Peripheral neuromorphological anomaly identification method, system and device, and storage medium

Abstract

The invention provides a peripheral neuromorphological anomaly recognition method, system and device and a storage medium, and the method comprises the steps: segmenting a frame image in an ultrasonic image sequence based on a three-dimensional deep learning model, and calculating corresponding neuromorphological information based on each segmentation mask, and analyzing continuous frames of images in the ultrasonic image sequence to obtain neuromorphic change information, and judging a morphological anomaly recognition result of the to-be-recognized peripheral nerves based on the neuromorphic change information. The three-dimensional deep learning model can process space and time dimensions at the same time, extract multi-scale spatial-temporal features in an ultrasonic image sequence and capture dynamic changes of inter-frame neural morphology, so that a neural region is segmented more accurately, the method adapts to conditions of complex anatomical structures and fuzzy boundaries, the segmentation precision is remarkably improved, and the neural region segmentation efficiency is improved. And particularly, high accuracy and robustness of morphological anomaly recognition can be kept under the condition of lesion.

Description

Technical Field

[0001] The present invention relates to the technical field of peripheral nerve identification, and in particular to a method, system, device and storage medium for identifying peripheral nerve morphological abnormalities. Background Art

[0002] Peripheral neuropathy (such as carpal tunnel syndrome and diabetic neuropathy) affects patients' motor and sensory functions, and early diagnosis is crucial to improving prognosis. Ultrasound imaging has become the preferred tool for peripheral nerve diagnosis due to its non-invasiveness, portability, and high resolution, and is widely used in hospitals and mobile medical devices. Clinically, doctors usually use ultrasound probes to dynamically scan peripheral nerves, analyzing changes in nerve continuity, cross-sectional morphology (cross-sectional area), and echo intensity of nerve structures frame by frame to identify peripheral nerve morphological abnormalities. However, manual diagnosis is time-consuming and is affected by the doctor's experience and image quality. The interpretation criteria of different doctors may be inconsistent, resulting in poor sensitivity and accuracy in the identification of morphological abnormalities.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to ordinary technicians in this field. Summary of the Invention

[0004] In response to the problems in the prior art, the purpose of the present invention is to provide a method, system, device and storage medium for identifying peripheral nerve morphological abnormalities, which overcome the difficulties of the prior art and can solve the current technical problem of poor accuracy in identifying peripheral nerve morphological abnormalities.

[0005] A first aspect of the present disclosure provides a method for identifying peripheral nerve morphological abnormalities, comprising:

[0006] Acquiring an ultrasound image sequence of the peripheral nerve to be identified;

[0007] Segmenting the frame images in the ultrasound image sequence based on a three-dimensional deep learning model to obtain a segmentation mask for each frame image, wherein the segmentation mask marks a peripheral nerve area;

[0008] Calculating corresponding neural morphology information based on each segmentation mask, and analyzing consecutive frame images in the ultrasound image sequence to obtain neural morphology change information, wherein the neural morphology information includes at least one of a neural cross-sectional area and an echo intensity;

[0009] Based on the nerve morphology change information, a morphological abnormality recognition result of the peripheral nerve to be identified is determined.

[0010] Optionally, acquiring an ultrasound image sequence of the peripheral nerve to be identified includes:

[0011] The ultrasound image sequence is obtained by dynamically scanning along the anatomical path of the peripheral nerve to be identified using an ultrasound probe.

[0012] Optionally, the three-dimensional deep learning model includes an encoder, a decoder, and a skip connection structure;

[0013] The segmenting of the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask for each frame image includes:

[0014] The encoder extracts a multi-scale spatiotemporal feature map of the frame image in the ultrasound image sequence through multi-layer three-dimensional convolution and pooling operations;

[0015] The skip connection structure splices the multi-scale spatiotemporal feature maps of each layer of the encoder to the feature maps of the corresponding layer of the decoder;

[0016] The decoder generates a probability map through deconvolution and upsampling operations, and generates a segmentation mask of the frame image through activation and thresholding processing.

[0017] Optionally, segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask for each frame image includes:

[0018] When set conditions are met, segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask of the peripheral nerve area in the frame image, and using the corresponding frame image as the previous frame image and the next frame image of the previous frame image as the current frame image;

[0019] For the current frame image, based on the segmentation mask of the previous frame image, determining a peripheral nerve region in the previous frame image as a region of interest;

[0020] estimating a displacement vector between the previous frame image and the current frame image by using a motion compensation technique;

[0021] Mapping the region of interest of the previous frame image to the current frame image based on the displacement vector to obtain the region of interest of the peripheral nerve region of the current frame image;

[0022] Using the region of interest of the current frame image as a segmentation mask of the current frame image, then using the current frame image as the previous frame image and the next frame image of the current frame as the current frame image, returning the current frame image, and determining the peripheral nerve region in the previous frame image as the region of interest based on the segmentation mask of the previous frame image;

[0023] Until the set conditions are met.

