Method and device for determining mitral valve prolapse and pathological segmentation based on echocardiographic images
By constructing a cardiac ultrasound image dataset and building a small-sample classification and pathological segmentation model, and using ResNet residual network for feature extraction, the problems of long processing time and low accuracy in medical image analysis were solved, and efficient mitral valve prolapse segmentation was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2023-03-31
- Publication Date
- 2026-06-09
AI Technical Summary
Existing medical image analysis and processing methods are time-consuming, inefficient, and have low accuracy in medical image interpretation.
By acquiring echocardiogram images, a dataset was constructed and denoised. A few-shot classification model and a pathological segmentation neural model were built. Features were extracted and classified and segmented using a few-shot learning network. A ResNet residual network was then used for feature extraction and pathological segmentation.
It achieves automated mitral valve prolapse detection and segmentation with short analysis and processing time, high efficiency, and high accuracy in medical image judgment.
Smart Images
Figure CN116563531B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical image analysis technology, and in particular to a method and device for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images. Background Technology
[0002] In current clinical practice and scientific research, medical images constitute the vast majority of medical images. These medical images mainly come from medical imaging equipment such as CT, MRI, and ultrasound. Many human pathological judgments require description and detection through these images so that doctors can make accurate diagnoses and treat patients.
[0003] Deep learning methods, as a highly effective approach in image information processing, have been widely used in numerous studies for image classification and segmentation. Many researchers have also applied its characteristics to medical image analysis, hoping to extract relevant functionalities from deep learning that are beneficial to medical image processing. Currently, due to limitations in deep learning algorithms for mitral valve prolapse in echocardiographic images, a sufficient amount of labeled data and appropriate model training are required to achieve high-precision semantic segmentation. Furthermore, the acquisition of medical images is limited, requiring specific medical imaging equipment. Existing medical image analysis processes are time-consuming, inefficient, and have low accuracy. Summary of the Invention
[0004] This application provides a method and apparatus for judging mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, in order to solve the problems of long processing time, slow efficiency, and low accuracy of existing medical image analysis.
[0005] In a first aspect, embodiments of this application provide a method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, the method comprising the following steps:
[0006] Acquire cardiac ultrasound images to build a cardiac ultrasound image dataset;
[0007] Denoising and image processing are performed on each image in the cardiac ultrasound image dataset;
[0008] Based on the denoised and preprocessed images, a few-shot classification model is trained and tested.
[0009] The existing cardiac ultrasound images are classified according to the small sample classification model to obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and then the images are processed.
[0010] Based on the aforementioned mitral valve prolapse image dataset, a pathological segmentation neural model was input, trained, and validated.
[0011] The medical images of the mitral valve and the images of mitral valve prolapse from the small sample classification model are input into the pathological segmentation neural model to perform preliminary verification of the pathological segmentation neural model and obtain the classification-segmentation model.
[0012] New medical ultrasound images are sequentially input into the classification-segmentation model to detect whether classification and pathological segmentation have been completed.
[0013] Furthermore, the image processing includes:
[0014] Perform image flipping, rotation, cropping, scaling, translation, and shaking operations on the image.
[0015] Furthermore, before training and testing the few-shot classification model based on the denoised and preprocessed images, the following steps are included:
[0016] A few-shot classification model is constructed, wherein the few-shot classification model includes an image feature convolutional neural network, a support set, and a query set.
[0017] Furthermore, the training and testing of the few-shot classification model based on the denoised and preprocessed images includes:
[0018] The images, after noise reduction and preprocessing, are input into a convolutional neural network for image feature extraction. The extracted image features are then input into a support set to obtain category prototypes, which are then input into a query set for cosine similarity calculation to obtain classification, thereby classifying medical images of the mitral valve and images of mitral valve prolapse.
[0019] Furthermore, before inputting the medical images of the mitral valve and the images of mitral valve prolapse from the small-sample classification model into the pathological segmentation neural model for preliminary verification of the pathological segmentation neural model, the process further includes:
[0020] A neural model for pathological segmentation is constructed, comprising a dual convolutional layer module, four downsampling modules, four upsampling modules, and a single convolutional layer.
