Method for intelligently identifying lesion area through endoscope based on AI technology

Abstract

The invention discloses a method for intelligently identifying a lesion area through an endoscope based on an AI technology, and relates to the technical field of lesion area identification, and the method comprises the following steps: carrying out the segmentation of a lesion area according to an endoscope image, and obtaining a lesion area mask image; performing multi-scale structural feature extraction according to the lesion area mask image to obtain structural features; lesion semantic category recognition is carried out according to the structural features, and lesion semantic categories are obtained; lesion risk level evaluation is carried out according to lesion semantic categories, and lesion risk levels are obtained; and generating a diagnosis suggestion according to the lesion risk level to obtain the diagnosis suggestion. According to the method, the structure change in the lesion area image is directly associated with the semantic category, so that the judgment accuracy and interpretability are improved, the requirement of a large-scale training data set is avoided, and the adaptability of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lesion area recognition, and specifically to a method for endoscopic intelligent recognition of lesion areas based on AI technology. Background Art

[0002] With the rapid development of medical imaging and artificial intelligence technologies, the application of endoscopic image analysis in clinical diagnosis has become increasingly important. As a common medical diagnostic method, endoscopy can directly obtain image information from the human body and provide more intuitive images of lesions. However, endoscopic images contain a large amount of detail, and image quality can be affected by factors such as the environment, equipment, and operation, making the manual analysis process cumbersome and error-prone. Therefore, the use of artificial intelligence (AI) technology, especially deep learning methods, to automatically analyze endoscopic images has become a research hotspot.

[0003] In existing technologies, there are deficiencies in the recognition of lesion semantic categories: existing technologies generally rely on deep learning methods, which tend to identify more common lesion types during training and ignore those rare or inconspicuous lesions, resulting in insufficient judgment accuracy. They often perform well on specific datasets, but when faced with endoscopic images from different devices, different patients or different hospitals, they may suffer from insufficient generalization performance, resulting in the model's performance on new datasets being lower than expected and insufficient adaptability. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a method for intelligently identifying lesion areas through endoscopy based on AI technology to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides a method for intelligently identifying lesion areas through an endoscope based on AI technology, comprising the following steps: S1. Segment the lesion area according to the endoscopic image to obtain a mask image of the lesion area; S2. Extract multi-scale structural features based on the lesion area mask image to obtain structural features; S3, identifying the semantic category of the lesion based on the structural features to obtain the semantic category of the lesion; S4. Evaluate the lesion risk level according to the lesion semantic category to obtain the lesion risk level; S5. Generate a diagnostic suggestion based on the lesion risk level to obtain a diagnostic suggestion.

[0006] To further optimize this technical solution, the multi-scale structural feature extraction in step S2 includes: Based on the obtained mask map of the lesion area, a multi-scale feature extraction model is used to extract the structural features of the lesion area at different scales, obtain structural features, reveal the morphological changes of the lesion area at different scales, and measure the structural complexity of the lesion area.

[0007] To further optimize this technical solution, the multi-scale feature extraction model includes:

[0008]

[0009] in: : structural energy at the kth scale; : Pixels at scales k and k+1 The structural residual at ; : Multi-scale structural feature vector.

[0010] Further optimizing this technical solution, the structural residual includes:

[0011] in: : The norm of the structure graph at the k+1th scale; : The norm of the structure graph at the kth scale; : The image of the structural residual between the kth and k+1th scales; By calculating the difference between two structural graphs at adjacent scales, the structural residual between the two scales is obtained.

[0012] To further optimize this technical solution, the structural diagram includes:

[0013] in: : Gaussian filter at the kth scale; : Mask image of lesion area; The mask image of the lesion area is blurred by a Gaussian filter to obtain structural image data at different scales.

[0014] To further optimize this technical solution, the lesion semantic category identification in step S3 includes: According to the obtained structural features, the structural semantic mapping model is used to map the structural features with the category knowledge space, so as to judge the lesion type and obtain the lesion semantic category.