[0024] Optionally, calculating corresponding neuromorphic information based on each segmentation mask includes:

[0025] Counting the number of pixels in the peripheral nerve region in the segmentation mask using a pixel counting method to calculate the nerve cross-sectional area;

[0026] The average grayscale value is calculated as the echo intensity by statistically analyzing the pixel value distribution of the peripheral nerve area in the segmentation mask.

[0027] Optionally, analyzing continuous frame images in an ultrasound image sequence to obtain information on nerve morphological changes includes:

[0028] The frame images in the ultrasound image sequence are processed using a sliding window technique, and the nerve cross-sectional area change and the echo intensity change are generated by comparing the nerve cross-sectional area and the echo intensity at upstream and downstream positions along the anatomical path.

[0029] Optionally, determining the morphological abnormality recognition result of the peripheral nerve to be identified based on the nerve morphological change information includes:

[0030] Based on the nerve cross-sectional area data of the historical frames corresponding to the frame images in the ultrasound image sequence, a dynamic threshold algorithm is used to calculate a determination threshold, wherein the determination threshold is a weighted moving standard deviation of a moving average of the nerve cross-sectional areas in the historical frames;

[0031] When the change in the nerve cross-sectional area exceeds the determination threshold, it is determined to be morphologically abnormal.

[0032] Optionally, the method for identifying peripheral nerve morphological abnormalities further includes:

[0033] When the morphological abnormality recognition result shows morphological abnormality, the morphological abnormality recognition result is displayed through a display interface.

[0034] Optionally, displaying the morphological abnormality recognition result through a display interface includes:

[0035] In the ultrasound image sequence, the abnormal nerve segments in the morphological abnormality identification result are visually marked.

[0036] Optionally, the peripheral nerve morphological abnormality identification method further includes: when the morphological abnormality identification results of the continuous multiple frames of images show morphological abnormality, sending a tactile feedback signal to the ultrasound device to prompt the operator.

[0037] Optionally, the method for identifying peripheral nerve morphological abnormalities further includes:

[0038] A clinical report is generated based on the morphological abnormality recognition result. The clinical report includes a key frame image, a nerve cross-sectional area change curve, and a schematic diagram of the anatomical position of the abnormal nerve segment. The key frame image is a frame image showing the morphological abnormality.

[0039] Optionally, before segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model, the peripheral nerve morphological abnormality identification method also includes: preprocessing the ultrasound image sequence, the preprocessing including Gaussian filtering and / or normalization operations to segment the preprocessed frame images based on the three-dimensional deep learning model.

[0040] A second aspect of the present disclosure provides a peripheral nerve morphological abnormality identification system, comprising:

[0041] an acquisition module, for acquiring an ultrasound image sequence of the peripheral nerve to be identified;

[0042] a segmentation module, which segments the frame images in the ultrasound image sequence based on a three-dimensional deep learning model to obtain a segmentation mask for each frame image, wherein the segmentation mask marks the peripheral nerve area;

[0043] an analysis module, calculating corresponding neural morphological information based on each segmentation mask, and analyzing consecutive frame images in the ultrasound image sequence to obtain neural morphological change information, wherein the neural morphological information includes at least one of neural cross-sectional area and echo intensity;

[0044] The determination module determines the morphological abnormality identification result of the peripheral nerve to be identified based on the nerve morphological change information.

[0045] A third aspect of the present disclosure provides an electronic device, comprising:

[0046] processor;

[0047] a memory storing executable instructions for the processor;

[0048] The processor is configured to execute the steps of the method for identifying peripheral nerve morphological abnormalities described in any of the above embodiments by executing the executable instructions.

[0049] A fourth aspect of the present disclosure provides a computer-readable storage medium for storing a program, which, when executed, implements the steps of the method for identifying peripheral nerve morphological abnormalities described in any embodiment.

