 and any form of data received by the network unit 130. For example, the memory 120 can include at least one type of storage medium such as a flash memory type, a hard disk type, a multimedia card micro type, a card type memory, a RAM (random access memory), an SRAM (static random access memory), a ROM (read-only memory), an EEPROM (electrically erasable programmable read-only memory), a PROM (programmable read-only memory), a magnetic memory, a magnetic disk, and an optical disk. Also, the memory 120 can include a database system for controlling and managing data in a predetermined system. Since the types of the memory 120 described above are merely examples, the types of the memory 120 can be configured in various ways within the scope understandable by those skilled in the art based on the content of the present disclosure.

[0045] The memory 120 can include endoscopic images and learning data generated based on the endoscopic images, and program code executable by the processor 110.

[0046] The network unit 130 according to an embodiment of the present disclosure can be understood as a component that transmits and receives data via any known wired or wireless communication system. For example, the network unit 130 can perform data transmission and reception using a wired or wireless communication system such as a local area network (LAN), wideband code division multiple access (WCDMA), long term evolution (LTE), wireless broadband internet (WiBro), fifth generation mobile communication (5G), ultrawide-band wireless communication, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity (Wi-Fi), near field communication (NFC), or Bluetooth (registered trademark). Since the above-described communication systems are merely examples, the wired or wireless communication system for data transmission and reception of the network unit 130 can be variously applied in addition to the above-described examples.

[0047] For example, the network unit 130 can receive an endoscopic image from an endoscopic device and provide the detection result of the artificial neural network model to the endoscopic device.

[0048] The data to be processed by the processor 110 can be stored in the memory 120 or received via the network unit 130, and the data generated by the processor 110 can be stored in the memory 120 or transmitted externally via the network unit 130.

[0049] FIG. 2 is a block configuration diagram of an artificial neural network model using learning data according to an embodiment of the present disclosure, FIG. 3 is a flowchart showing a method for a computing device to train an artificial neural network model according to an embodiment of the present disclosure, and FIG. 4 is an exemplary diagram showing an endoscopic image used as learning data according to an embodiment of the present disclosure.

[0050] Referring to FIG. 1 and FIGS. 2 to 4 together, the computing device 100 can generate learning data 220 based on the endoscopic image 210 in order to train the artificial neural network model 200. Then, the trained artificial neural network model 200 can detect a lesion or an image without a lesion from the endoscopic image 210.

[0051] The computing device 100 can generate learning data 220 including labels for lesions based on the endoscopic image 210 based on the characteristics of the site where the endoscopic image 210 is taken and the characteristics of the lesions (S110). Specifically, the computing device 100 can generate the learning data 220 so as to include lesion learning data including lesions and normal learning data without lesions at a specific ratio. Here, the ratio can be determined by the clinical characteristics of the lesions. The clinical characteristics of the lesions can include the frequency of occurrence of the lesions or the types of the lesions.

[0052] For example, the computing device 100 can determine a ratio in consideration of the characteristics of the body part to be the subject of the endoscopic procedure. Taking a gastric endoscope as an example, the ratio between the lesion learning data and the normal learning data can be determined based on the types and occurrence ratios of the lesions occurring in the stomach. Then, for the lesion learning data, the ratio of the endoscopic images 210 including lesions can be determined based on the morbidity rate according to the types of lesions. Here, the determined ratio may not match the number of the lesion learning data and the normal learning data. For example, the number of the lesion learning data may be insufficient. The computing device 100 can use a data augmentation technique to generate the insufficient learning data 220. Here, the computing device 100 can use an augmentation technique that maintains the image characteristics of the lesions. For example, the computing device 100 can further generate the endoscopic images 210 by using image processing methods such as Rotation, Flipping, Crop, and Mosaic on the endoscopic images 210. And an augmentation technique that changes the color sense of the endoscopic images 210, such as the hue, saturation, lightness, brightness, and luminance of the endoscopic images 210, or the characteristics generated by the light source, may not be used.

[0053] The computing device 100 can include learning data, verification data, and evaluation data based on the endoscopic images 210 for the learning and verification of the artificial neural network model 200. That is, the learning data 220 can include the learning data, the verification data, and the evaluation data. Here, the computing device 100 can configure the learning data, the verification data, and the evaluation data based on the patient information from which the learning data 220 is obtained.

