Medical image recognition method, electronic device, and storage medium

Abstract

The present invention relates to the field of artificial intelligence, and in particular to a medical image recognition method, electronic device, and storage medium. The medical image recognition method according to the present invention includes: obtaining an image to be recognized, the image to be recognized includes a medical image of a target to be detected; segmenting the image to be recognized to generate a target image block; using a preamble codec predictor to extract features of the target image block to generate a feature extraction vector; based on the feature extraction vector, the target to be detected is recognized to obtain a recognition result. The medical image recognition method in the present invention utilizes a preamble codec predictor to process the image to be recognized. Since the preamble codec predictor can process the image based on the correlation between similar pixels in the image to be recognized, the medical image recognition method in the present invention can improve the accuracy of the medical image recognition process.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a medical image recognition method, electronic equipment, and storage medium. Background Art

[0002] The Transformer is a model that uses the attention mechanism to improve training speed. The attention mechanism is a mechanism that improves the recognition accuracy of the model by adjusting the attention to the input weight distribution. The Transformer model encodes inputs and calculates outputs based on the attention mechanism, allowing the model to be trained in parallel and have global information, which helps improve the recognition accuracy of the model.

[0003] In recent years, the application of image processing technology in natural images has made great progress, but there are still many difficulties in the processing of medical images. For example, the semantic information contained in medical images is relatively simple, so it is difficult to obtain effective information in medical images. How to identify medical images in a more accurate way is still a problem that needs to be solved by industry insiders. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a medical image recognition method, an electronic device, and a storage medium, which can improve the accuracy of medical image recognition.

[0005] A medical image recognition method according to a first aspect of the present invention includes:

[0006] Acquire an image to be identified, wherein the image to be identified includes a medical image of a target to be detected;

[0007] Segmenting the image to be identified to generate target image blocks;

[0008] Using a preamble codec predictor to perform feature extraction on the target image block to generate a feature extraction vector;

[0009] The target to be detected in the image to be identified is identified based on the feature extraction vector to obtain an identification result.

[0010] Optionally, according to some embodiments of the present invention, the method further includes:

[0011] Recognize the unlabeled preset data set by using a basic recognition model to obtain a pseudo-labeled data set, wherein the basic recognition model is obtained by pre-training the original preset model;

[0012] Taking a standard data set with preset labels as positive sample data and the pseudo-label data set as negative sample data, the positive sample data and the negative sample data are mixed to form an input sample set;

[0013] The basic recognition model is optimized and trained using the input sample set to obtain the preamble codec predictor.

[0014] Optionally, according to some embodiments of the present invention, the method of taking a standard data set with preset labels as positive sample data and taking the pseudo-label data set as negative sample data, and mixing the positive sample data with the negative sample data to form an input sample set includes:

[0015] Mixing the positive sample data and the negative sample data to obtain mixed data;

[0016] Adjusting the proportions of the positive sample data and the negative sample data in the mixed data;

[0017] When the proportion of the positive sample data in the mixed data is greater than the proportion of the negative sample data in the mixed data, the adjusted mixed data is used as the input sample set.

[0018] Optionally, according to some embodiments of the present invention, the optimizing and training the basic recognition model using the input sample set to obtain the preamble codec predictor includes:

[0019] Inputting the input sample set into the basic recognition model to perform a first type of iterative training;

[0020] After each round of the first type of iterative training, the recognition accuracy of the basic recognition model is calculated and the first type of parameters of the basic recognition model are updated;

[0021] The variation of the recognition accuracy rate in each round of the first type of iterative training is counted, and when the recognition accuracy rate converges to a first constant value, the trained preamble codec predictor is obtained.

[0022] Optionally, according to some embodiments of the present invention, segmenting the image to be identified to generate a target image block includes:

[0023] Preliminarily determining the tissue structure features of the target to be detected in the image to be identified, and obtaining an image segmentation type that matches the tissue structure features;

[0024] The image to be identified is segmented according to the image segmentation type to generate the target image block.

[0025] Optionally, according to some embodiments of the present invention, the preliminarily determining the tissue structure feature of the target to be detected in the image to be identified and acquiring the image segmentation type matching the tissue structure feature includes:

[0026] Obtaining pixel distribution in the image to be identified, and performing the preliminary determination on the feature type of the tissue structure feature according to the pixel distribution;

[0027] The image segmentation type is acquired according to the feature type of the tissue structure feature.

[0028] Optionally, according to some embodiments of the present invention, the image segmentation type includes strip sampling segmentation, and the preliminarily determining the tissue structure feature of the target to be detected in the image to be identified and acquiring the image segmentation type matching the tissue structure feature further includes:

[0029] When the tissue structure feature is a hierarchical distribution feature type, the strip sampling segmentation is obtained to define a segmentation sampling shape, the length of the segmentation sampling shape is greater than the width, and the length direction of the segmentation sampling shape is consistent with the extension direction of the tissue structure hierarchical boundary line.

[0030] Optionally, according to some embodiments of the present invention, segmenting the to-be-recognized image according to the image segmentation type to generate the target image block includes:

[0031] Segmenting the image to be identified based on the segmentation sampling shape defined by the strip sampling segmentation to obtain a preliminary image block;

[0032] The size of the preliminary image block is adjusted to a preset size by zero-filling to obtain the target image block.

[0033] In a second aspect, an embodiment of the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the medical image recognition method as described in any one of the embodiments of the first aspect of the present invention.

[0034] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement a medical image recognition method as described in any one of the embodiments of the first aspect of the present invention.

[0035] The medical image recognition method, electronic device, and storage medium according to the embodiments of the present invention have at least the following beneficial effects:

[0036] The medical image recognition method of the present invention first obtains an image to be recognized including a target tissue structure to be detected, and then segments the image to be recognized to generate a target image block. Further, a preamble codec predictor is used to extract features of the target image block to generate a feature extraction vector, and then the tissue structure of the target to be detected is recognized based on the feature extraction vector to obtain a recognition result. The medical image recognition method of the present invention utilizes a preamble codec predictor to process the image to be recognized. Since the preamble codec predictor can process the image based on the correlation between similar pixels in the image to be recognized, the medical image recognition method of the present invention can improve the accuracy of the medical image recognition process.