[0050] The peripheral nerve morphological abnormality identification method, system, device, and storage medium provided by the embodiments of the present disclosure have the following advantages:

[0051] Based on the three-dimensional deep learning model, the frame images in the ultrasound image sequence are segmented, the corresponding neural morphological information is calculated based on each segmentation mask, and the continuous frame images in the ultrasound image sequence are analyzed to obtain the neural morphological change information. Based on the neural morphological change information, the morphological abnormality recognition result of the peripheral nerve to be identified is determined. In this way, the segmentation of the frame images in the ultrasound image sequence is realized by the three-dimensional deep learning model. The three-dimensional deep learning model can simultaneously process the spatial and temporal dimensions, extract the multi-scale spatiotemporal features in the ultrasound image sequence, and capture the dynamic changes of neural morphology between frames, thereby more accurately segmenting the nerve area, adapting to complex anatomical structures and blurred boundaries, significantly improving the segmentation accuracy, and maintaining a high accuracy and robustness of morphological abnormality recognition, especially in the case of lesions. At the same time, the embodiment of the present disclosure also solves the problem that traditional methods require a large amount of labeled data. The three-dimensional deep learning model can learn richer features from limited labeled data through multi-scale spatiotemporal feature extraction, and use inter-frame information to reduce the dependence on single-frame annotation, thereby reducing the demand for labeled data. At the same time, because the three-dimensional deep learning model can better capture dynamic changes and adapt to complex scenes, it is more robust to noise interference and image quality changes, enabling the model to adapt to ultrasound images of different devices, scanning parameters and image quality, thereby improving the generalization ability.

[0052] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Other features, objects and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0054] Figure 1 A schematic diagram of an ultrasound image of a peripheral nerve is shown;

[0055] Figure 2 One of the flow charts showing the method for identifying peripheral nerve morphological abnormalities provided by an embodiment of the present disclosure;

[0056] Figure 3 exhibit Figure 2 Figure 3D U-Net architecture used in the peripheral nerve morphological abnormality recognition method shown;

[0057] Figure 4 A second flowchart showing a method for identifying peripheral nerve morphological abnormalities provided by an embodiment of the present disclosure;

[0058] Figure 5 A schematic diagram showing the module structure of a peripheral nerve morphological abnormality identification system provided by an embodiment of the present disclosure;

[0059] Figure 6A schematic diagram showing the structure of an electronic device provided by an embodiment of the present disclosure;

[0060] Figure 7 It is a schematic diagram of the structure of a computer program product according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0061] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0062] In addition, the accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0063] In the existing technology, edge detection and threshold segmentation image processing techniques are used to outline peripheral nerves, and manually designed features and rules are combined to identify morphological abnormalities of peripheral nerves. Although this method has high computational efficiency, the edge detection method relies on the gradient information of the image. In ultrasound images, the boundaries of nerves are often blurred due to noise or echo interference, resulting in edge detection failure. In addition, traditional methods cannot effectively handle the diversity of nerve morphology, especially in the case of pathological conditions, when the nerve outline is blurred or irregular, the segmentation accuracy is significantly reduced. More importantly, traditional methods usually process single-frame images and lack the real-time analysis capabilities during dynamic scanning.

[0064] Existing technologies have attempted to incorporate deep learning techniques, such as using two-dimensional deep learning methods for image segmentation and recognition, which has improved segmentation accuracy. However, these methods still have significant shortcomings in their application to peripheral nerve ultrasound images. First, two-dimensional deep learning models require a large amount of labeled data for training, and the high cost and time-consuming process of labeling medical images limits their widespread application.

[0065] Secondly, existing models are mainly targeted at static tasks. Figure 1 As shown in the figure, the anatomical structure of the peripheral nerve (the area s pointed by the arrow in the figure) is complex and has fuzzy boundaries. Traditional two-dimensional deep learning models are difficult to accurately extract the nerve area and cannot effectively capture the spatiotemporal changes between frames, making it difficult to accurately identify morphological abnormalities.

[0066] Therefore, the embodiments of the present disclosure introduce an automatic identification scheme for peripheral nerve morphological abnormalities based on a three-dimensional deep learning model, aiming to overcome the defects of the existing technology in ultrasound image segmentation accuracy and poor recognition accuracy.

[0067] like Figure 2 As shown, the embodiment of the present disclosure provides a method for identifying peripheral nerve morphological abnormalities, which includes but is not limited to the following steps:

[0068] Step 210: Acquire an ultrasound image sequence of the peripheral nerve to be identified;

[0069] Step 220: Segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask for each frame image, wherein the segmentation mask marks the peripheral nerve area;

[0070] Step 230: Calculating corresponding neural morphological information based on each segmentation mask, and analyzing consecutive frame images in the ultrasound image sequence to obtain neural morphological change information, wherein the neural morphological information includes at least one of neural cross-sectional area and echo intensity;

[0071] Step 240: Based on the nerve morphology change information, determine the morphological abnormality recognition result of the peripheral nerve to be identified.