[0015] To further optimize this technical solution, the structural semantic mapping model includes:

[0016] in: : The final predicted lesion semantic category label; : The predicted probability of belonging to the semantic category of the t-th lesion.

[0017] Further optimizing this technical solution, the predicted probability includes:

[0018] in: : Similarity sensitivity adjustment factor; : The similarity score between the multi-scale structural feature vector and the semantic prototype vector of the t-th type of lesion; The similarity scores are normalized by the Softmax function to obtain the predicted probability of each semantic category.

[0019] To further optimize this technical solution, the similarity score includes:

[0020] in: : the Euclidean norm of the semantic prototype vector of the t-th lesion; : the inner product of the multi-scale structural feature vector and the semantic prototype vector of the t-th type of lesion; The similarity between vectors is calculated by the inner product of the lesion semantic prototype vector and the multi-scale structural feature vector and the product of the Euclidean norm of the two vectors.

[0021] To further optimize this technical solution, the generation of diagnostic suggestions in step S5 includes: Based on the obtained lesion risk level, a series of diagnostic suggestion templates that match the risk level, semantic category, and structural characteristics are preset to generate diagnostic suggestions with clear logic and accurate content. It is also ensured that the output diagnostic suggestions have clear semantics, standardized grammar, and conform to medical expression habits.

[0022] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of a method for endoscopic intelligent identification of lesion areas based on AI technology as described in the first aspect of the present invention are implemented.

[0023] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of a method for endoscopic intelligent identification of lesion areas based on AI technology as described in the first aspect of the present invention are implemented.

[0024] Compared with the existing technology, the present invention provides a method for intelligently identifying lesion areas through endoscopy based on AI technology, which has the following beneficial effects: This AI-based method for endoscopic intelligent identification of lesion areas uses a structural semantic mapping model and structural similarity mapping to directly associate structural changes in lesion area images with semantic categories, thereby improving judgment accuracy and interpretability. By constructing a semantic prototype, it avoids the need for large-scale training datasets and improves the adaptability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 This is a flow chart of a method for intelligently identifying lesion areas through endoscopy based on AI technology proposed in the present invention; Figure 2 This is a flow chart of a multi-scale feature extraction model for an AI-based endoscopic intelligent lesion area recognition method proposed in the present invention; Figure 3 This is a flow chart of a structural semantic mapping model for an AI-based method for intelligently identifying lesion areas through endoscopy proposed in the present invention; Figure 4 This is a flow chart of a risk assessment model for a method for intelligent endoscopic identification of lesion areas based on AI technology proposed in the present invention. DETAILED DESCRIPTION

[0027] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0028] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0029] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it constitute a separate or selective embodiment that is mutually exclusive with other embodiments.

[0030] Example 1:

[0031] Reference Figures 1 to 4 , which is the first embodiment of the present invention, provides a method for intelligently identifying lesion areas through endoscopy based on AI technology, comprising the following steps: S1. Segment the lesion area according to the endoscopic image to obtain a mask image of the lesion area.

[0032] In this embodiment, the lesion area segmentation includes: Identifying lesions in endoscopic images is a core issue in medical image analysis. Traditional manual analysis not only relies on the physician's experience but can also be affected by visual fatigue, interpretation speed, and variations in interpretation standards, resulting in reduced diagnostic accuracy. To improve the efficiency and accuracy of disease diagnosis, especially the identification of early-stage lesions, the U-Net architecture is used to automatically segment and identify lesions. The U-Net architecture utilizes a symmetrical encoder-decoder structure, which fully extracts and restores image contextual information to the pixel level. Consequently, the U-Net architecture effectively segments lesions and provides precise image data for subsequent analysis and diagnosis.