[0054] For example, when the patient information is the same, that is, the endoscopic images 210 generated from the same patient can be configured to belong to any one of the groups of learning data, verification data, and evaluation data. Since the endoscopic images 210 obtained from the same patient can have similar image characteristics or clinical characteristics, when learning, verification, and evaluation are performed using similar images, the accuracy of inference of the artificial neural network model 200 may be higher than intended. To prevent this, the data of the same patient can be included in a single group.

[0055] The computing device 100 can train the artificial neural network model 200 to detect lesions in the endoscopic image 210 based on the training data 220 (S120).

[0056] The artificial neural network model 200 can include at least one neural network. The neural network can include, but is not limited to, network models such as DNN (Deep Neural Network), RNN (Recurrent Neural Network), BRDNN (Bidirectional Recurrent Deep Neural Network), MLP (Multilayer Perceptron), CNN (Convolutional Neural Network), and transformer.

[0057] The computing device 100 can train the artificial neural network model 200 through supervised learning using the training data 220 as input values. Alternatively, the artificial neural network model 200 can be trained through unsupervised learning by self-learning the types of data required for data recognition without any teacher, so as to find the criteria for data recognition. Alternatively, the artificial neural network model 200 can be trained through reinforcement learning that uses feedback on whether the result of data recognition by learning is correct.

[0058] Exemplarily, the artificial neural network model 200 of the present disclosure can be composed of an architecture based on a one-layer perception algorithm. The artificial neural network model 200 can include a backbone that extracts and classifies features from the input training data 220, and a detector that senses objects. The detector includes a loss function for predicting the coordinates of the bounding box. Here, the computing device 100 can specifically determine the type of the loss function of the artificial neural network model 200 in order to improve the accuracy of lesion detection, that is, to improve the accuracy of the bounding box. That is, the computing device 100 can determine the loss function of the artificial neural network model 200 according to the clinical characteristics of the lesion or the image characteristics of the lesion.

[0059] Exemplarily, the types of loss functions include the Generalized IoU (GIoU) loss function that considers the distance between two objects while maintaining an attribute with a constant scale based on the IoU (Intersection over Union) between overlapping objects, the Distance IoU (DIoU) loss function that adds a term for the penalty corresponding to the center point distance between two objects to the GIoU loss function to improve the convergence speed, and the Complete IoU (CIoU) loss function that adds a penalty term considering the aspect ratio between two objects to the DIoU loss function so as to enable faster convergence when the objects do not overlap.

[0060] The artificial neural network model 200 of the present disclosure can use a loss function with a DIoU structure as the loss function. In the case of lesion learning data, the bounding box including the lesion can include a plurality of lesions. Alternatively, depending on the direction of the scope for acquiring the endoscopic image 210, the direction of the image capturing the lesion may change, and the aspect ratio of the bounding box may change. This is because the feature of the aspect ratio of the bounding box is less likely to be a factor improving the accuracy of lesion detection.

[0061] The computing device 100 can train the artificial neural network model 200 to detect an endoscopic image 210 without a lesion based on the learning data 220. In actual endoscopic surgery, the ratio of detecting an endoscopic image 210 without a lesion may be higher than the ratio of detecting an endoscopic image 210 with a lesion. Therefore, it is important to ensure not only the accuracy of detecting an endoscopic image 210 with a lesion but also the accuracy of detecting an endoscopic image 210 without a lesion for the artificial neural network model 200.

[0062] For this purpose, the computing device 100 can use, as learning data 220, an endoscopic image 210 that does not contain a lesion and has no bounding box as shown in FIG. 4(a). Further, the computing device 100 can design a loss function based on the loss function for detecting a lesion and the loss function for detecting an image without a lesion, such that the artificial neural network model 200 detects a lesion or detects an image without a lesion, and use this for learning.

[0063] The computing device 100 can learn the artificial neural network model 200 by assigning weight values so as to determine the inference result of the artificial neural network model 200 reflecting the clinical characteristics of the lesion. For example, the computing device 100 can assign a high weight value to the learning data 220 including a lesion with high reading difficulty. The reading difficulty can be determined by the image characteristics indicating the occurrence frequency, form, color, shape, size, etc. of the lesion and the information of the patient from whom the endoscopic image 210 was obtained.