[0072] In this embodiment, the ultrasound image sequence refers to a set of continuous frame images collected by an ultrasound probe, which reflects the spatiotemporal information of the peripheral nerves. Subsequently, the frame images in the ultrasound image sequence are segmented based on the three-dimensional deep learning model to generate a segmentation mask of the peripheral nerve area. The segmentation mask refers to an image that marks the peripheral nerve area and is used to distinguish the nerve area from the background area. Next, the nerve morphology is calculated and analyzed by the segmentation mask to obtain its changes. The nerve cross-sectional area refers to the cross-sectional area of ​​the peripheral nerve in a certain frame image, and the echo intensity refers to the reflection intensity reflected by the pixel grayscale value of the nerve area in the ultrasound image. Finally, based on the changes in the nerve cross-sectional area and the echo intensity, the morphological abnormality recognition result of the peripheral nerve is determined. Morphological abnormality refers to abnormalities in the shape or structure of the peripheral nerve, such as swelling or stenosis.

[0073] The above-mentioned technical solution provided by the embodiment of the present disclosure achieves high-precision segmentation through a three-dimensional deep learning model, which solves the problem that traditional two-dimensional deep learning methods cannot capture spatiotemporal information. Traditional two-dimensional deep learning methods only perform segmentation on single-frame images and cannot utilize inter-frame information in ultrasound image sequences. Ultrasound image sequences are continuous frame images, and there is spatiotemporal correlation between frames. The three-dimensional deep learning model can process spatial and temporal dimensions at the same time, extract multi-scale spatiotemporal features in ultrasound image sequences, and capture the dynamic changes of neural morphology between frames, thereby more accurately segmenting neural areas and adapting to complex anatomical structures and blurred boundaries. Compared with traditional two-dimensional deep learning methods, the embodiment of the present disclosure significantly improves segmentation accuracy, especially in the case of lesions, it can also maintain a high accuracy and robustness of morphological abnormality recognition.

[0074] At the same time, the disclosed embodiments also solve the problem that traditional methods require a large amount of labeled data. Traditional two-dimensional deep learning methods rely on a large amount of labeled data for training, and the labeling of medical images is costly and time-consuming, which limits their widespread application. The three-dimensional deep learning model can learn richer features from limited labeled data through multi-scale spatiotemporal feature extraction, and use inter-frame information to reduce dependence on single-frame labeling, thereby reducing the demand for labeled data. At the same time, because the three-dimensional deep learning model can better capture dynamic changes and adapt to complex scenes, it is more robust to noise interference and image quality changes, enabling the model to adapt to ultrasound images of different devices, scanning parameters and image quality, thereby improving generalization capabilities.

[0075] Below Figure 2 The method for identifying abnormal peripheral nerve morphology is described in detail.

[0076] In an embodiment of the present disclosure, before segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model, the peripheral nerve morphological abnormality identification method also includes: preprocessing the ultrasound image sequence, and the preprocessing includes Gaussian filtering and / or normalization operations to segment the preprocessed frame images based on the three-dimensional deep learning model.

[0077] Gaussian filtering is a smoothing filtering technique used to reduce salt-and-pepper noise, ensuring the accuracy of subsequent neuromorphic abnormality identification. Normalization is the process of normalizing pixel values ​​to a specific range. This process uses a linear transformation to normalize pixel values ​​to the range (0-1). This preprocessing improves the robustness of 3D deep learning models.

[0078] In the embodiment of the present disclosure, executing Figure 2 Step 210 specifically includes the following steps:

[0079] The ultrasound image sequence is obtained by dynamically scanning along the anatomical path of the peripheral nerve to be identified using an ultrasound probe.

[0080] This embodiment allows for real-time acquisition of ultrasound image sequences and further immediate identification of morphological abnormalities, enabling real-time dynamic scanning and immediate identification of morphological abnormalities. The anatomical pathway refers to the natural course of peripheral nerves, such as the path of the median nerve from the wrist to the forearm.

[0081] In another embodiment, it is also possible to extract an acquired ultrasound image sequence from a storage database to obtain an ultrasound image sequence of the peripheral nerve to be identified.

