Medical image recognition method, device, equipment and storage medium

Abstract

The present invention provides a medical image recognition method, device, equipment and storage medium, belonging to the field of image recognition technology, and the method includes: segmenting a medical image based on an edgeless active contour segmentation model to obtain a lesion contour curve; determining a first training image and a second training image based on the lesion contour curve, the first training image only including a lesion area within the lesion contour curve, and the second training image including a lesion area within the lesion contour curve and an associated area adjacent to the lesion area outside the lesion contour curve; training a medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model; and performing medical image recognition on a medical image to be recognized based on the target medical image recognition model. The present invention improves the accuracy of medical image recognition by setting an edgeless active contour segmentation model and training the medical image recognition model based on the second training image and the first training image.

Description

Technical Field

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

[0002] With the development of computer image processing technology, computer-aided diagnosis systems have been widely used in medical image processing. They can realize classification based on medical images and have important practical significance and clinical research value.

[0003] However, due to the differences in lesion location, size, and edge fuzziness of many lesion areas, the medical image recognition models in the prior art often encounter difficulties in accurately locating the lesion area and distinguishing the boundary between the normal area and the lesion area. Secondly, the lesion areas of different types of lesions may share similar apparent features, further increasing the difficulty of accurately classifying the lesion areas. In other words, the medical image recognition technology based on deep learning in the prior art has demonstrated certain recognition capabilities, but in actual applications, the recognition accuracy is low.

[0004] There is an urgent need to provide a medical image recognition method, device, equipment and storage medium to accurately distinguish normal areas from lesion areas, and to accurately distinguish lesion types with similar appearance characteristics, so as to improve the recognition accuracy of medical images. Summary of the invention

[0005] In view of this, it is necessary to provide a medical image recognition method, device, equipment and storage medium to solve the problems existing in the prior art that normal areas and lesion areas cannot be accurately distinguished, and lesion types with similar apparent characteristics cannot be accurately distinguished, which in turn leads to the technical problem of low recognition accuracy of medical images.

[0006] On the one hand, in order to solve the above technical problems, the present invention provides a medical image recognition method, comprising:

[0007] Segment the medical image based on the edgeless active contour segmentation model to obtain the lesion contour curve;

[0008] Determine a first training image and a second training image based on the lesion contour curve, wherein the first training image only includes the lesion area within the lesion contour curve, and the second training image includes the lesion area within the lesion contour curve and an associated area outside the lesion contour curve and adjacent to the lesion area;

[0009] Training a medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model;

[0010] Medical image recognition is performed on the medical image to be recognized based on the target medical image recognition model.

[0011] In a possible implementation, determining the first training image and the second training image based on the lesion contour curve includes:

[0012] Using the image within the lesion contour curve as the first training image;

[0013] The minimum rectangular enclosing area of ​​the lesion contour curve is determined, and the minimum rectangular enclosing area is used as the second training image.

[0014] In a possible implementation, the edgeless active contour segmentation model is:

[0015]

[0016] In the formula, is the energy functional of the lesion image within the lesion contour curve; is the level set function; is the Heaviside function; is the average gray value of the image within the lesion contour curve; is the average gray value of the image outside the lesion contour curve; is the image domain; is the first coefficient; is the second coefficient; is the third coefficient; is the gradient symbol.

[0017] In a possible implementation, the medical image recognition model includes a first feature extraction module, a second feature extraction module, a feature splicing module and a prediction module;

[0018] The first feature extraction module is used to extract the first image feature of the first training image;

[0019] The second feature extraction module is used to extract the second image feature of the second training image;

[0020] The feature stitching module is used to stitch the first image feature and the second image feature to obtain a stitching feature;

[0021] The prediction module is used to predict the splicing features to obtain a medical image recognition result.

[0022] In a possible implementation, the first feature extraction module includes a first convolution layer, multiple moving flip bottleneck convolution modules, a second convolution layer and a pooling layer connected in sequence; the multiple moving flip bottleneck convolution modules include multiple first moving flip bottleneck convolution modules with a dilation rate of 1 and multiple second moving flip bottleneck convolution modules with a dilation rate of 6.