[0037] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0039] Figure 1 A flow chart of a medical image recognition method provided by an embodiment of the present invention;

[0040] Figure 2 A flow chart of another medical image recognition method provided by an embodiment of the present invention;

[0041] Figure 3 A flow chart of another medical image recognition method provided by an embodiment of the present invention;

[0042] Figure 4 A flow chart of another medical image recognition method provided by an embodiment of the present invention;

[0043] Figure 5 A flow chart of another medical image recognition method provided by an embodiment of the present invention;

[0044] Figure 6 A flow chart of another medical image recognition method provided by an embodiment of the present invention;

[0045] Figure 7 A flow chart of another medical image recognition method provided by an embodiment of the present invention;

[0046] Figure 8 A flow chart of another medical image recognition method provided by an embodiment of the present invention;

[0047] Fig. 9 A schematic diagram of segmentation processing of a medical image recognition method provided by an embodiment of the present invention;

[0048] Fig.10A schematic diagram of an electronic device for executing a medical image recognition method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0050] In the description of the present invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0051] In the description of the present invention, it should be understood that descriptions involving orientation, such as up, down, left, right, front, back, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0052] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0053] In the description of the present invention, it should be noted that, unless otherwise clearly defined, the terms such as setting, installing, connecting, etc. should be understood in a broad sense, and the technical personnel in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention in combination with the specific content of the technical solution. In addition, the identification of specific steps below does not represent a limitation on the order of steps and execution logic, and the execution order and execution logic between steps should be understood and inferred with reference to the corresponding illustrative expressions.

[0054] The Transformer is a model that uses the attention mechanism to improve training speed. The attention mechanism is a mechanism that improves the recognition accuracy of the model by adjusting the attention to the input weight distribution in the encoder-decoder. The Transformer model encodes inputs and calculates outputs based on the attention mechanism, allowing the model to be trained in parallel and have global information, which helps improve the recognition accuracy of the model.

[0055] The uses of medical image recognition include: before diagnosis, it can be used to predict individual or group diseases and give health advice; during diagnosis, artificial intelligence can assist in diagnosis and treatment to reduce the misdiagnosis rate; after diagnosis, it can ensure the authenticity of patients' medication through channels such as computer vision, image recognition and video analysis, and assist doctors in supervising patients' medication compliance. The uses of medical image recognition are not limited to the above examples.

[0056] In recent years, the application of image processing technology in natural images has made great progress, but there are still many difficulties in the processing of medical images. For example, the semantic information contained in medical images is relatively simple, so it is difficult to obtain effective information in medical images. How to identify medical images in a more accurate way is still a problem that needs to be solved by industry insiders.

[0057] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a medical image recognition method, an electronic device, and a storage medium, which can improve the accuracy of medical image recognition.

[0058] Further description is given below with reference to the accompanying drawings.

[0059] A medical image recognition method according to a first aspect of the present invention includes:

[0060] Step S101, obtaining an image to be identified, where the image to be identified is a medical image including an object to be detected;

[0061] It should be noted that a medical image is an image rich in anatomical information, and this anatomical information is called anatomical context. In the field of medical image recognition, medical images usually need to contain specific tissue structures, such as pancreatic tissue, brain tissue, lung tissue, heart tissue, breast tissue, etc. The main task of medical image recognition is to analyze the anatomical information in the medical image, match the anatomical information with the Foundational Model of Anatomy (FMA), and convert the anatomical information into a form that humans can understand, and then identify the lesion area from the specific tissue structure to help diagnose the disease. It should be understood that the purpose of obtaining the image to be identified is to provide a recognition basis for the subsequent recognition and processing of the image, and based on the detection requirements of different symptoms, the tissue structure of the target to be detected can have a variety of structural features, such as hierarchical distribution of structural features, regional distribution of structural features (such as lung shadows, pancreatic spots, etc.) and other types.

[0062] According to some embodiments provided by the present invention, the image to be identified can be a medical image including the target tissue structure to be detected, which is obtained based on optical coherence tomography technology. It should be noted that optical coherence tomography (OCT) is an imaging technology that has developed rapidly in recent years. It uses the basic principle of weak coherence light interferometer to detect the back reflection or several scattered signals of the incident weak coherent light at different depth levels of biological tissue, and obtains a two-dimensional or three-dimensional structural image of the biological tissue by scanning. Since OCT can achieve real-time non-destructive non-contact imaging with an axial resolution of micrometers within a depth range of millimeters, it can effectively fill the gap between ultrasound and confocal microscopy, so that the OCT image has the advantages of fast imaging speed, real-time dynamic imaging, non-contact and non-invasive. Therefore, in some preferred embodiments of the present invention, OCT images are selected as images to be identified.

[0063] Step S102, segmenting the image to be identified to generate target image blocks;

[0064] In some embodiments of the present invention, the medical image recognition method needs to first perform basic detection and positioning on the target to be detected, complete the recognition of the target to be detected, and after the recognition or detection is completed, obtain the outer bounding box (Bounding Box) of the target tissue structure to be detected, that is, the target image block. It should be noted that in some embodiments of the present invention, after obtaining the target image block, the precise boundary of the target tissue structure to be detected is further found in the target image block based on the segmentation task. When there are multiple target objects in the image, the segmentation of multiple targets becomes a task of semantic parsing, that is, assigning semantic labels (Semantic Labels) to pixels in a two-dimensional image or a three-dimensional image. By labeling the pixels or voxels of the same target with the same label, the segmentation of the target is completed. It should be emphasized that since the semantic information of medical images is relatively simple, and the construction of negative example pairs in the three-dimensional medical image recognition process is judged based on the vertical distance after the slice segmentation, this is not applicable in the recognition of two-dimensional medical images. Therefore, in some embodiments of the present invention, the method of constructing negative example pairs by strip cutting sampling is used to strengthen the neighborhood association in the image to be recognized and improve the pixel segmentation accuracy in the segmentation process.