[0082] In the embodiment of the present disclosure, the three-dimensional deep learning model can select one of 3D convolutional neural network (3DCNN), 3DU-Net, V-Net, 3D ResNet / 3D DenseNet.

[0083] Among them, combined Figure 3 As shown, 3D U-Net is a three-dimensional image segmentation model based on deep learning, which includes an encoder, a decoder, and a skip connection structure. The encoder includes a bottleneck layer and an output layer. In this case, the above step 220 specifically includes:

[0084] The encoder extracts multi-scale spatiotemporal feature maps of frame images in the ultrasound image sequence through multi-layer three-dimensional convolution and pooling operations, and forms the bottleneck layer at the bottom layer. The bottleneck layer contains high-semantic abstract features of the global structure. The feature map output by the bottleneck layer is passed to the decoder for layer-by-layer recovery of the spatial structure.

[0085] The skip connection structure splices the multi-scale spatiotemporal feature maps of each layer of the encoder to the feature maps of the corresponding layer of the decoder, preserving edge information and spatial details;

[0086] The decoder restores the spatial resolution layer by layer through deconvolution and upsampling operations, and finally generates a probability map, which is then activated and thresholded to generate a segmentation mask for the frame image.

[0087] Combine Figure 3 As shown in the figure, the encoder is on the left, which consists of multiple 3D convolutional layers and pooling layers. Through continuous convolution and pooling operations, the features of the frame image are gradually extracted while reducing the image resolution, enabling the model to capture multi-scale spatiotemporal feature maps, which contain feature information at different scales.

[0088] The decoder is symmetrical with the encoder. Through a series of deconvolution layers and upsampling operations, it gradually restores the low-resolution feature maps output by the encoder to the resolution of the original frame image and performs image segmentation prediction. In the decoder, feature maps from corresponding levels of the encoder are also fused to fully utilize the image context and improve segmentation accuracy.

[0089] The bottleneck layer processes high-level semantic features, while the skip connection merges detail and overall shape information by splicing feature maps from each encoder layer to the corresponding decoder layer.

[0090] 3D U-Net uses skip connections to directly connect feature maps at different levels in the encoder to the corresponding levels in the decoder. This connection method effectively solves the problem of detailed information lost during convolution and pooling, allowing the model to better utilize global and local information in the image, thereby improving segmentation accuracy.

[0091] like Figure 3 As shown in the figure, dotted lines represent skip connections. For example, the outputs of the three convolutional layers above the encoder are skip-connected to the upsampled inputs of the decoder, while the last pooling layer (downsampling) of the encoder is connected to the last three upsampling (output) layers of the decoder. Skip connections can directly transmit some of the spatial details lost during the encoding process (due to the loss of resolution caused by convolution and pooling) to the decoder, helping to more precisely restore object boundaries, which is particularly important for small objects such as neural contours.

[0092] Optionally, the decoder's output layer uses a sigmoid activation function and thresholding to generate a segmentation mask. The segmentation mask is a binary image, and a subregion or bounding box of the segmentation mask contains the surrounding nerve area. For example, white areas (pixel value 1) represent the surrounding nerve, and black areas represent the background.

[0093] Through the encoder-decoder structure and skip connections, 3D U-Net can simultaneously extract the spatial and temporal features of ultrasound images, adapt to the diversity of neural morphology, and maintain high segmentation accuracy and robustness, especially in the case of pathological conditions (such as blurred or irregular neural contours).

[0094] In one embodiment, the 3D U-Net model function is y = f(x; θ), where x is the ultrasound image sequence input D, θ is the model parameter, and y = M(x, y, t) is the segmentation mask at time t. For the first frame (t = 1), the segmentation mask M(x, y, 1) is directly generated by the 3D U-Net. The corresponding segmentation masks are then generated sequentially for multiple frames.

[0095] In another embodiment, executing step 220 can be divided into the following specific steps:

[0096] When set conditions are met, segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask of the peripheral nerve area in the frame image, and using the corresponding frame image as the previous frame image and the next frame image of the previous frame image as the current frame image;

[0097] For the current frame image, based on the segmentation mask of the previous frame image, determining a peripheral nerve region in the previous frame image as a region of interest;

[0098] Estimate the displacement vector between the previous frame image and the current frame image through motion compensation technology;

[0099] Mapping the region of interest of the previous frame image to the current frame image based on the displacement vector to obtain the region of interest of the peripheral nerve region of the current frame image;

[0100] Using the region of interest of the current frame image as a segmentation mask of the current frame image, then using the current frame image as the previous frame image and the next frame image of the current frame as the current frame image, returning the current frame image, and determining the peripheral nerve region in the previous frame image as the region of interest based on the segmentation mask of the previous frame image;

[0101] Until the set conditions are met.