[0021] Furthermore, the step of inputting, training, and validating the pathological segmentation neural model based on the mitral valve prolapse image dataset includes:
[0022] A pathological segmentation neural model was trained using 70% of the aforementioned mitral valve prolapse image dataset. The performance of the pathological segmentation neural model was then verified using the remaining 30% of the aforementioned mitral valve prolapse image dataset. The internal parameters of the network were fine-tuned based on the network performance, and the weights of the pathological segmentation neural model were updated using the backpropagation algorithm.
[0023] Furthermore, the acquisition of cardiac ultrasound images to construct a cardiac ultrasound image dataset includes:
[0024] Based on existing open-source cardiac ultrasound image datasets and cardiac ultrasound images provided by hospitals, a cardiac ultrasound image dataset was constructed by integrating the large-scale open-source ImageNet dataset.
[0025] In a second aspect, embodiments of this application also provide a device for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, comprising:
[0026] The image acquisition module is used to acquire cardiac ultrasound images to build a cardiac ultrasound image dataset;
[0027] The image processing module is used to perform noise reduction and image processing on each image in the cardiac ultrasound image dataset.
[0028] The first training module is used to train and test a few-shot classification model based on the denoised and preprocessed images.
[0029] The image classification module is used to classify existing cardiac ultrasound images according to the small sample classification model, obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and perform image processing.
[0030] The second training module is used to input, train, and validate the pathological segmentation neural model based on the mitral valve prolapse image dataset.
[0031] The model validation module is used to input the medical images of the mitral valve and the mitral valve prolapse images of the small sample classification model into the pathological segmentation neural model, so as to perform preliminary validation of the pathological segmentation neural model and obtain the classification-segmentation model.
[0032] The model detection module is used to input new medical ultrasound images sequentially into the classification-segmentation model to detect whether classification and pathological segmentation have been completed.
[0033] In a third aspect, embodiments of this application also provide a computer device, including: a memory and one or more processors;
[0034] The memory is used to store one or more programs;
[0035] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-described method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images.
[0036] In a fourth aspect, embodiments of this application also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images as described above.
[0037] This application embodiment acquires cardiac ultrasound images to construct a cardiac ultrasound image dataset; performs noise reduction and image processing on each image in the cardiac ultrasound image dataset; trains and tests a few-shot classification model based on the denoised and preprocessed images; classifies existing cardiac ultrasound images according to the few-shot classification model to obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and performs image processing on them; inputs, trains, and validates a pathological segmentation neural model based on the mitral valve prolapse image dataset; inputs the mitral valve medical images and mitral valve prolapse images from the few-shot classification model into the pathological segmentation neural model to perform preliminary validation of the pathological segmentation neural model, obtaining a classification-segmentation model; inputs new medical ultrasound images sequentially into the classification-segmentation model to detect whether classification and pathological segmentation are completed; achieves automated mitral valve prolapse judgment and segmentation of cardiac ultrasound images, with short analysis and processing time, high efficiency, and high accuracy in medical image judgment. Attached Figure Description
[0038] Figure 1 This is a flowchart of a method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, provided in an embodiment of this application.
[0039] Figure 2 This is a schematic diagram of a device for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, provided in an embodiment of this application.
[0040] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0041] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.
[0042] This application establishes a method for judging mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, which solves the problems of long processing time, slow efficiency, and low accuracy of existing medical image analysis.
[0043] The method for determining mitral valve prolapse and pathological segmentation based on echocardiographic images provided in this embodiment can be executed by a device for determining mitral valve prolapse and pathological segmentation based on echocardiographic images. This device can be implemented through software and / or hardware and integrated into a device for determining mitral valve prolapse and pathological segmentation based on echocardiographic images. The device for determining mitral valve prolapse and pathological segmentation based on echocardiographic images can be a computer or similar equipment.
[0044] Figure 1 This document provides a flowchart of a method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, as exemplified in this application. (Reference) Figure 1 The method includes the following steps:
[0045] Step 110: Acquire cardiac ultrasound images to build a cardiac ultrasound image dataset.
[0046] Specifically, based on existing open-source cardiac ultrasound image datasets and cardiac ultrasound images provided by hospitals, a cardiac ultrasound image dataset was constructed by integrating the large-scale open-source ImageNet dataset.
[0047] Step 120: Denoise and image processing are performed on each image in the cardiac ultrasound image dataset.