[0064] Exemplarily, referring to FIG. 4, FIGS. 4(b) and 4(c) are endoscopic images 210 including a pedunculated polyp. On the other hand, FIG. 4(d) is an endoscopic image 210 including a sessile polyp and a flat polyp. Since the flat polyp has a flat form, there is a high possibility that the endoscopic operator misses the detection. To prevent this, the artificial neural network model 200 must detect the lesion with higher sensitivity than a lesion with low reading difficulty. Therefore, the computing device 100 can use a separate weight value for a lesion with high reading difficulty.

[0065] FIG. 5 is a flowchart showing the operation of the artificial neural network model according to an embodiment of the present disclosure.

[0066] Referring to FIGS. 2 and 5, the computing device 100 can detect a lesion or an image without a lesion in the endoscopic image 210 by using an artificial neural network model 200 trained with learning data 220 generated based on at least one of the characteristics of the site where the endoscopic image 210 is taken, the clinical characteristics of the lesion, and the imaging characteristics of the lesion (S210). The computing device 100 can receive the endoscopic image 210 from the endoscopic device, infer the presence or absence of a lesion therein, and provide the inference result to the endoscopic device.

[0067] In addition, the computing device 100 can use an object detection model together with the trained artificial neural network model 200. The object detection model can track the bounding box including the lesion detected by the artificial neural network model 200. Therefore, a smoother detection result can be provided in the real-time video during which the endoscopic procedure is performed.

[0068] The various embodiments of the present disclosure described above can be combined with additional embodiments and can be modified within the scope understandable by those skilled in the art from the detailed description above. It should be understood that the embodiments of the present disclosure are illustrative in all aspects and not restrictive. For example, each component described as a single type can also be implemented in a distributed manner, and similarly, the components described as distributed can also be implemented in a combined form. Therefore, all changes or modifications derived from the meaning, scope, and equivalent concept of the claims of the present disclosure should be construed as being included within the scope of the present disclosure.

Claims

1. 1. A method for training an artificial neural network model for detecting lesions in endoscopic video, the method being performed by a computing device including at least one processor, the method comprising: generating learning data including a label for the lesion based on the endoscopic image according to a characteristic of a region where the endoscopic image is captured and a characteristic of the lesion; training an artificial neural network model to detect the lesion in the endoscopic image based on the training data; A method comprising:

2. The learning data includes lesion learning data including a lesion and normal learning data not including a lesion at a specific ratio, The method of claim 1 , wherein the percentage is determined by clinical characteristics of the lesion.

3. The method of claim 2 , wherein the clinical characteristics of the lesion include the frequency of occurrence of the lesion or the type of the lesion.

4. The method of claim 2 , wherein the lesion training data is generated by an enhancement technique that preserves image characteristics of the lesions.

5. The learning data includes learning data, verification data, and evaluation data, The method according to claim 1 , wherein the training data, the validation data, and the evaluation data are configured based on patient information from which the training data was obtained.

6. The step of training the artificial neural network model comprises:

2. The method of claim 1, further comprising the step of: assigning weights to reflect clinical characteristics of the lesion to determine an inference result of the artificial neural network model.

7. The step of assigning weights includes: The method of claim 6, further comprising the step of weighting training data containing lesions that are more difficult to interpret with a higher weighting.

8. The method of claim 1 , wherein the type of loss function of the artificial neural network model is determined by clinical characteristics of the lesion or image characteristics of the lesion.

9. The method of claim 8 , wherein the loss function of the artificial neural network model comprises a Distance IoU (DIoU) structured loss function.

10. The method of claim 1 , wherein the artificial neural network model is trained to detect lesion-free images in endoscopic images based on the training data.

11. The method of claim 10 , wherein a loss function of the artificial neural network model is determined based on a loss function for detecting the lesion and a loss function for detecting the image without the lesion.

12. 1. A method for detecting lesions in endoscopic images, the method being performed by a computing device including at least one processor, the method comprising: A method comprising: detecting a lesion or an image not including a lesion in an endoscopic image using an artificial neural network model trained with learning data generated based on at least one of characteristics of a region where an endoscopic image is captured, clinical characteristics of a lesion, and image characteristics of the lesion.

13. 1. A computing device for training an artificial neural network model for detecting lesions in endoscopic images, comprising: A processor including at least one core; a memory containing program code executable by the processor; Including, The processor generates learning data including a label for the lesion based on the endoscopic image according to characteristics of the area where the endoscopic image is captured and characteristics of the lesion, and trains an artificial neural network model to detect lesions in the endoscopic image based on the learning data.

Images