[0065] Step S103, using a preamble codec predictor to perform feature extraction on the target image block to generate a feature extraction vector;

[0066] It should be noted that the pre-order codec predictor (Transformer) is a model that uses an attention mechanism to improve the training speed, wherein the attention mechanism is a mechanism that improves the model recognition accuracy by adjusting the attention to the input weight distribution in the encoder-decoder (Encoder-Decoder). The Transformer model encodes the input and calculates the output based on the attention mechanism, so that the model can be trained in parallel and have global information, which helps to improve the model recognition accuracy. Some embodiments of the present invention introduce the pre-order codec predictor Transformer into the image recognition process, which can be achieved by introducing the self-attention mechanism (Self-Attention) in the Convolutional Neural Networks (CNN), or by directly replacing the convolution block with the Transformer. The vision pre-order codec predictor (Vision Transformer, ViT) is a visual model based on the Transformer architecture. ViT represents the input image as a series of tiles, similar to a series of word embeddings used when applying the Transformer to text, and can directly predict the category label of the image. When trained with sufficient data, ViT can show excellent performance, better than its equivalent advanced CNN, and the required resources are only one-quarter of the latter. Therefore, in some preferred embodiments of the present invention, a visual preamble codec predictor Vision Transformer with better performance is selected as the preamble codec predictor Transformer of the present invention. It should be understood that the visual preamble codec predictor is one type of preamble codec predictor, and the visual preamble codec predictor can use the anatomical structure information in the image to be recognized to recognize the medical image, thereby making the medical image recognition method have a higher accuracy rate.

[0067] In some embodiments of the present invention, a preamble codec predictor is used to perform feature extraction on a target image block to generate a feature extraction vector. This can be performed in an encoder in a visual preamble codec predictor Vision Transformer. It should be noted that when the encoder performs feature extraction on an image to be recognized, the generated latent code (LatentCode) is the feature extraction vector.

[0068] Step S104: identifying the target to be detected in the image to be identified based on the feature extraction vector to obtain a recognition result.

[0069] In some embodiments of the present invention, the recognition result is obtained by recognizing the target to be detected based on the feature extraction vector, which is obtained by the encoder Decoder in the visual preamble codec predictor Vision Transformer decoding the latent code serving as the feature extraction vector and attaching a classification label.

[0070] The medical image recognition method of the present invention first obtains an image to be recognized including a target to be detected, and then segments the image to be recognized to generate a target image block. Further, a preamble codec predictor Transformer is used to extract features of the target image block to generate a feature extraction vector, and then the target to be detected is recognized based on the feature extraction vector to obtain a recognition result. The medical image recognition method of the present invention utilizes a preamble codec predictor Transformer to process the image to be recognized. Since the preamble codec predictor Transformer can process the image based on the correlation between similar pixels in the image to be recognized, the medical image recognition method of the present invention can improve the accuracy of the medical image recognition process. It should be understood that in steps S101 to S104, steps S101 and S102 can be completed in the preamble codec predictor Transformer, or can be completed in other ways.

[0071] Reference Figure 2 According to some embodiments of the present invention, segmenting the image to be identified to generate target image blocks includes:

[0072] Step S201, preliminarily determining the tissue structure features of the target to be detected in the image to be identified, and obtaining the image segmentation type matching the tissue structure features;

[0073] It should be noted that, since the semantic information of medical images is relatively simple, and the construction of negative pairs in the process of three-dimensional medical image recognition is judged based on the vertical distance after slice segmentation, this is not applicable in the recognition of two-dimensional medical images. Therefore, in some embodiments of the present invention, the method of setting the image segmentation type for sampling, thereby constructing a negative pair is used to enhance the neighborhood association in the image to be identified and improve the pixel segmentation accuracy in the segmentation process. Among them, the image segmentation type refers to the mode type of image segmentation. The image segmentation type is determined according to the tissue structure of the target to be detected in the image to be identified, and the image segmentation type matches the tissue structure of the target to be detected. For example, when the tissue structure of the target to be detected in the image to be identified has a regional distribution structural feature (such as lung shadows, pancreatic spots, etc.), the image segmentation type is set to square sampling; for another example, when the tissue structure of the target to be detected in the image to be identified has a hierarchical distribution structural feature, the image segmentation type is set to strip sampling. It should be understood that the image segmentation type is determined according to the tissue structure of the target to be detected in the image to be identified, and can match the tissue structure features of the target to be detected to a large extent, so that the target image block obtained by segmentation can be more accurately identified and classified based on the attention mechanism in the preamble codec predictor Transformer. It should be noted that the preliminary determination of the tissue structure features of the target to be detected in the image to be identified can be performed in a variety of ways, including: obtaining the tissue structure features of the target to be detected through the pixel distribution in the image to be identified, obtaining the tissue structure features of the target to be detected through the annotation tags of the image to be identified, and obtaining the tissue structure features of the target to be detected based on the contrast or color difference in the image to be identified. It should be understood that the preliminary determination of the tissue structure features of the target to be detected in the image to be identified is not limited to the specific embodiments listed above.

[0074] Step S202: segment the image to be identified according to the image segmentation type to generate target image blocks.