[0048] For example, the image processing includes performing operations such as image flipping, rotation, cropping, scaling, translation, and shaking on the image to make the trained pathological segmentation model more effective and accurate.
[0049] Step 130: Based on the denoised and preprocessed images, train and test the few-shot classification model.
[0050] Optionally, a few-shot classification model can be built, and trained and tested based on the denoised and preprocessed images. The few-shot classification model includes an image feature convolutional neural network, a support set, and a query set.
[0051] Specifically, the images, after noise reduction and preprocessing, are input into an image feature convolutional neural network for image feature extraction. The extracted image features are then input into a support set to obtain category prototypes, which are then input into a query set for cosine similarity calculation to obtain classification, thereby classifying medical images of the mitral valve and images of mitral valve prolapse.
[0052] For example, the image feature convolutional neural network is sequentially linked as one convolutional layer, one max pooling layer, eight residual network modules, one global average pooling layer, and one fully connected layer, specifically:
[0053] Convolutional layer c1 has a kernel size of 5*5, a stride of 1, uses BN normalization, employs ReLU activation function, has 64 kernels, and uses zero padding to fill the outer layer.
[0054] The max pooling layer has a window size of 3*3 and a stride of 2. It uses a 3*3 convolutional kernel to generate 64 new feature maps and uses BN normalization and ReLU function as activation functions.
[0055] Eight residual network modules are used, each with two convolutional layers. The first residual network module uses two 3x3 kernels (64 kernels total) with a stride of 1, an outer layer padded with zeros, and normalized using Batch Normalization (BN). The activation function is ReLU. The second residual network module uses two 3x3 kernels (64 kernels total) with a stride of 1, an outer layer padded with zeros, and normalized using BN. The activation function is ReLU. The third residual network module uses two 3x3 kernels (128 kernels total). The first convolutional layer has a stride of 2, an outer layer padded with zeros, and normalized using BN. The activation function is ReLU. The second convolutional layer has a stride of 1, and other parameters remain unchanged. The third residual network module includes a downsampling layer that directly passes the data output from the second residual network module to its output. The downsampling layer uses the same number of convolutional kernels as the residual network module it's currently in, i.e., the third layer, which has 128 kernels. It uses a stride of 2 and employs Batch Normalization (BN) to ensure that the downsampled feature map from the second residual module is the same size as the output feature map from the third module. The fourth residual network module uses two convolutional layers, each with 128 3x3 kernels. Both layers have a stride of 1, are padded with zeros on the outer layer, and are normalized using BN with the ReLU activation function. (This last sentence is a repetition of the previous one and can be omitted.) The fifth residual network module has a similar structure to the third layer, with 256 3x3 convolutional kernels. The first convolutional layer has a stride of 2, uses zero-padding on the outermost layer, and employs Batch Normalization (BN) with ReLU activation. The second convolutional layer has a stride of 1, and other settings remain unchanged. The sixth module uses 256 3x3 convolutional kernels, with other settings unchanged. The seventh module is similar to the third and fifth, with 512 convolutional kernels, and the downsampling layer also uses 512 convolutional kernels. The eighth module uses 512 3x3 convolutional kernels, with other settings unchanged.
[0056] Global average pooling layer.
[0057] A linear layer is used to return the number of categories for feature extraction.
[0058] For example, the few-shot learning network, based on the convolutional neural network for extracting image features, needs to provide a query set and a support set, and calculates the features to obtain the few-shot classification learning network. The specific operation is as follows:
[0059] The support set is constructed using labeled echocardiogram images of mitral valve prolapse and non-prolapse. The query set is queried using large, general image datasets such as ImageNet, and unlabeled echocardiogram images of the mitral valve are also included as a complete query set.
[0060] By calculating the features of the support set and the query set, the category prototype of the support set is obtained, and the categories in the query set are classified by calculating the cosine similarity. Finally, a classification network is obtained that can classify mitral valve prolapse or non-prolapse in echocardiographic images.
[0061] Step 140: Classify the existing cardiac ultrasound images according to the small sample classification model to obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and perform image processing.
[0062] For example, the image processing includes performing operations such as image flipping, rotation, cropping, scaling, translation, and shaking on the image to make the trained pathological segmentation model more effective and accurate.