[0102] In this implementation, after obtaining a segmentation mask for one frame using a 3D deep learning model, motion compensation is used to estimate the segmentation mask for each subsequent frame until the specified conditions are met again. Compared to using a 3D deep learning model, motion compensation can simplify computation and improve segmentation and recognition efficiency.

[0103] In this embodiment, the ROI (Region of Interest) refers to the primary area where the peripheral nerve is located, which generally includes a cross-section of the nerve structure. Because the peripheral nerve moves during the scanning process, the ROI area is dynamically adjusted to ensure that subsequent analysis is performed in the actual peripheral nerve area.

[0104] Optionally, the motion compensation technology uses an optical flow method (such as the Lucas-Kanade algorithm), which refers to a method for estimating pixel displacement between frame images. The optical flow method is a method for estimating motion based on changes in pixels in an image sequence. It assumes that pixels between adjacent frame images are translated on the surface of the peripheral nerve, and this translation can be calculated by analyzing the changes in pixels over time. Specifically, by comparing the motion of two frames, a displacement vector is obtained, which represents the translation of the peripheral nerve area in the current frame image relative to the previous frame image in the ultrasound image sequence. It is further used to estimate the segmentation mask of the adjacent next frame image based on the position of the peripheral nerve area in the adjacent previous frame image, so that the content changes between adjacent frames in the ultrasound image sequence remain smooth and coherent, avoiding abrupt jumps or flickers.

[0105] Specifically, starting with the second frame (t=2), the optical flow method is applied to align the peripheral nerve region based on the displacement between it and the first frame to improve inter-frame consistency and segmentation stability. The goal of motion compensation is to stabilize the segmentation result of the current frame by aligning the peripheral nerve region of the adjacent previous frame image with the current frame image, ensuring that the changes in nerve cross-sectional area and echo intensity are calculated based on the consistent peripheral nerve region. The specific steps are as follows:

[0106] Based on the segmentation mask M(x,y,t-1) of the adjacent previous frame image, the peripheral nerve area in the adjacent previous frame image is determined as the region of interest (ROI). The ROI refers to the area with a pixel value of 1 in M(x,y,t-1), indicating the main area where the peripheral nerve is located.

[0107] Optical flow method calculates displacement: Estimate the displacement vector between the frame image I(x,y,t-1) at time t-1 and the frame image I(x,y,t) at the current time t where u and represent the horizontal and vertical displacements respectively;

[0108] Based on this displacement vector Map the region of interest of the adjacent previous frame image to the current frame image to obtain the region of interest of the current frame image. Specifically, for each pixel (x, y) with a pixel value of 1 in M(x, y, t-1), calculate its new position in the current frame image according to the above displacement vector Thus, the predicted ROI of the current frame image is obtained.

[0109] The predicted ROI of the current frame image is converted into a segmentation mask M(x,y,t). Specifically, (x,y,t) is a binary image, where the pixel values ​​inside the predicted ROI are set to 1 (indicating the surrounding nerve area) and the pixel values ​​outside the ROI are set to 0 (indicating the background).

[0110] Motion compensation aligns the peripheral nerve regions at time t-1 and time t, stabilizing the segmentation mask for the current frame and preventing jumps in nerve region position due to probe movement. This ensures that subsequent calculations of nerve cross-sectional area and echo intensity are based on a consistent peripheral nerve region. Motion-compensated ROI mapping improves inter-frame segmentation consistency. After motion compensation, inter-frame consistency is significantly improved, meaning the overlap rate of segmentation masks between adjacent frames increases.

[0111] In an optional embodiment, the setting condition may include at least one of the following situations:

[0112] A large displacement error is detected (e.g. tracking failure, severe mask deviation);

[0113] Recalibrate with 3D U-Net every fixed number of frames;

[0114] Abnormal morphological changes are detected, requiring high-precision reconfirmation of the segmentation.