[0023] In one possible implementation, the mobile flip bottleneck convolution module includes a third convolution layer, a separable convolution layer, a coordinate attention mechanism layer, a fourth convolution layer, a normalization layer, a discard layer and a superposition layer connected in sequence, and the superposition layer is used to superimpose the output of the discard layer and the input of the third convolution layer.

[0024] In a possible implementation, the method further includes:

[0025] An initial medical image is acquired, and preprocessed to obtain the medical image, wherein the preprocessing includes bottom-hat operation, data enhancement and normalization processing.

[0026] On the other hand, the present invention also provides a medical image recognition device, comprising:

[0027] An image segmentation unit, used for segmenting the medical image based on an edgeless active contour segmentation model to obtain a lesion contour curve;

[0028] a training image determining unit, configured to determine a first training image and a second training image based on the lesion contour curve, wherein the first training image only includes a lesion area within the lesion contour curve, and the second training image includes the lesion area within the lesion contour curve and an associated area outside the lesion contour curve and adjacent to the lesion area;

[0029] A recognition model training unit, configured to train a medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model;

[0030] The medical image recognition unit is used to perform medical image recognition on the medical image to be recognized based on the target medical image recognition model.

[0031] On the other hand, the present invention also provides a medical image recognition device, including a memory and a processor, wherein:

[0032] The memory is used to store programs;

[0033] The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the medical image recognition method described in any one of the possible implementations above.

[0034] On the other hand, the present invention also provides a computer-readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the medical image recognition method described in any possible implementation method mentioned above are implemented.

[0035] The beneficial effects of the present invention are as follows: the medical image recognition method provided by the present invention takes into account that the boundary of the lesion area is usually fuzzy, and the difference between the internal structure of the lesion and the surrounding normal area may not be simply reflected in the edge strength. By segmenting the medical image based on the edgeless active contour segmentation model, it breaks through the dependence of the traditional edge detection-based method on clear and obvious boundary conditions, realizes the accurate segmentation of medical images with unclear boundaries, improves the segmentation accuracy of the lesion area and the normal area in the medical image, and thus improves the accuracy of medical image recognition. Secondly, it can also avoid the need for precise edge detection, and can better handle the boundary uncertainty problem caused by factors such as lighting changes and imaging equipment differences, thereby improving the robustness of medical image recognition.

[0036] Furthermore, the present invention obtains a target medical image recognition model by determining a first training image and a second training image based on the lesion contour curve, wherein the first training image is a lesion image within the lesion contour curve, and the second training image includes the lesion image and a surrounding associated image outside the lesion contour curve, that is, the medical image recognition model is trained by training a second training image that retains the interaction information between the lesion and the surrounding area and a first training image that only contains lesion information, thereby providing a more comprehensive image feature representation for the identification of the lesion area, thereby further improving the accuracy of medical image recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in 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 those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 A schematic diagram of a flow chart of an embodiment of a medical image recognition method provided by the present invention;

[0039] Figure 2 For the present invention Figure 1 A schematic flow chart of an embodiment of S102;

[0040] Figure 3 A schematic diagram of an embodiment of the first training image and the second training image provided by the present invention;

[0041] Figure 4 A schematic diagram of the structure of an embodiment of the medical image recognition model provided by the present invention;

[0042] Figure 5 A schematic diagram of the structure of an embodiment of the mobile flip bottleneck convolution module provided by the present invention;

[0043] Figure 6 A schematic diagram of the structure of an embodiment of a medical image recognition device provided by the present invention;

[0044] Figure 7 A schematic diagram of the structure of an embodiment of the medical image recognition device provided by the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0046] It should be understood that the schematic drawings are not drawn to scale. The flowchart used in the present invention shows the operations implemented according to some embodiments of the present invention. It should be understood that the operations of the flowchart can be implemented out of order, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art, under the guidance of the content of the present invention, can add one or more other operations to the flowchart, and can also remove one or more operations from the flowchart. Some of the block diagrams shown in the accompanying drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor systems and / or microcontroller systems.

[0047] Reference to an "embodiment" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0048] The present invention provides a medical image recognition method, apparatus, device and storage medium, which are described below respectively.