[0033] The core purpose of this step is to identify and segment the lesion area from the endoscopic image through the detailed pixel-level segmentation of the U-Net architecture, so as to accurately identify the lesion area in the endoscopic image, such as polyps, ulcers, and bleeding spots, etc., reduce the errors that may occur when doctors observe the image, improve the accuracy and consistency of diagnosis, increase the speed of image analysis, reduce the workload of doctors, and provide data support for subsequent steps.

[0034] This step is implemented using the U-Net architecture. U-Net is a convolutional neural network (CNN) architecture suitable for semantic segmentation tasks in medical images. Its structure consists of two main parts: an encoder and a decoder. Skip connections are used to combine low-level features with high-level features. This allows the network to better restore image details during upsampling during decoding, especially edge details, effectively improving segmentation accuracy. The steps include: Input endoscopic image: The input image is the original image taken by the endoscope, generally a three-channel RGB color image; Encoder usage: The encoder is responsible for extracting high-level features from the input image. It gradually downsamples the image through multiple convolutional layers and pooling layers to obtain the global information of the image. With each convolution operation, the spatial resolution of the image gradually decreases, but the dimension of the feature gradually increases. This discards redundant details in the input image space, retains the key semantic content, and captures the semantic features of the image. Decoder usage: The decoder uses a deconvolution operation to upsample the image, restore the features extracted by the encoder to an image of the same size as the input image, and restore its spatial resolution. It then outputs the segmentation result of the lesion area, i.e., a mask image with the same size as the input image, with the lesion area marked as 1 and the normal area as 0. Model training: A large-scale endoscopic image dataset is used for training. Network parameters are optimized through multiple rounds of iterations, enabling the model to accurately learn how to identify lesions from endoscopic images. The training process uses standard gradient descent optimization algorithms, such as the Adam optimizer, to minimize the loss function until the model converges. The loss function uses a combination of Dice Loss and cross-entropy loss functions to effectively address class imbalance. Segmentation accuracy is improved by optimizing the overlap between lesion and background areas. Binary classification training is performed on each pixel to ensure that the boundary between lesion and normal areas can be accurately segmented.

[0035] S2. Extract multi-scale structural features based on the lesion area mask image to obtain structural features.

[0036] In this embodiment, the multi-scale structural feature extraction includes: In step S1, a mask map of the lesion area is obtained. The structural changes at different scales in the lesion area reflect its potential pathological state and are important indicators for disease classification and risk inference. Therefore, this step uses a multi-scale feature extraction model based on the obtained mask map of the lesion area to extract structural features at different scales in the lesion area, obtain structural features, reveal the morphological changes of the lesion area at different scales, and measure the structural complexity of the lesion area for subsequent lesion semantic type recognition and risk level assessment.

[0037] Furthermore, the multi-scale feature extraction model includes:

[0038]

[0039] in: : structural energy at the kth scale; : Pixels at scales k and k+1 The structural residual at ; : Multi-scale structural feature vector.

[0040] Furthermore, the structural residual includes:

[0041] in: : The norm of the structure graph at the k+1th scale, which represents the absolute value of each pixel in the image; : The norm of the structure graph at the kth scale, which represents the absolute value of each pixel in the image; : The image of the structural residual between the kth and k+1th scales, which contains the values ​​of the structural residual at all pixels; By calculating the difference between two structural graphs at adjacent scales, the structural residual between the two scales is obtained.

[0042] Furthermore, the structure diagram includes:

[0043] in: : Gaussian filter at the kth scale, used for blurring at different scales; : The lesion area mask image, obtained from step S1; The mask image of the lesion area is blurred by a Gaussian filter to obtain structural image data at different scales.

[0044] This model describes how to extract multi-scale structural features from the lesion area mask image and obtain the structural features of the lesion area at different scales.