[0115] In the embodiment of the present disclosure, the above-mentioned three-dimensional deep learning model is a trained model. In model training and application, multi-task learning and data enhancement technology can be combined to improve the generalization ability of the model and data utilization. Multi-task learning refers to the technology of improving the overall performance of the model by simultaneously optimizing the tasks of frame image segmentation, nerve cross-sectional area calculation and echo intensity analysis. Data enhancement technology refers to a method of increasing the diversity of training data through operations such as rotation, scaling, and noise addition, specifically including random rotation, scaling, and adding Gaussian noise to ultrasound images. These technologies reduce the model's dependence on labeled data. When the labeled data is reduced by 50%, the segmentation accuracy of the model only decreases by about 5%, significantly reducing the cost of data annotation. At the same time, it enables the model to adapt to ultrasound images under different devices and scanning parameters, thereby improving generalization ability.

[0116] In the disclosed embodiment, the corresponding neuromorphic information is calculated based on each segmentation mask, including:

[0117] Counting the number of pixels in the peripheral nerve region in the segmentation mask using a pixel counting method to calculate the cross-sectional area of ​​the nerve;

[0118] The average gray value is calculated as the echo intensity by statistically analyzing the pixel value distribution of the surrounding nerve area in the segmentation mask.

[0119] In this embodiment, when calculating the cross-sectional area and echo intensity of the nerve, the pixel counting method is used to count the number of pixels in the peripheral nerve region in the segmentation mask. The pixel counting method is a method of calculating the area of ​​the region by accumulating pixel values. Assume that the peripheral nerve region contains n pixels, and each pixel p iIf is 1, then the cross-sectional area of ​​the nerve is:

[0120]

[0121] According to the image resolution, it is converted into physical area. The echo intensity is calculated by statistically analyzing the pixel value distribution of the surrounding nerve area in the segmentation mask. Assume that the pixel values ​​are I i , then the echo intensity is:

[0122]

[0123] The unit is grayscale value.

[0124] In an embodiment of the present disclosure, analyzing the continuous frame images in the ultrasound image sequence to obtain the nerve morphology change information includes:

[0125] The frame images in the ultrasound image sequence are processed using a sliding window technique, and the nerve cross-sectional area change and the echo intensity change are generated by comparing the nerve cross-sectional area and the echo intensity at upstream and downstream positions along the anatomical path.

[0126] When analyzing changes in nerve cross-sectional area and echo intensity, a sliding window technique is used to process frame images in an ultrasound image sequence. Sliding window technology refers to a technique that moves a fixed-size window frame by frame to analyze changes in consecutive frame images. By comparing the cross-sectional area and echo intensity of nerves at upstream and downstream positions (such as the proximal and distal ends) along the anatomical path, a change sequence is generated. A change sequence refers to a data sequence that reflects the changes in nerve cross-sectional area and echo intensity over time and space, including spatial differences (upstream and downstream) and temporal differences (between frames), providing a basis for abnormality judgment. This technology solves the problem of insufficient real-time performance of traditional methods and existing deep learning models in dynamic scanning, and can improve the efficiency of identifying morphological abnormalities.

[0127] In an embodiment of the present disclosure, determining the morphological abnormality recognition result of the peripheral nerve to be identified based on the nerve morphological change information includes:

[0128] Based on the historical data of neural morphological changes of historical frames corresponding to frame images in the ultrasound image sequence, a dynamic threshold algorithm is used to obtain a judgment threshold. When the neural morphological change information exceeds the judgment threshold, it is determined to be morphological abnormality.

[0129] The dynamic threshold algorithm refers to a method for dynamically calculating the threshold based on historical data. For example, the threshold is the weighted moving standard deviation of the moving average of the nerve cross-sectional area and echo intensity in the historical frame. Suppose the nerve cross-sectional area of ​​the current frame image is A t , the previous frame image is A t-1 , then the rate of change of nerve cross-sectional area is:

[0130]

[0131] Similarly, let the echo intensity of the current frame image be I t , the previous frame image is I t-1 , then the echo intensity change rate is:

[0132]

[0133] In this embodiment, the dynamic threshold is simple to calculate and is suitable for real-time processing of ultrasound image sequences. By adopting the dynamic threshold, the threshold can be adaptively adjusted according to the actual situation of the patient. The peripheral nerve characteristics of different patients (such as nerve cross-sectional area, echo intensity) may vary greatly. The dynamic threshold can reduce misjudgment and improve the personalized accuracy of the judgment. In addition, the ultrasound image sequence is generated by dynamic scanning, and the nerve cross-sectional area and echo intensity will change with time and anatomical position. By analyzing historical frame data, the dynamic threshold can better reflect these dynamic change trends and avoid misjudgment caused by local noise or short-term fluctuations. For example, if the overall cross-sectional area of ​​a patient's nerve is larger, the dynamic threshold will be increased accordingly to avoid misjudging normal fluctuations as abnormalities.