[0049] Figure 1A schematic flow chart of an embodiment of the medical image recognition method provided by the present invention is as follows: Figure 1 As shown, the medical image recognition method includes:

[0050] S101, segmenting the medical image based on an edgeless active contour segmentation model to obtain a lesion contour curve;

[0051] S102, determining a first training image and a second training image based on the lesion contour curve, wherein the first training image includes only the lesion area within the lesion contour curve, and the second training image includes the lesion area within the lesion contour curve and an associated area outside the lesion contour curve and adjacent to the lesion area;

[0052] S103, training the medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model;

[0053] S104. Perform medical image recognition on the medical image to be recognized based on the target medical image recognition model.

[0054] Among them, the number of medical images in step S101 should be multiple, and each medical image is processed by steps S101-S102 to train the medical image recognition model through multiple first training images and multiple second training images to improve the accuracy of the obtained target medical image recognition model.

[0055] The medical image may be an image obtained from a data set. Specifically, the medical image may be a skin image, that is, the type of skin lesions is identified.

[0056] In a specific embodiment of the present invention, the data set is the HAM10000 data set, which covers seven common skin diseases, namely: actinic keratosis (AKIEC), basal cell carcinoma (BCC), benign keratosis-like lesion (BKL), dermatofibroma (DF), melanoma (MEL), melanocytic nevus (NV) and vascular lesion (VASC).

[0057] Compared with the prior art, the medical image recognition method provided by the embodiment of the present invention takes into account that the boundary of the lesion area is usually fuzzy, and the difference between the internal structure of the lesion and the surrounding normal area may not be simply reflected in the edge strength. By segmenting the medical image based on the edgeless active contour segmentation model, it breaks through the dependence of the traditional edge detection-based method on clear and obvious boundary conditions, realizes the accurate segmentation of medical images with unclear boundaries, improves the segmentation accuracy of the lesion area and the normal area in the medical image, and thus improves the accuracy of medical image recognition. In addition, it can also avoid the need for precise edge detection, and can better handle the boundary uncertainty problem caused by factors such as lighting changes and imaging equipment differences, thereby improving the robustness of medical image recognition.

[0058] Furthermore, an embodiment of the present invention obtains a target medical image recognition model by determining a first training image and a second training image based on a lesion contour curve, and the first training image is a lesion image within the lesion contour curve, and the second training image includes the lesion image and a surrounding associated image outside the lesion contour curve, that is: by training the medical image recognition model with a second training image that retains the interaction information between the lesion and the surrounding area and a first training image that only contains lesion information, a more comprehensive image feature representation can be provided for the recognition of lesion areas in medical images, thereby further improving the accuracy of medical image recognition.

[0059] In some embodiments of the present invention, Figure 2 As shown, step S102 includes:

[0060] S201, using the image within the lesion contour curve as the first training image;

[0061] S202: Determine the minimum rectangular enclosing area of ​​the lesion contour curve, and use the minimum rectangular enclosing area as the second training image.

[0062] In a specific embodiment of the present invention, Figure 3 As shown in the figure, the area enclosed by the irregular solid line is the first training image, and the first training image and the shaded part together constitute the second training image. That is, the second training image includes the information of the area surrounding the lesion. When training the medical image recognition model, the lesion features and the interactive information between the lesion and the surrounding area can be integrated to improve the comprehensiveness of feature learning.

[0063] In a specific embodiment of the present invention, the edgeless active contour segmentation model is:

[0064]

[0065] In the formula, is the energy functional of the lesion image within the lesion contour curve; is the level set function, i.e., the lesion contour evolution curve; is the Heaviside function; is the average gray value of the image within the lesion contour curve; is the average gray value of the image outside the lesion contour curve; is the image domain; is the first coefficient; is the second coefficient; is the third coefficient; is the gradient symbol.

[0066] Step S101 specifically includes: determining the contour of the lesion with the goal of minimizing the energy functional, that is, determining the contour curve of the lesion.