[0075] In some embodiments of the present invention, the medical image recognition method needs to first perform basic detection and positioning of the target to be detected, complete the recognition of the target to be detected, and after the recognition or detection is completed, obtain the outer frame of the target tissue structure to be detected, that is, the target image block. It should be noted that since the semantic information of medical images is relatively simple, and the construction of negative examples in the three-dimensional medical image recognition process is based on the vertical distance after slice segmentation, this is not applicable in the recognition of two-dimensional medical images. Therefore, in some embodiments of the present invention, the image to be recognized is segmented by image segmentation type to obtain the target image block, and then the target image block is further extracted. The neighborhood correlation in the image to be recognized can be enhanced and the pixel segmentation accuracy in the segmentation process can be improved. It should be understood that the image segmentation type depends on the tissue structure of the target to be detected in the image to be identified. For example, when the tissue structure of the target to be detected in the image to be identified has regional distribution structural features (such as lung shadows, pancreatic spots, etc.), then setting the image segmentation type to square sampling can make the sampling frame include as many discrete pixel areas (such as lung shadows, pancreatic spots, etc.) as possible; for another example, when the tissue structure of the target to be detected in the image to be identified has hierarchical distribution structural features, setting the image segmentation type to strip sampling can make the sampling frame contain as many pixels of the same layer as possible. The strip cutting method also allows the attention mechanism in the pre-order codec predictor Transformer to play a role, thereby enhancing the neighborhood correlation in the image and improving the pixel segmentation accuracy.

[0076] Reference Figure 3 According to some embodiments of the present invention, a preliminary determination is made on the tissue structure features of the target to be detected in the image to be identified, and an image segmentation type matching the tissue structure features is obtained, including:

[0077] Step S301, obtaining pixel distribution in the image to be identified, and making a preliminary determination of the feature type of the tissue structure feature according to the pixel distribution;

[0078] According to some embodiments provided by the present invention, it is necessary to make a preliminary judgment on the tissue structure characteristics of the target to be detected by the pixel distribution in the image to be identified. For example, when the tissue structure characteristics of the target to be detected are the characteristic type of hierarchical distribution, the pixel distribution in the image to be identified will form a dividing line with a relatively obvious contrast in the image to be identified. Therefore, when it is detected that there is a dividing line with a relatively obvious contrast in the pixel distribution in the image to be identified, it can be determined that the tissue structure characteristics of the target to be detected are the characteristic type of hierarchical distribution; for another example, when the tissue structure characteristics of the target to be detected in the image to be identified are the characteristic type of regional distribution, the pixel distribution in the image to be identified will form discrete areas, and the more typical manifestations include but are not limited to: lung shadow areas, pancreatic spot areas, etc. Therefore, when it is detected that there are discrete areas in the pixel distribution in the image to be identified, it can be determined that the tissue structure characteristics of the target to be detected in the image to be identified are the characteristic type of regional distribution. It should be understood that the determination of the tissue feature type of the target to be detected according to the pixel distribution in the image to be identified is not limited to the above-mentioned embodiments.

[0079] Step S302, obtaining the image segmentation type according to the feature type of the tissue structure feature;

[0080] According to some embodiments provided by the present invention, the image segmentation type includes strip sampling segmentation. When the tissue structure feature is a hierarchically distributed feature type, strip sampling segmentation is obtained to define a segmentation sampling shape. The length of the segmentation sampling shape is greater than the width, and the length direction of the segmentation sampling shape is consistent with the extension direction of the tissue structure hierarchy dividing line.

[0081] It should be noted that in the image to be identified, there are boundaries with relatively obvious contrast between the various levels of the organizational structure with hierarchical distribution structural features, which are called hierarchical boundaries of the organizational structure. When the organizational structure feature of the target to be detected in the image to be identified is a hierarchical distribution feature type, the image segmentation type is set to strip sampling, so that there are more pixels of the same layer in the sampling frame. The strip cutting method allows the attention mechanism in the pre-order codec predictor Transformer to play a role. A more specific embodiment is used below to illustrate. When the extension direction of the hierarchical boundary line of the organizational structure of the target to be detected is consistent with the horizontal direction of the image to be identified, the layers are divided from top to bottom: the first layer of the organizational structure of the target to be detected is located at the top of the image to be identified, the second layer of the organizational structure of the target to be identified is located in the middle of the image to be identified, and the third layer of the organizational structure of the target to be identified is located at the bottom of the image to be identified, wherein there is a boundary line with relatively obvious contrast between the first layer and the second layer of the organizational structure of the target to be identified, and there is also a boundary line with relatively obvious contrast between the second layer and the third layer. When horizontal prediction is performed at the same height in the image to be identified, the pixel category of the strip sampling block on one side of the horizontal direction will give favorable guidance to the strip sampling block on the other side of the horizontal direction, because it is highly likely that they belong to the same layer, and when vertical prediction is performed at different heights, most of the pixel points in the strip sampling blocks at the upper and lower positions are more likely to be of non-same layer, so the pixel category of the strip sampling block on one side of the vertical direction will give guidance to distinguish the strip sampling block on the other side of the vertical direction.

[0082] According to some embodiments provided by the present invention, when the medical image recognition method of the present invention is applied to recognize retinal medical images of the eye, setting the image segmentation type to strip sampling segmentation is a more preferred embodiment. The reason is that the retinal medical images of the eye have typical hierarchical distribution structural characteristics, and the pixel distribution in the image to be recognized will form a boundary line with a relatively obvious contrast in the image to be recognized. At this time, setting the image segmentation type to strip sampling segmentation can make as many pixels of the same layer as possible exist in the sampling frame. The strip cutting method also allows the attention mechanism in the preceding codec predictor Transformer to play a role, thereby enhancing the neighborhood correlation in the image and improving the pixel segmentation accuracy. It should be noted that the length of the segmentation sampling shape is greater than the width, and the length direction of the segmentation sampling shape is consistent with the extension direction of the organizational structure hierarchy boundary line, which refers to the description of the segmentation sampling shape in the case of strip sampling segmentation.