[0045] Most traditional lesion area feature extraction methods focus on feature extraction at a single scale, which makes it difficult to capture subtle changes in the lesion area at different scales and ignores the multi-dimensional changes in the lesion area at different scales. This model performs multi-scale processing on the lesion area mask map. By performing residual modeling on the changes in the structural features of the lesion area at different scales, it reveals the morphological changes of the lesion area at different scales and evaluates the structural complexity and irregularity of the lesion area. It provides accurate and rich data support for subsequent semantic classification, risk assessment and other tasks, and improves the accuracy and reliability of the overall diagnosis.

[0046] The steps for using this model include: Multi-scale processing: through Gaussian filter The lesion area mask obtained in step S1 Perform multi-scale processing to obtain structural diagrams at different scales ; Residual calculation: According to the structural diagram at different scales , calculate the structural residual between adjacent scales , reveals the morphological changes of lesion areas at different scales, especially the differences in edge changes and subtle lesion areas, helping to identify lesion areas with complex morphology; Feature vector construction: Based on the structural residuals between adjacent scales , through energy quantification, we can further evaluate the structural complexity and irregularity of the lesion area, and the value of structural energy The larger the value, the more complex the structural changes in the area, and the higher the risk of lesions. Finally, the structural energy features at each scale are integrated into a multi-scale structural feature vector. This feature vector integrates the multi-dimensional structural information of the lesion area and provides key data support for subsequent semantic classification and risk assessment.

[0047] S3. Identify the semantic category of the lesion based on the structural features to obtain the semantic category of the lesion.

[0048] In this embodiment, the lesion semantic category identification includes: The structural features of the lesion area are obtained from step S2. This step needs to convert this structural feature into a lesion semantic category with clinical diagnostic value, such as benign polyps, malignant tumors, inflammatory erosions, etc., so as to realize automatic judgment of the lesion type and assist doctors in quickly identifying the lesion type.

[0049] This step uses a structural semantic mapping model to map the structural features with the category knowledge space based on the obtained structural features, thereby judging the lesion type and obtaining the lesion semantic category to assist doctors in quickly identifying the lesion type.

[0050] Furthermore, the structural semantic mapping model includes:

[0051] in: : The final predicted lesion semantic category label; : The predicted probability of belonging to the semantic category of the t-th lesion.

[0052] Furthermore, the predicted probability includes:

[0053] in: : Similarity sensitivity adjustment factor, used to control the steepness of the output distribution of the Softmax function. By adjusting this value, the model's prediction tendency for the closest category can be changed. The value range is between 0.1 and 10. The larger the value, the steeper the Softmax function, indicating stronger discrimination between categories. The smaller the value, the weaker the difference between categories. : The similarity score between the multi-scale structural feature vector and the semantic prototype vector of the t-th lesion, indicating the similarity between the multi-scale structural feature vector and the semantic prototype vector of the lesion; The similarity scores are normalized by the Softmax function to obtain the predicted probability of each semantic category.

[0054] Furthermore, the similarity score includes:

[0055] in: : The Euclidean norm of the semantic prototype vector of the t-th lesion. The dimension of the lesion semantic prototype vector is the same as that of the multi-scale structural feature vector. : the inner product of the multi-scale structural feature vector and the semantic prototype vector of the t-th type of lesion; The similarity between vectors is calculated by the inner product of the lesion semantic prototype vector and the multi-scale structural feature vector and the product of the Euclidean norm of the two vectors.

[0056] This model describes how to determine the lesion type based on the structural characteristics of the lesion area.

[0057] Traditional lesion semantic category recognition methods have difficulty explicitly modeling the logical connection between structural changes and semantic categories, have difficulty explaining the separability of categories, require a large number of labeled samples for training, and require more computing resources and time. However, this model directly associates structural changes in images with semantic categories through structural similarity mapping, improving judgment accuracy and interpretability. By constructing semantic prototypes, it avoids the need for large-scale training datasets and improves the adaptability of the model.