[0067] In some embodiments of the present invention, Figure 4 As shown, the medical image recognition model includes a first feature extraction module, a second feature extraction module, a feature splicing module and a prediction module;

[0068] The first feature extraction module is used to extract the first image feature of the first training image;

[0069] The second feature extraction module is used to extract the second image feature of the second training image;

[0070] The feature stitching module is used to stitch the first image feature and the second image feature to obtain a stitching feature;

[0071] The prediction module is used to predict the splicing features and obtain the medical image recognition results.

[0072] The embodiment of the present invention sets two parallel first feature extraction modules and second feature extraction modules to respectively extract features from the first training image and the second training image, thereby maintaining the spatial structural integrity of the first training image and the second training image and increasing the number of channels of the medical image recognition model. The medical image recognition model can learn more cross-modal interaction information, provide a more comprehensive image feature representation for the prediction module, and improve the recognition accuracy of the medical image recognition model.

[0073] The prediction module includes a fully connected layer and an activation function layer connected in sequence, and the activation function of the activation function layer is softmax. The concatenated features are linearly transformed based on the fully connected layer, and the prediction probability of each type of lesion is output through the activation function layer to achieve accurate classification of lesion types in medical images.

[0074] It should be noted that the model structures of the first feature extraction module and the second feature extraction module are exactly the same, only the input images are different. The embodiment of the present invention is described with the first feature extraction module. Specifically, the first feature extraction module includes a first convolution layer, multiple moving flip bottleneck convolution modules, a second convolution layer and a pooling layer connected in sequence; the multiple moving flip bottleneck convolution modules include multiple first moving flip bottleneck convolution modules with a dilation rate of 1 and multiple second moving flip bottleneck convolution modules with a dilation rate of 6.

[0075] Among them, the second convolution module includes a second convolutional layer, a second pooling layer and a second fully connected layer which are connected in sequence.

[0076] In a specific embodiment of the present invention, as shown in Table 1, the first feature extraction module includes a first convolution layer with a convolution kernel of 3×3, a dilation rate of 1, 4 first moving flip bottleneck convolution modules with a convolution kernel of 3×3, and a dilation rate of 6, 51 second moving flip bottleneck convolution modules with a convolution kernel of 3×3.

[0077] Table 1 Model structure of the first feature extraction module

[0078]

[0079] In order to verify the superiority of the first feature extraction module proposed in the embodiment of the present invention, the embodiment of the present invention tested and compared its recognition performance with the common VGG16, ResNet50, DenseNet121 and the Inception ResNet v2 model combined with the soft attention mechanism (abbreviated as IRv2-SA). The test results are shown in Table 2:

[0080] Table 2 Test comparison between the first feature extraction module and other feature extraction modules

[0081]

[0082] It can be seen from Table 2 that the first feature extraction model in the embodiment of the present invention is superior to the other four common feature extraction models in terms of the five evaluation indicators of accuracy, precision, F1 score (F1), recall, and mean average precision (mAP). That is, the first feature extraction model proposed in the embodiment of the present invention has a higher recognition accuracy in medical image recognition.

[0083] In some embodiments of the present invention, Figure 5As shown in the figure, the mobile flip bottleneck convolution module includes a third convolution layer, a separable convolution layer, a coordinate attention mechanism layer (CA), a fourth convolution layer, a normalization layer, a discard layer and a stacking layer, which are connected in sequence. The stacking layer is used to stack the output of the discard layer and the input of the third convolution layer.

[0084] The embodiment of the present invention sets the attention mechanism in the moving flip bottleneck convolution module as a coordinate attention mechanism, which can effectively guide the model to dynamically adjust the degree of emphasis on different channel features. Especially when processing medical images, it can highlight the key features of the lesion site and reduce the impact of background noise, thereby improving the accuracy of judging the lesion type in the medical image.

[0085] The difference between the use of the attention mechanism in the embodiment of the present invention and the traditional technology is that in the traditional technology, the attention mechanism is usually applied to the entire medical image, and the learning effect is enhanced by giving greater weight to the local area of ​​the image. However, the embodiment of the present invention adopts a more sophisticated method: first, the medical image is segmented for lesions, and then feature extraction is performed on two different training images respectively. The extracted features are fused through concat, an operation that particularly emphasizes the information interaction between the lesion area and its surrounding associated areas. This strategy not only increases the attention to the lesion area, but also significantly improves the accuracy and reliability of recognition by enhancing information exchange between regions.