[0083] The following is a description of some specific embodiments of the present invention. Fig. 9 , Fig. 9901a, 901b, 902a, 902b, and 903 are shown as the left end of the first level boundary line, the right end of the first level boundary line, the left end of the second level boundary line, and the right end of the second level boundary line. Fig. 9 The organizational structure feature of the target to be detected is a hierarchical distribution feature type. According to the left end 901a of the first level dividing line, the right end 901b of the first level dividing line, the left end 902a of the second level dividing line, and the right end 902b of the second level dividing line, the extension direction of the organizational structure level dividing line can be determined to be from the left side to the right side of the image to be identified, or from the left side to the right side of the image to be identified. Therefore, Fig. 9 Therefore, the length direction of the segmentation sampling shape 903 is consistent with the extension direction of the organizational level boundary line, which means that the length direction of the multiple segmentation sampling shapes 903 obtained by delineation is horizontal (such as Fig. 9 It should be understood that in some embodiments of the present invention, the segmentation sampling shape 903 includes but is not limited to a regular strip sampling frame, and may also be an irregular strip sampling frame. In addition, the length direction of the segmentation sampling shape 903 is consistent with the extension direction of the organizational structure level boundary line, which should not be narrowly understood as the two being completely the same. The situation where the length direction of the segmentation sampling shape 903 is approximately consistent with the extension direction of the organizational structure level boundary line is also included in the embodiments of the present invention.

[0084] In some other embodiments provided by the present invention, when the hierarchical boundary between the target tissue structures to be detected in the image to be identified connects the top and bottom of the image to be identified, the extension direction of the hierarchical boundary is the longitudinal direction, so the extension direction of the tissue structure hierarchical boundary is the longitudinal direction, and the length direction of the corresponding segmentation sampling shape is also the longitudinal direction. It should be understood that the length direction of the segmentation sampling shape depends on the extension direction of the tissue structure hierarchical boundary in the image to be identified, so when identifying different images to be identified, the length direction of the segmentation sampling shape may also be different. It should be emphasized that the length of the segmentation sampling shape is greater than the width, and the length direction of the segmentation sampling shape is consistent with the extension direction of the tissue structure hierarchical boundary, which is intended to indicate that the strip sampling frame is determined based on the structural characteristics of the target tissue structure to be detected, so the length direction of the segmentation sampling shape is consistent with the extension direction of the tissue structure hierarchical boundary, which should not be understood as the same direction, but should also include the situation where the length direction of the segmentation sampling shape is roughly the same as the extension direction of the tissue structure hierarchical boundary.

[0085] Reference Figure 4 According to some embodiments of the present invention, segmenting the image to be identified according to the image segmentation type to generate a target image block includes:

[0086] Step S401, segmenting the image to be identified based on the segmentation sampling shape defined by the strip sampling segmentation to obtain a preliminary image block;

[0087] It should be understood that the purpose of segmenting the image to be identified based on the segmentation sampling shape defined by the strip sampling segmentation is to segment the image to be identified with the strip sampling frame, thereby obtaining a strip-shaped target image block, so that the attention mechanism in the preceding codec predictor Transformer can play a role, thereby improving the accuracy of the medical image recognition method of the present invention. It should be noted that the preliminary image block refers to an image block that has not been resized. According to some embodiments provided by the present invention, the image to be identified is segmented based on the segmentation sampling shape defined by the strip sampling segmentation. If the identification is directly based on the preliminary image block that has not been resized, the size may not match in the subsequent input encoder process, so that there will be recognition obstacles when the preliminary image block is identified.

[0088] Step S402 , adjusting the size of the preliminary image block to a preset size by zero-filling to obtain a target image block.

[0089] In order to avoid the situation that the size may not match in the subsequent input process of the encoder, so that there is a recognition obstacle when the preliminary image block is recognized, according to some embodiments provided by the present invention, it is necessary to fill the size of the target image block after the segmentation process. There are many methods for filling the size of the target image block after the segmentation process, including but not limited to zero value filling, repeated filling (filling to be consistent with the pixels around the target image block), and constant filling (specifying a certain constant for filling). In some preferred embodiments of the present invention, the target image block after the segmentation process can be adjusted to a preset size for generation by zero value filling, that is, by zero value filling, so that the image input size of the encoder is corresponding to the image input size of the encoder in the subsequent input process of the encoder. It should be understood that the preset size refers to a pre-set size corresponding to the image input of the encoder. Since the pixel value of the filling is 0, the feature extraction after filling will not be affected, that is, the target image block obtained by segmentation is still strip-shaped, and such sampling will allow as many pixels of the same layer as possible to exist in a target image block. The strip cutting method allows the attention mechanism in the pre-order codec predictor Transformer to play a role, strengthens the neighborhood association in the image and improves the recognition accuracy of the image recognition process.

[0090] Reference Figure 5 According to some embodiments of the present invention, the medical image recognition method further includes:

[0091] Step S501, identifying an unlabeled preset data set by using a basic recognition model to obtain a pseudo-labeled data set, wherein the basic recognition model is obtained by pre-training an original preset model;

[0092] According to some embodiments provided by the present invention, the basic recognition model is obtained by pre-training the original preset model, wherein the original recognition model refers to an image recognition model that has not been pre-trained, and pre-training refers to preliminary training of the image recognition model based on a standard data set with preset labels, so that the image recognition model has a basic medical image recognition level. It should be understood that the basic medical image recognition level refers to the recognition level of a common medical image recognition model. In some embodiments of the present invention, pre-training needs to be completed through supervised training. Specifically, during the supervised training process, the original recognition model is based on a standard data set with preset labels to guide the encoder-decoder to learn the features of medical images, so that the pre-trained basic recognition model can recognize the unlabeled preset data set to obtain a pseudo-label data set. It should be noted that supervised learning is like "learning based on labeled examples". The model is trained using labeled data, so they have been carefully labeled so that each example is associated with a specific category. By studying the features of various examples of each class, the model learns to generalize and can even classify data it has never seen. In order to apply this method, well-labeled data is therefore required, and the standard data set with preset labels as positive sample data can be used as the above-mentioned well-labeled data for pre-training.