[0063] Specifically, an image classification network trained on a small number of medical images is used to classify existing cardiac ultrasound images to obtain a mitral valve medical image dataset. The labeled data is then integrated with the newly obtained data as a new dataset. When a new ultrasound image is input, the image classification network can automatically identify whether to include it in the mitral valve medical image dataset, as this image is unlabeled.
[0064] Step 150: Input, train, and validate the pathological segmentation neural model based on the mitral valve prolapse image dataset.
[0065] Optionally, a pathological segmentation neural model is constructed, and the model is input, trained, and validated based on the aforementioned mitral valve prolapse image dataset. The pathological segmentation neural model comprises one double convolutional layer module, four downsampling modules, four upsampling modules, and one single convolutional layer.
[0066] Specifically, 70% of the aforementioned mitral valve prolapse image dataset is used as the training set to train a pathological segmentation neural model. The remaining 30% of the mitral valve prolapse image dataset is then used to verify the performance of the pathological segmentation neural model. The network parameters are fine-tuned based on the network performance, and the weights of the pathological segmentation neural model are updated using the backpropagation algorithm.
[0067] For example, the dual-layer convolutional module used includes two convolutional layers, using 3*3 convolutional kernels, 64 convolutional kernels, both using BN normalization and ReLU activation function, and both filled with a layer of all zeros.
[0068] The downsampling modules used contain one max pooling layer and two convolutional layers. The two convolutional layers in the first module are configured with 128 3x3 convolutional kernels; the two convolutional layers in the second module are configured with 256 3x3 convolutional kernels; the two convolutional layers in the third module are configured with 512 3x3 convolutional kernels; and the two convolutional layers in the fourth module are configured with 1024 3x3 convolutional kernels.
[0069] The upsampling module used consists of one upsampling layer and two convolutional layers. The upsampling layer has a magnification factor of 2, a bilinear mode, and the input corner pixels are aligned with the output tensor. Both convolutional layers use 3x3 kernels: the first upsampling layer uses 512 kernels, the second uses 256 kernels, the third uses 128 kernels, and the fourth uses 64 kernels.
[0070] The parameters for the single-layer convolutional layer used are: a 1x1 kernel, with a maximum of 1 kernel. This is used for displaying pixels in the segmented image region.
[0071] Step 160: Input the medical images of the mitral valve and the images of mitral valve prolapse from the small sample classification model into the pathological segmentation neural model to perform preliminary verification of the pathological segmentation neural model and obtain the classification-segmentation model.
[0072] Select mitral valve ultrasound images (unlabeled) classified by a small sample classification model and existing mitral valve pathological ultrasound images that need to be segmented and input them into the pathological segmentation neural model to verify the pathological segmentation ability of the neural model.
[0073] Step 170: Input the new medical ultrasound images into the classification-segmentation model in sequence to detect whether classification and pathological segmentation have been completed.
[0074] The above-described embodiments of this application provide a method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images. First, a few-shot classification learning network is used to classify input medical cardiac ultrasound images, separating out relevant medical ultrasound images containing mitral valve information. Then, the classified images are input into a segmentation network to achieve automatic pathological segmentation. The few-shot learning network uses a ResNet residual network to extract feature structures. The purpose is twofold: first, to address the problem of gradient vanishing due to excessive network layers, allowing for more layer stacking and more accurate feature extraction; second, by learning features from a large number of general images, a feature learner is obtained that can be applied to the classification of a small number of samples. By simultaneously extracting features from support set images and comparing them with those from the query set images, accurate classification can be achieved with a small number of samples, overcoming the drawback of insufficient medical image samples. Finally, the method addresses the pathological segmentation problem of mitral valve prolapse. This segmentation model is a lightweight segmentation model. It is trained on medically labeled images of mitral valve prolapse to obtain a deep learning network that can automatically segment mitral valve prolapse pathology. Combined with a few-shot learning classification network, it can easily realize a streamlined "classification-segmentation" cardiac mitral valve prolapse segmentation system, reducing the workload of doctors and greatly improving their work efficiency.
[0075] Based on the above embodiments, please refer to Figure 2 This application provides a device for judging mitral valve prolapse and pathological segmentation based on cardiac ultrasound images. The device specifically includes: an image acquisition module 201, an image processing module 202, a first training module 203, an image classification module 204, a second training module 205, a model verification module 206, and a model detection module 207.