[0086] In order to verify the superiority of the mobile flip bottleneck convolution module using the coordinate attention (CA) mechanism provided by the embodiment of the present invention over the Squeeze-and-Excitation (SE) attention, the Efficient Channel Attention (ECA) and the Convolutional Block Attention Module (CBAM), the embodiment of the present invention compares the performance of the mobile flip bottleneck convolution module integrating the four mechanisms. The comparison test results are shown in Table 3:

[0087] Table 3 Comparative test results of mobile flip bottleneck convolution modules integrating different attention mechanisms

[0088]

[0089] It can be seen from Table 3 that compared with other attention mechanisms, the performance of the moving flip bottleneck convolution module fused with the coordinate attention mechanism is better. This is because the coordinate attention mechanism can finely model and optimize the spatial coordinate information of the input image. In medical image recognition, the location, shape and relationship of the lesion site with the surrounding tissue are often crucial to diagnosis. The coordinate attention mechanism strengthens the contrast and connection between the lesion area and the normal area by focusing on the feature correlation in the spatial coordinate dimension, so that the model can more keenly capture these key spatial distribution features, accurately focus on and strengthen the key features of the lesion area, and thus improve the overall performance of the model in disease classification and localization.

[0090] Since the diversity and accuracy of medical images used for training are crucial to improving the robustness and generalization ability of the target medical image recognition model, in some embodiments of the present invention, before step S101, the medical image recognition method further includes:

[0091] An initial medical image is acquired, and preprocessed to obtain the medical image, wherein the preprocessing includes preprocessing methods such as bottom hat operation, data enhancement and normalization processing.

[0092] The embodiments of the present invention can remove background noise and hair interference in the initial medical image through bottom-hat operation, make up for the problem of insufficient samples through data enhancement, and adjust the pixel values ​​of all medical images to a unified standard range through normalization processing, thereby reducing the impact of data range differences, thereby improving the accuracy and standardization of the obtained medical images, thereby improving the accuracy and reliability of the target medical image recognition model obtained through medical image training.

[0093] In a specific embodiment of the present invention, the bottom hat operation is specifically as follows: first, the initial medical image is processed by a closing operation. The closing operation is a sequence of operations that combines dilation and erosion, in which the image is first dilated and then the dilation result is eroded. This dual operation can effectively eliminate small holes, breaks, and bridges connecting adjacent bright areas in the image while maintaining the continuity of dark areas. The closing operation helps to smooth and fill the bright areas in the background while maintaining the integrity of the dark areas. Next, the difference between the initial medical image and the image processed by the closing operation is calculated. The resulting difference image is called a bottom hat image, which only contains those pixels whose brightness in the original image is significantly lower than that of their surrounding neighborhoods. The bottom hat image effectively suppresses background noise and over-brightness interference, thereby highlighting the detailed features of dark colors.

[0094] In a specific embodiment of the present invention, data enhancement specifically includes: processing the initial medical image using random horizontal flipping, rotation, scaling and other techniques. For categories with a small number of samples, a higher upper limit of data enhancement times is set to compensate for the lack of samples, and for categories with a large number of samples, only slight enhancement is performed to avoid over-amplifying the impact of a certain category.

[0095] In summary, the medical image recognition method proposed in the embodiment of the present invention extracts the features of medical images based on a feature extraction module fused with a coordinate attention mechanism, emphasizes the adaptive weighting of key feature channels, and enables the model to better focus on the lesion area in the context of complex medical images, thereby significantly improving the diagnostic accuracy.

[0096] Furthermore, in the segmentation method based on the edgeless active contour segmentation model, when processing medical images, the accuracy and completeness of the lesion boundary definition are crucial for accurate diagnosis. The method based on edgeless segmentation proposed in the embodiment of the present invention is different from traditional segmentation technology. It pays more attention to the complete preservation and natural transition of the internal structure information of the lesion, reducing the risk of misdiagnosis due to unclear boundaries.