[0093] It should be noted that the unlabeled preset data set is composed of unlabeled medical images in a preset training direction, wherein the preset training direction can be determined according to the application scenario of the pre-order codec predictor Transformer, for example: if the pre-order codec predictor Transformer needs to be used to detect retinal medical images, then the unlabeled preset data set is an unlabeled data set composed of retinal medical images; if the pre-order codec predictor Transformer needs to be used to detect lung medical images, then the unlabeled preset data set is an unlabeled data set composed of lung medical images. In some embodiments of the present invention, the pseudo-label data set is obtained by attaching classification labels to the unlabeled preset data set via the basic recognition model. Although the classification labels of the pseudo-label data set are not completely matched with the corresponding medical images relative to the standard data set with preset labels, they still have reference significance for optimizing the training process. Therefore, the pseudo-label data set can be used as negative sample data, and mixed with the standard data set with preset labels as positive sample data to form an input sample set. Through the reasonable expansion of the training data by the pseudo-label data set, the basic recognition model can be further trained with the expanded data set, thereby improving the recognition accuracy of the pre-order codec predictor Transformer.

[0094] Step S502, using a standard data set with preset labels as positive sample data and a pseudo-label data set as negative sample data, and mixing the positive sample data with the negative sample data to form an input sample set;

[0095] It should be noted that the standard data set with preset labels as positive sample data refers to a data set composed of medical images with preset labels, wherein the classification labels of the standard data set with preset labels completely match the corresponding medical images. In some embodiments of the present invention, the standard data set with preset labels generally uses an existing data set or a data set commonly used in cooperation with hospitals. The standard data set for retinal medical images includes but is not limited to the retinal OCT image data set Kaggle provided by Duke University Retinal OCT Images. It should be understood that the classification labels of the pseudo-label data set are not completely matched with the corresponding medical images relative to the standard data set with preset labels, but it still has reference significance for optimizing the training process. Therefore, the pseudo-label data set can be used as negative sample data and mixed with the standard data set with preset labels as positive sample data to form an input sample set. Through the reasonable expansion of the training data by the pseudo-label data set, the basic recognition model can be further trained with the expanded data set, thereby improving the medical image recognition accuracy of the pre-order codec predictor Transformer.

[0096] Step S503: Optimize and train the basic recognition model using the input sample set to obtain a preamble codec predictor Transformer.

[0097] It should be noted that the reason for using the input sample set to optimize the training of the basic recognition model is that, since the standard data set with preset labels in the input sample set is positive sample data and the pseudo-label data set is negative sample data, and the input sample set formed by the mixture of positive sample data and negative sample data can reasonably expand the training amount of the basic recognition model in the optimization training process, so that the medical image recognition ability of the basic recognition model can be more effectively trained. The preamble codec predictor Transformer obtained through optimization training will also have a more precise medical image recognition accuracy.

[0098] Reference Figure 6 According to some embodiments of the present invention, a standard data set with preset labels is used as positive sample data, a pseudo-label data set is used as negative sample data, and the positive sample data and the negative sample data are mixed to form an input sample set, including:

[0099] Step S601, mixing positive sample data and negative sample data to obtain mixed data;

[0100] It should be noted that the standard data set with preset labels refers to a data set composed of medical images with preset labels, wherein the classification labels of the standard data set with preset labels completely match the corresponding medical images; and the pseudo-label data set is obtained by attaching classification labels to the unlabeled preset data set through the basic recognition model. Although the classification labels of the pseudo-label data set are not completely matched with the corresponding medical images compared with the standard data set with preset labels, they still have reference significance for the optimization training process. Therefore, the positive sample data in the input sample set is sampled from the standard data set with preset labels, and the negative sample data in the input sample set is sampled from the pseudo-label data set. The input sample set mixed by the above method is used for optimization training, which can reasonably expand the training amount of the basic recognition model in the optimization training process, so that the medical image recognition ability of the basic recognition model can be trained more effectively, and the preamble codec predictor Transformer obtained through optimization training will also have a more accurate medical image recognition accuracy.

[0101] Step S602, adjusting the proportion of positive sample data and negative sample data in the mixed data;

[0102] According to some embodiments provided by the present invention, if the proportion of positive sample data in the mixed data is too small and the proportion of negative sample data is large, the training direction of the basic recognition model will be close to the negative sample data. The pseudo-label data set is obtained by attaching classification labels to the unlabeled preset data set through the basic recognition model, and the classification labels of the pseudo-label data set are not completely matched with the corresponding medical images. In contrast, the standard data set with preset labels refers to a data set composed of medical images with preset labels, wherein the classification labels of the standard data set with preset labels completely match the corresponding medical images. Therefore, the standard data set with preset labels is more suitable as the training direction of the basic recognition model than the pseudo-label data set. Therefore, the proportion of positive sample data in the mixed data should not be too small. In some embodiments of the present invention, before generating an input sample set based on the mixed data, it is necessary to first increase the proportion of positive sample data in the mixed data and reduce the proportion of negative sample data in the mixed data to form an input data set that helps to improve the recognition accuracy of the basic recognition model. It should be clear that the proportion of positive sample data in the mixed data refers to the ratio of positive sample data to the total amount of mixed data, and the proportion of negative sample data in the mixed data refers to the ratio of positive sample data to the total amount of mixed data.

[0103] Step S603: When the proportion of positive sample data in the mixed data is greater than the proportion of negative sample data in the mixed data, the adjusted mixed data is used as an input sample set.