[0058] The steps for using the above model include: Similarity calculation: Get the multi-scale structural feature vector from step S2 , calculate the similarity between it and all kinds of lesion semantic prototype vectors, and obtain the multi-scale structural feature vector Similarity scores with all kinds of lesion semantic prototype vectors; Category probability calculation: Based on the calculated similarity scores, all similarities are normalized using the Softmax function to obtain the probability that the structural features belong to various lesion semantic categories; Semantic category output: According to the probability that the obtained structural features belong to various types of lesion semantic categories, the lesion semantic category with the maximum probability is selected Output as a result.

[0059] S4. Evaluate the lesion risk level according to the lesion semantic category to obtain the lesion risk level.

[0060] In this embodiment, the lesion risk level assessment includes: The lesion semantic category is obtained from step S3. The purpose of this step is to quantitatively analyze the severity of the lesion area, that is, the risk, to help doctors make a more accurate diagnosis and ultimately assess the risk level of the lesion.

[0061] This step uses a risk assessment model based on the obtained lesion semantic category, combined with structural similarity and lesion area ratio to conduct a more comprehensive risk analysis, thereby reflecting the risk of the lesion area more comprehensively and accurately and obtaining a more accurate lesion risk level.

[0062] Furthermore, the risk assessment model includes:

[0063] in: : Comprehensive risk score; : the total number of lesion semantic categories; : Lesion semantic category The risk sensitivity coefficient ranges from 0 to 1 and is used to measure the contribution of the similarity score to the risk. The larger the value, the more significant the impact of the similarity between the structural features and the lesion semantic category on the risk increase. It is set based on historical case statistics according to different lesion semantic categories. : Lesion semantic category The area weight coefficient ranges from 0 to 1 and is used to measure the area ratio to the lesion semantic category. The risk-added effect is set based on historical case statistics according to different lesion semantic categories; : The normalized area ratio of the lesion area is calculated by dividing the number of pixels in the lesion area mask by the total number of pixels in the entire image, which is used to reflect the extent of lesion expansion.

[0064] The model describes how to assess lesion risk levels based on lesion semantic categories.

[0065] Traditional lesion risk assessment methods are mostly based on a single indicator and fixed assessment rules, which are difficult to adapt to diverse lesions, have low adaptability, and it is difficult to quantify the specific contribution of each factor, and lack interpretability. This model simultaneously considers the semantic category, structural similarity and area ratio of lesions, making the assessment more comprehensive. It can also adjust the model parameters, clearly showing the contribution of each information to the final risk assessment, and improving the adaptability and interpretability of the model.

[0066] The steps for using the model include: Data input: The probability of each lesion semantic category obtained from step S3 and similarity score , according to the lesion area mask obtained in step S1, the normalized area ratio is obtained ; Risk calculation: For each lesion area category, calculate the contribution of the category to the risk score and sum them to obtain a comprehensive risk score. ; Risk level classification: A threshold is set for the risk score based on experts and historical statistics, and three levels of risk are divided: high risk, medium risk, and low risk. The resulting comprehensive risk score is compared with the threshold to obtain the lesion risk level.

[0067] S5. Generate a diagnostic suggestion based on the lesion risk level to obtain a diagnostic suggestion.

[0068] In this embodiment, the generating of the diagnostic suggestion includes: At present, doctors face a large number of image judgment tasks in their actual work. It is difficult to quickly form a diagnostic judgment by relying solely on the lesion risk level value. Using the risk level to further generate diagnostic recommendations can significantly reduce the burden on doctors and improve the efficiency and consistency of medical decision-making.

[0069] The lesion risk level is obtained from step S4. The purpose of this step is to convert the obtained lesion risk level into a diagnostic suggestion with clear logic and accurate content, build an intelligent auxiliary diagnostic tool with basic medical logical reasoning capabilities, and provide assistance to doctors in decision-making.