[0097] Furthermore, taking into account the comprehensive characteristics of medical image recognition, the embodiment of the present invention also adopts an image feature fusion strategy of the first training image and the second training image, which deeply fuses the visual image information of the lesion after borderless segmentation with the information around the lesion, further enhancing the ability of the medical image recognition model in comprehensively understanding and diagnosing the type of lesions, thereby improving the recognition accuracy of medical images.

[0098] In order to better implement the medical image recognition method in the embodiment of the present invention, based on the medical image recognition method, the embodiment of the present invention also provides a medical image recognition device, such as Figure 6 As shown, the medical image recognition device 600 includes:

[0099] An image segmentation unit 601 is used to segment the medical image based on an edgeless active contour segmentation model to obtain a lesion contour curve;

[0100] A training image determining unit 602 is used to determine a first training image and a second training image based on the lesion contour curve, wherein the first training image only includes the lesion area within the lesion contour curve, and the second training image includes the lesion area within the lesion contour curve and an associated area outside the lesion contour curve and adjacent to the lesion area;

[0101] A recognition model training unit 603 is used to train the medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model;

[0102] The medical image recognition unit 604 is configured to perform medical image recognition on the medical image to be recognized based on the target medical image recognition model.

[0103] The medical image recognition device 600 provided in the above embodiment can implement the technical solution described in the above medical image recognition method embodiment. The specific implementation principles of the above modules or units can refer to the corresponding contents in the above medical image recognition method embodiment, which will not be repeated here.

[0104] like Figure 7 As shown, the present invention also provides a medical image recognition device 700. The medical image recognition device 700 includes a processor 701, a memory 702 and a display 703. Figure 7 Only some components of the medical image recognition apparatus 700 are shown, but it should be understood that it is not required to implement all of the shown components, and more or fewer components may be implemented instead.

[0105] In some embodiments, the processor 701 may be a central processing unit (CPU), a microprocessor or other data processing chip, and is used to run program codes or process data stored in the memory 702, such as the medical image recognition method of the present invention.

[0106] In some embodiments of the present invention, the processor 701 may be a single server or a server group. The server group may be centralized or distributed. In some embodiments, the processor 701 may be local or remote. In some embodiments, the processor 701 may be implemented in a cloud platform. In one embodiment, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination thereof.

[0107] In some embodiments, the memory 702 may be an internal storage unit of the medical image recognition device 700, such as a hard disk or memory of the medical image recognition device 700. In other embodiments, the memory 702 may also be an external storage device of the medical image recognition device 700, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the medical image recognition device 700.

[0108] Furthermore, the memory 702 may include both an internal storage unit of the medical image recognition device 700 and an external storage device. The memory 702 is used to store application software installed in the medical image recognition device 700 and various data.

[0109] In some embodiments, the display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 703 is used to display information of the medical image recognition device 700 and to display a visual user interface. The components 701-703 of the medical image recognition device 700 communicate with each other via a system bus.

[0110] In some embodiments of the present invention, when the processor 701 executes the medical image recognition program in the memory 702, the following steps may be implemented:

[0111] Segment the medical image based on the edgeless active contour segmentation model to obtain the lesion contour curve;

[0112] Determine a first training image and a second training image based on the lesion contour curve, wherein the first training image only includes the lesion area within the lesion contour curve, and the second training image includes the lesion area within the lesion contour curve and an associated area outside the lesion contour curve and adjacent to the lesion area;

[0113] Training the medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model;

[0114] Medical image recognition is performed on the medical image to be identified based on the target medical image recognition model.

[0115] It should be understood that: when the processor 701 executes the medical image recognition program in the memory 702, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiment above.

[0116] Furthermore, the embodiment of the present invention does not specifically limit the type of the medical image recognition device 700 mentioned. The medical image recognition device 700 may be a portable medical image recognition device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of portable medical image recognition devices include but are not limited to portable medical image recognition devices equipped with IOS, Android, Microsoft or other operating systems. The above-mentioned portable medical image recognition device may also be other portable medical image recognition devices. It should also be understood that in some other embodiments of the present invention, the medical image recognition device 700 may not be a portable medical image recognition device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0117] Correspondingly, an embodiment of the present invention also provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the program or instructions are executed by a processor, it can implement the steps or functions of the medical image recognition method provided by the above-mentioned method embodiments.