[0104] According to some embodiments provided by the present invention, when the proportion of positive sample data in mixed data is greater than the proportion of negative sample data in mixed data, the adjusted mixed data is used as an input sample set, wherein the obtained input sample set can correctly guide the basic recognition model to optimize the training in the direction of the standard data set with preset labels, thereby improving the recognition accuracy of the basic recognition model. It should be noted that the proportion of positive sample data in mixed data is not the larger the better. If the proportion of positive sample data in the mixed data is too large and the proportion of negative sample data is too small, the recognition processing of the pseudo-label data set in the negative sample data by the basic recognition model will start from a lower level, and the model obtained by the optimized training will be difficult to resist the interference of noise data in complex detection situations, and the robustness is weak, which is not conducive to improving the recognition accuracy of the basic recognition model. Therefore, the proportion of positive sample data in mixed data should not be too large. Therefore, in some preferred embodiments of the present invention, before generating an input sample set based on mixed data, in addition to increasing the proportion of positive sample data in mixed data and reducing the proportion of negative sample data in mixed data, it is also necessary to control the proportion of positive sample data to negative sample data at an appropriate ratio, for example, the proportion of negative sample data in mixed data is 15%, and the pseudo-label data set in the negative sample data is inserted into the positive sample data by class interpolation, thereby reducing the hard negative cases that may appear in the basic recognition model and enhancing the robustness of the basic recognition model. It should be understood that the appropriate proportion of negative sample data in mixed data can be tested during the optimization training process.

[0105] Reference Figure 7 According to some embodiments of the present invention, the basic recognition model is optimized and trained using the input sample set to obtain a preamble codec predictor Transformer, including:

[0106] Step S701, inputting the input sample set into the basic recognition model to perform the first type of iterative training;

[0107] Step S702, after each round of first-type iterative training, calculating the recognition accuracy of the basic recognition model and updating the first-type parameters of the basic recognition model;

[0108] According to some embodiments provided by the present invention, the training process of the preamble codec predictor Transformer includes: inputting the input sample set into the basic recognition model for the first type of iterative training, and after each round of the first type of iterative training, calculating the recognition accuracy of the basic recognition model and updating the basic recognition model. Among them, the purpose of the first type of iterative training is to continuously optimize the medical image recognition ability of the basic recognition model. After each round of the first type of iterative training, the recognition accuracy of the basic recognition model needs to be calculated to clarify that the recognition accuracy of the basic recognition model is gradually improving during the optimization training process. It should be noted that in the process of optimization training, after each round of the first type of iterative training, the basic recognition model needs to be updated once, so that the basic recognition model can have better performance and improved recognition ability in the recognition processing of medical images in the next round of the first type of iterative training. In the first type of iterative training, the updating of the basic recognition model is mainly completed by adjusting the first type of parameters of the basic recognition model, and the first type of parameters are internal parameters related to the recognition accuracy rate within the basic recognition model.

[0109] Step S703 , counting the changes in the recognition accuracy in each round of the first type of iterative training, and when the recognition accuracy converges to a first constant value, a trained preamble codec predictor Transformer is obtained.

[0110] It should be noted that the first fixed value refers to: after several rounds of adjustment of the internal parameters of the basic recognition model, the convergence value of the medical image recognition accuracy of the basic recognition model. In some embodiments provided by the present invention, the medical image recognition accuracy of the basic recognition model will be stabilized within a certain error interval, and this fixed value is the first fixed value. For example, if the medical image recognition accuracy fluctuates in the range of 84% to 86%, the first fixed value can be considered to be 85%. It should be understood that the first fixed value is not an exact constant value, but a value that changes with training conditions. When the medical image recognition accuracy converges to the first fixed value, it is determined that the optimization training has achieved a better effect, and the first type of iterative training can be stopped, wherein the medical image recognition model obtained after stopping the first type of iterative training is the preamble codec predictor Transformer, and the preamble codec predictor Transformer obtained by the optimization training will have a more accurate medical image recognition accuracy.

[0111] Reference Figure 8 According to some embodiments of the present invention, the basic recognition model is optimized and trained using the input sample set to obtain a preamble codec predictor Transformer, further comprising:

[0112] Step S801, inputting the input sample set into the basic recognition model to perform the second type of iterative training;

[0113] Step S802, after each round of second-type iterative training, calculating the output value of the loss function of the basic recognition model and updating the second-type parameters of the basic recognition model;

[0114] According to some embodiments provided by the present invention, the training process of the preamble codec predictor Transformer includes: inputting the input sample set into the basic recognition model for second-type iterative training, and after each round of second-type iterative training, calculating the loss function output value of the basic recognition model and updating the basic recognition model. Among them, the purpose of the second type of iterative training is to continuously optimize the robustness and anti-interference ability of the basic recognition model. After each round of second-type iterative training, it is necessary to calculate the loss function output value of the basic recognition model to clarify that the loss function output value of the basic recognition model is gradually improving during the optimization training process. It should be noted that in the process of optimization training, after each round of second-type iterative training, the basic recognition model needs to be updated once, so that the basic recognition model can have better robustness performance and anti-interference ability in the recognition and processing of medical images in the next round of second-type iterative training. In the second type of iterative training, the updating of the basic recognition model is mainly completed by adjusting the second type of parameters of the basic recognition model, and the second type of parameters are internal parameters related to the output value of the loss function within the basic recognition model.

[0115] Step S803, counting the changes in the loss function output value in each round of the second type of iterative training, when the loss function output value converges to the second constant value, a trained preamble codec predictor Transformer is obtained.

[0116] According to some embodiments of the present invention, in addition to using the medical image recognition accuracy as a reference benchmark, the output value of the loss function can also be used as a reference benchmark to determine that the optimization training has achieved a better effect. Specifically, when the output value of the loss function converges to the second fixed value, the robustness of the basic recognition model is improved to a higher level, so that the optimization training is determined to have achieved a better effect, and the second type of iterative training can be stopped, wherein the medical image recognition model obtained after stopping the second type of iterative training is the pre-order codec predictor Transformer. It should be understood that the reference benchmark for determining that the optimization training has achieved a better effect can be selected from the convergence of the medical image recognition accuracy to the first fixed value, or the convergence of the loss function output value to the second fixed value, or a composite standard, that is, the convergence of the medical image recognition accuracy to the first fixed value and the convergence of the loss function output value to the second fixed value. It should be understood that the second fixed value is not an exact constant value, but also a value that changes with the training situation. It should be emphasized that the reference benchmark for determining that the optimization training has achieved a better effect in the embodiment of the present invention includes but is not limited to the above examples.