[0070] This step presets a series of diagnostic suggestion templates that match the risk level, semantic category, and structural features based on the obtained lesion risk level. For example, for a lesion area with a structural rule of "mucosal damage + vascular distortion", a semantic category of "gastric erosion", and a high risk score, the suggestion "immediate gastroscopic biopsy and pathological analysis are recommended to rule out the possibility of malignancy" is generated. This generates diagnostic suggestions with clear logic and accurate content based on the obtained lesion risk level, and ensures that the output diagnostic suggestions are semantically clear, grammatically standardized, and conform to medical expression habits.

[0071] Example 2: This embodiment also provides a computer device, which is suitable for a method for endoscopic intelligent identification of lesion areas based on AI technology, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a method for endoscopic intelligent identification of lesion areas based on AI technology as proposed in the above embodiment.

[0072] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, a method for endoscopic intelligent identification of lesion areas based on AI technology as proposed in the above embodiment is implemented.

[0073] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0074] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0075] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0076] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0077] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0078] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligently identifying lesion areas through endoscopy based on AI technology, characterized in that: The following steps are involved: S1. Segment the lesion area according to the endoscopic image to obtain a mask image of the lesion area; S2. Extract multi-scale structural features based on the lesion area mask image to obtain structural features; S3, identifying the semantic category of the lesion based on the structural features to obtain the semantic category of the lesion; S4. Evaluate the lesion risk level according to the lesion semantic category to obtain the lesion risk level; S5. Generate a diagnostic suggestion based on the lesion risk level to obtain a diagnostic suggestion.

2. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 1, characterized in that: The multi-scale structural feature extraction in step S2 includes: Based on the obtained mask map of the lesion area, a multi-scale feature extraction model is used to extract the structural features of the lesion area at different scales, obtain structural features, reveal the morphological changes of the lesion area at different scales, and measure the structural complexity of the lesion area.

3. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 2, characterized in that: The multi-scale feature extraction model includes: ， ， in: : structural energy at the kth scale; : Pixels at scales k and k+1 The structural residual at ; : Multi-scale structural feature vector.

4. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 3, characterized in that: The structural residuals include: ， in: : The norm of the structure graph at the k+1th scale; : The norm of the structure graph at the kth scale; : The image of the structural residual between the kth and k+1th scales; By calculating the difference between two structural graphs at adjacent scales, the structural residual between the two scales is obtained.

5. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 4, characterized in that: The structure diagram includes: ， in: : Gaussian filter at the kth scale; : Mask image of lesion area; The mask image of the lesion area is blurred by a Gaussian filter to obtain structural image data at different scales.

6. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 1, characterized in that: The lesion semantic category identification in step S3 includes: According to the obtained structural features, the structural semantic mapping model is used to map the structural features with the category knowledge space, so as to judge the lesion type and obtain the lesion semantic category.

7. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 6, characterized in that: The structural semantic mapping model includes: ， in: : The final predicted lesion semantic category label; : The predicted probability of belonging to the semantic category of the t-th lesion.

8. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 7, characterized in that: The predicted probabilities include: ， in: : Similarity sensitivity adjustment factor; : The similarity score between the multi-scale structural feature vector and the semantic prototype vector of the t-th type of lesion; The similarity scores are normalized by the Softmax function to obtain the predicted probability of each semantic category.

9. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 8, characterized in that: The similarity score includes: ， in: : the Euclidean norm of the semantic prototype vector of the t-th lesion; : the inner product of the multi-scale structural feature vector and the semantic prototype vector of the t-th type of lesion; The similarity between vectors is calculated by the inner product of the lesion semantic prototype vector and the multi-scale structural feature vector and the product of the Euclidean norm of the two vectors.

10. The method for intelligently identifying lesion areas through endoscopy based on AI technology according to claim 1, characterized in that: The generation of the diagnostic suggestion in step S5 includes: Based on the obtained lesion risk level, a series of diagnostic suggestion templates that match the risk level, semantic category, and structural characteristics are preset to generate diagnostic suggestions with clear logic and accurate content. It is also ensured that the output diagnostic suggestions have clear semantics, standardized grammar, and conform to medical expression habits.

Images