[0076] The system comprises the following modules: an image acquisition module 201 for acquiring echocardiogram images to construct an echocardiogram image dataset; an image processing module 202 for denoising and processing each image in the echocardiogram image dataset; a first training module 203 for training and testing a small-sample classification model based on the denoised and preprocessed images; an image classification module 204 for classifying existing echocardiogram images according to the small-sample classification model to obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and performing image processing; a second training module 205 for inputting, training, and validating a pathological segmentation neural model based on the mitral valve prolapse image dataset; a model validation module 206 for inputting the mitral valve medical images and mitral valve prolapse images from the small-sample classification model into the pathological segmentation neural model to perform preliminary validation of the pathological segmentation neural model and obtain a classification-segmentation model; and a model detection module 207 for sequentially inputting new medical ultrasound images into the classification-segmentation model to detect whether classification and pathological segmentation have been completed.
[0077] The above describes a process involving: acquiring echocardiogram images to construct an echocardiogram image dataset; performing noise reduction and image processing on each image in the dataset; training and testing a few-shot classification model based on the denoised and preprocessed images; classifying existing echocardiogram images using the few-shot classification model to obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and performing image processing on these datasets; inputting, training, and validating a pathological segmentation neural model based on the mitral valve prolapse image dataset; inputting the mitral valve medical images and mitral valve prolapse images from the few-shot classification model into the pathological segmentation neural model for preliminary validation, resulting in a classification-segmentation model; and sequentially inputting new medical ultrasound images into the classification-segmentation model to detect whether classification and pathological segmentation are complete. This process achieves automated mitral valve prolapse judgment and segmentation of echocardiogram images, with short processing time, high efficiency, and high accuracy in medical image judgment.
[0078] The device for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in this application embodiment can be used to execute the method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in the above embodiment, and has corresponding functions and beneficial effects.
[0079] This application also provides a computer device that can integrate the mitral valve prolapse and pathological segmentation device based on cardiac ultrasound images provided in this application. Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. (Reference) Figure 3The computer device includes an input device 53, an output device 54, a memory 52, and one or more processors 51. The memory 52 stores one or more programs. When the one or more programs are executed by the one or more processors 51, the one or more processors 51 implement the method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in the above embodiments. The input device 53, output device 54, memory 52, and processors 51 can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.
[0080] The processor 51 executes various functional applications and data processing of the device by running software programs, instructions and modules stored in the memory 52, thereby realizing the above-mentioned method for judging mitral valve prolapse and pathological segmentation based on cardiac ultrasound images.
[0081] The computer equipment provided above can be used to execute the method for judging mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in the above embodiments, and has the corresponding functions and beneficial effects.
[0082] This application embodiment also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to execute a method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images. This method includes: acquiring cardiac ultrasound images to construct a cardiac ultrasound image dataset; performing noise reduction and image processing on each image in the cardiac ultrasound image dataset; training and testing a small-sample classification model based on the denoised and preprocessed images; classifying existing cardiac ultrasound images according to the small-sample classification model to obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and performing image processing; inputting, training, and validating a pathological segmentation neural model based on the mitral valve prolapse image dataset; inputting the mitral valve medical images and mitral valve prolapse images from the small-sample classification model into the pathological segmentation neural model to perform preliminary validation of the pathological segmentation neural model, obtaining a classification-segmentation model; and sequentially inputting new medical ultrasound images into the classification-segmentation model to detect whether classification and pathological segmentation are completed.
[0083] Storage medium – any type of memory device or storage device. The term “storage medium” is intended to include: mounting media, such as CD-ROM, floppy disk, or magnetic tape devices; computer device memory or random access memory, such as DRAM, DDRRAM, SRAM, EDORAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. Storage medium may also include other types of memory or combinations thereof. Furthermore, storage medium may reside in a first computer device in which a program is executed, or it may reside in a different second computer device connected to the first computer device via a network (such as the Internet). The second computer device may provide program instructions to the first computer for execution. The term “storage medium” may include two or more storage media that may reside in different locations (e.g., in different computer devices connected via a network). Storage medium may store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors.