[0117] Fig.10The electronic device 1000 provided by an embodiment of the present invention is shown. The electronic device 1000 includes: a processor 1001, a memory 1002, and a computer program stored in the memory 1002 and executable on the processor 1001, and the computer program is used to execute the above-mentioned medical image recognition method when it is executed.

[0118] The processor 1001 and the memory 1002 may be connected via a bus or other means.

[0119] The memory 1002 is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs and non-transitory computer executable programs, such as the medical image recognition method described in the embodiment of the present invention. The processor 1001 implements the above-mentioned medical image recognition method by running the non-transitory software program and instructions stored in the memory 1002.

[0120] The memory 1002 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function. The data storage area may store and execute the above-mentioned medical image recognition method. In addition, the memory 1002 may include a high-speed random access memory 1002, and may also include a non-transitory memory 1002, such as at least one storage device storage device, a flash memory device or other non-transitory solid-state storage device. In some embodiments, the memory 1002 may optionally include a memory 1002 remotely arranged relative to the processor 1001, and these remote memories 1002 may be connected to the electronic device 1000 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0121] The non-transient software programs and instructions required to implement the above-mentioned medical image recognition method are stored in the memory 1002. When executed by one or more processors 1001, the above-mentioned medical image recognition method is executed, for example, Figure 1 Steps S101 to S104 of the method, Figure 2 Steps S201 to S202 of the method, Figure 3 Steps S301 to S302 of the method, Figure 4 Steps S401 to S402 of the method, Figure 5 Steps S501 to S503 of the method, Figure 6 Steps S601 to S603 of the method, Figure 7 Steps S701 to S703 of the method, Figure 8 Method steps S801 to S802 in .

[0122] The embodiment of the present invention further provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the above-mentioned medical image recognition method.

[0123] In one embodiment, the computer readable storage medium stores computer executable instructions, which are executed by one or more control processors, for example, Figure 1 Steps S101 to S104 of the method, Figure 2 Steps S201 to S202 of the method, Figure 3 Steps S301 to S302 of the method, Figure 4 Steps S401 to S402 of the method, Figure 5 Steps S501 to S503 of the method, Figure 6 Steps S601 to S603 of the method, Figure 7 Steps S701 to S703 of the method, Figure 8 Method steps S801 to S802 in .

[0124] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, i.e., may be located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. It will be appreciated by those of ordinary skill in the art that all or some of the steps and systems in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, storage device storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. It should also be understood that the various implementations provided in the embodiments of the present invention can be arbitrarily combined to achieve different technical effects.

[0125] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions under the shared conditions without violating the spirit of the present invention. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A medical image recognition method, characterized in that: include: The unlabeled preset data set is recognized by a basic recognition model to obtain a pseudo-labeled data set; wherein the basic recognition model is obtained by preliminarily training the original preset model based on a standard data set with preset labels; Taking the standard data set as positive sample data and the pseudo-label data set as negative sample data, the positive sample data and the negative sample data are mixed to form an input sample set; Optimizing and training the basic recognition model using the input sample set to obtain a preamble codec predictor; Acquire an image to be identified, wherein the image to be identified includes a medical image of a target to be detected; Preliminarily determine the tissue structure feature of the target to be detected in the image to be identified, and obtain an image segmentation type that matches the tissue structure feature; wherein, when the tissue structure feature is a hierarchical distribution feature type, obtain the image segmentation type of strip sampling segmentation to define a segmentation sampling shape, wherein the length of the segmentation sampling shape is greater than the width, and the length direction of the segmentation sampling shape is consistent with the extension direction of the tissue structure hierarchy boundary line; Segmenting the image to be identified according to the image segmentation type to generate a target image block; Using the preamble codec predictor to perform feature extraction on the target image block to generate a feature extraction vector; The target to be detected in the image to be identified is identified based on the feature extraction vector to obtain an identification result.

2. The method according to claim 1, characterized in that The method of taking the standard data set as positive sample data and the pseudo-label data set as negative sample data, and mixing the positive sample data with the negative sample data to form an input sample set includes: Mixing the positive sample data and the negative sample data to obtain mixed data; Adjusting the proportions of the positive sample data and the negative sample data in the mixed data; When the proportion of the positive sample data in the mixed data is greater than the proportion of the negative sample data in the mixed data, the adjusted mixed data is used as the input sample set.

3. The method according to claim 1, characterized in that The step of optimizing and training the basic recognition model using the input sample set to obtain the preamble codec predictor comprises: Inputting the input sample set into the basic recognition model to perform a first type of iterative training; After each round of the first type of iterative training, the recognition accuracy of the basic recognition model is calculated and the first type of parameters of the basic recognition model are updated; The variation of the recognition accuracy rate in each round of the first type of iterative training is counted, and when the recognition accuracy rate converges to a first constant value, the trained preamble codec predictor is obtained.

4. The method according to any one of claims 1 to 3, characterized in that: The preliminarily determining the tissue structure feature of the target to be detected in the image to be identified and obtaining the image segmentation type matching the tissue structure feature includes: Obtaining pixel distribution in the image to be identified, and performing the preliminary determination on the feature type of the tissue structure feature according to the pixel distribution; The image segmentation type is acquired according to the feature type of the tissue structure feature.

5. The method according to claim 4, characterized in that The step of segmenting the image to be identified according to the image segmentation type to generate the target image block includes: Segmenting the image to be identified based on the segmentation sampling shape defined by the strip sampling segmentation to obtain a preliminary image block; The size of the preliminary image block is adjusted to a preset size by zero-filling to obtain the target image block.

6. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and the processor implements the medical image recognition method according to any one of claims 1 to 5 when executing the computer program.

7. A computer-readable storage medium, characterized in that: The storage medium stores a program, and the program is executed by a processor to implement the medical image recognition method according to any one of claims 1 to 5.

Images