[0118] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0119] The above is a detailed introduction to a medical image recognition method, device, equipment and storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A medical image recognition method, characterized in that: include: Segment the medical image based on the edgeless active contour segmentation model to obtain the lesion contour curve; Determine a first training image and a second training image based on the lesion contour curve, wherein the first training image only includes the lesion area within the lesion contour curve, and the second training image includes the lesion area within the lesion contour curve and an associated area outside the lesion contour curve and adjacent to the lesion area; Training a medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model; Performing medical image recognition on the medical image to be recognized based on the target medical image recognition model; The edgeless active contour segmentation model is: In the formula, is the energy functional of the lesion image within the lesion contour curve; is the level set function; is the Heaviside function; is the average gray value of the image within the lesion contour curve; is the average gray value of the image outside the lesion contour curve; is the image domain; is the first coefficient; is the second coefficient; is the third coefficient; is the gradient symbol; The medical image recognition model includes a first feature extraction module, a second feature extraction module, a feature splicing module and a prediction module; The first feature extraction module is used to extract the first image feature of the first training image; The second feature extraction module is used to extract the second image feature of the second training image; The feature stitching module is used to stitch the first image feature and the second image feature to obtain a stitching feature; The prediction module is used to predict the splicing features to obtain a medical image recognition result.

2. The medical image recognition method according to claim 1, characterized in that: The determining of the first training image and the second training image based on the lesion contour curve comprises: Using the image within the lesion contour curve as the first training image; The minimum rectangular enclosing area of ​​the lesion contour curve is determined, and the minimum rectangular enclosing area is used as the second training image.

3. The medical image recognition method according to claim 1, characterized in that: The first feature extraction module includes a first convolution layer, multiple moving flip bottleneck convolution modules, a second convolution module and a pooling layer connected in sequence; the multiple moving flip bottleneck convolution modules include multiple first moving flip bottleneck convolution modules with a dilation rate of 1 and multiple second moving flip bottleneck convolution modules with a dilation rate of 6.

4. The medical image recognition method according to claim 3, characterized in that: The mobile flip bottleneck convolution module includes a third convolution layer, a separable convolution layer, a coordinate attention mechanism layer, a fourth convolution layer, a normalization layer, a discard layer and a superposition layer connected in sequence, and the superposition layer is used to superimpose the output of the discard layer and the input of the third convolution layer.

5. The medical image recognition method according to claim 1, characterized in that: The method further comprises: An initial medical image is acquired, and preprocessed to obtain the medical image, wherein the preprocessing includes bottom-hat operation, data enhancement and normalization processing.

6. A medical image recognition device, characterized in that: include: An image segmentation unit, used for segmenting the medical image based on an edgeless active contour segmentation model to obtain a lesion contour curve; a training image determining unit, configured to determine a first training image and a second training image based on the lesion contour curve, wherein the first training image only includes a lesion area within the lesion contour curve, and the second training image includes the lesion area within the lesion contour curve and an associated area outside the lesion contour curve and adjacent to the lesion area; A recognition model training unit, configured to train a medical image recognition model based on the first training image and the second training image to obtain a target medical image recognition model; A medical image recognition unit, configured to perform medical image recognition on the medical image to be recognized based on the target medical image recognition model; The edgeless active contour segmentation model is: In the formula, is the energy functional of the lesion image within the lesion contour curve; is the level set function; is the Heaviside function; is the average gray value of the image within the lesion contour curve; is the average gray value of the image outside the lesion contour curve; is the image domain; is the first coefficient; is the second coefficient; is the third coefficient; is the gradient symbol; The medical image recognition model includes a first feature extraction module, a second feature extraction module, a feature splicing module and a prediction module; The first feature extraction module is used to extract the first image feature of the first training image; The second feature extraction module is used to extract the second image feature of the second training image; The feature stitching module is used to stitch the first image feature and the second image feature to obtain a stitching feature; The prediction module is used to predict the splicing features to obtain a medical image recognition result.

7. A medical image recognition device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store programs; The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the medical image recognition method described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the medical image recognition method described in any one of claims 1 to 5 are implemented.

Images