[0084] Of course, the computer-executable instructions provided in the embodiments of this application are not limited to the method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images as described above, but can also perform related operations in the method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in any embodiment of this application.
[0085] The device, storage medium, and computer equipment for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in the above embodiments can execute the method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in any embodiment of this application. For technical details not described in detail in the above embodiments, please refer to the method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images provided in any embodiment of this application.
[0086] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.
Claims
1. A method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, characterized in that, The method includes the following steps: Acquire cardiac ultrasound images to build a cardiac ultrasound image dataset; Denoising and image processing are performed on each image in the cardiac ultrasound image dataset; Based on the denoised and preprocessed images, a few-shot classification model is trained and tested. The existing cardiac ultrasound images are classified according to the small sample classification model to obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and then the images are processed. Based on the aforementioned mitral valve prolapse image dataset, a pathological segmentation neural model was input, trained, and validated. The medical images of the mitral valve and the images of mitral valve prolapse from the small sample classification model are input into the pathological segmentation neural model to perform preliminary verification of the pathological segmentation neural model and obtain the classification-segmentation model. New medical ultrasound images are sequentially input into the classification-segmentation model to detect whether classification and pathological segmentation have been completed; Before training and testing the few-shot classification model based on the denoised and preprocessed images, the following steps are included: A few-shot classification model is constructed, wherein the few-shot classification model includes an image feature convolutional neural network, a support set, and a query set; The training and testing of a few-shot classification model based on the denoised and preprocessed images includes: The images, after noise reduction and preprocessing, are input into a convolutional neural network for image feature extraction. The extracted image features are then input into a support set to obtain category prototypes. These prototypes are then input into a query set for cosine similarity calculation to obtain classification, thereby classifying medical images of the mitral valve and images of mitral valve prolapse. Before inputting the medical images of the mitral valve and the images of mitral valve prolapse from the small-sample classification model into the pathological segmentation neural model for preliminary verification of the pathological segmentation neural model, the process further includes: A neural model for pathological segmentation is constructed, comprising a dual convolutional layer module, four downsampling modules, four upsampling modules, and a single convolutional layer.
2. The method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images according to claim 1, characterized in that, The image processing includes: Perform image flipping, rotation, cropping, scaling, translation, and shaking operations on the image.
3. The method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images according to claim 1, characterized in that, The step of inputting, training, and validating the pathological segmentation neural model based on the mitral valve prolapse image dataset includes: A pathological segmentation neural model was trained using 70% of the aforementioned mitral valve prolapse image dataset. The performance of the pathological segmentation neural model was then verified using the remaining 30% of the aforementioned mitral valve prolapse image dataset. The internal parameters of the network were fine-tuned based on the network performance, and the weights of the pathological segmentation neural model were updated using the backpropagation algorithm.
4. The method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images according to claim 1, characterized in that, The acquisition of cardiac ultrasound images to construct a cardiac ultrasound image dataset includes: Based on existing open-source cardiac ultrasound image datasets and cardiac ultrasound images provided by hospitals, a cardiac ultrasound image dataset was constructed by integrating the large-scale open-source ImageNet dataset.
5. A device for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images, implemented based on the method described in claim 1, characterized in that, include: The image acquisition module is used to acquire cardiac ultrasound images to build a cardiac ultrasound image dataset; The image processing module is used to perform noise reduction and image processing on each image in the cardiac ultrasound image dataset. The first training module is used to train and test a few-shot classification model based on the denoised and preprocessed images. The image classification module is used to classify existing cardiac ultrasound images according to the small sample classification model, obtain a mitral valve medical image dataset and a mitral valve prolapse image dataset, and perform image processing. The second training module is used to input, train, and validate the pathological segmentation neural model based on the mitral valve prolapse image dataset. The model validation module is used to input the medical images of the mitral valve and the mitral valve prolapse images of the small sample classification model into the pathological segmentation neural model, so as to perform preliminary validation of the pathological segmentation neural model and obtain the classification-segmentation model. The model detection module is used to input new medical ultrasound images sequentially into the classification-segmentation model to detect whether classification and pathological segmentation have been completed.
6. A computer device, characterized in that, include: Memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images as described in any one of claims 1-4.
7. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform a method for determining mitral valve prolapse and pathological segmentation based on cardiac ultrasound images as described in any one of claims 1-4.