[0134] Furthermore, dynamic thresholds smooth out the effects of noise and outliers through moving average and standard deviation calculations, making them more robust than fixed thresholds. In ultrasound images, noise, probe movement, or image quality variations can cause data fluctuations. Dynamic thresholds can reduce the impact of these interferences on abnormality determination, maintaining stability. Dynamic thresholds are automatically calculated based on real-time data, eliminating the need for subjective threshold setting and providing a more objective judgment standard.

[0135] In summary, dynamic thresholds can more accurately and stably identify peripheral nerve morphological abnormalities through adaptive adjustment, capturing dynamic changes, improving robustness, reducing subjectivity, and supporting real-time processing, significantly improving the reliability and efficiency of diagnosis.

[0136] For example, if the threshold for nerve cross-sectional area change is 20% and the threshold for echo intensity change is 15%, a morphological abnormality is determined if either the nerve cross-sectional area change exceeds 20% or the echo intensity change exceeds 15%. Conversely, if neither the nerve cross-sectional area change nor the echo intensity change exceeds the corresponding threshold, the morphology is determined to be normal.

[0137] like Figure 4 As shown, the embodiment of the present disclosure also provides another method for identifying peripheral nerve morphological abnormalities, which includes but is not limited to the following steps:

[0138] Step 410: Acquire an ultrasound image sequence of the peripheral nerve to be identified;

[0139] Step 420: Segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask of the peripheral nerve region in the frame image;

[0140] Step 430: Calculating corresponding neural morphological information based on each segmentation mask, and analyzing consecutive frame images in the ultrasound image sequence to obtain neural morphological change information, wherein the neural morphological information includes at least one of neural cross-sectional area and echo intensity;

[0141] Step 440: determining a morphological abnormality recognition result of the peripheral nerve to be identified based on the nerve morphological change information;

[0142] Step 450: When the morphological abnormality recognition result shows morphological abnormality, the morphological abnormality recognition result is displayed through a display interface.

[0143] In this implementation, the display includes keyframe images of abnormal nerve segments and related parameters (such as changes in nerve cross-sectional area). Keyframe images are frames selected from ultrasound image sequences that display morphological abnormalities, facilitating intuitive judgment by the operator. The display interface supports interactive operations, such as zooming in on keyframe images or adjusting display parameters, further enhancing the user experience.

[0144] In an embodiment of the present disclosure, in an ultrasound image sequence, abnormal nerve segments in the morphological abnormality identification results are visually marked, for example, using a color ribbon marking method. The color ribbon refers to a visual marker that highlights the abnormal segment and is used to highlight the area of ​​the abnormal nerve segment. For example, swollen abnormal nerve segments are marked in red, and stenosis abnormal nerve segments are marked in blue. The algorithm is integrated into the ultrasound device or computer system, supports real-time processing, and provides an API interface, which refers to an application programming interface, for facilitating integration with other medical systems, further improving the practicality and scalability of the technology.

[0145] In an embodiment of the present disclosure, the method for identifying peripheral nerve morphological abnormalities further includes: when the morphological abnormality identification results of multiple consecutive frames of images show morphological abnormalities, sending a tactile feedback signal to the ultrasound device to prompt the operator.

[0146] In the disclosed embodiments, a clinical report is automatically generated based on the results of morphological abnormality recognition. The clinical report is a document summarizing the abnormality recognition results. The clinical report includes key frame images, a nerve cross-sectional area change curve, and a schematic diagram of the anatomical location of the abnormal nerve segment. The nerve cross-sectional area change curve is a curve generated based on the nerve cross-sectional area change of the frame image, and the anatomical location schematic diagram is a schematic diagram that marks the anatomical location of the abnormal nerve segment, such as the abnormal location of the distal end of the median nerve.

[0147] Figure 5 FIG. 1 is a schematic diagram of a module of an embodiment of a peripheral nerve morphological abnormality identification system provided by the present disclosure. Figure 5 As shown, the peripheral nerve morphological abnormality identification system 500 of the present disclosure includes but is not limited to:

[0148] An acquisition module 510 acquires an ultrasound image sequence of the peripheral nerve to be identified;

[0149] a segmentation module 520 , which segments the frame images in the ultrasound image sequence based on a three-dimensional deep learning model to obtain a segmentation mask for each frame image, wherein the segmentation mask marks the peripheral nerve area;

[0150] an analysis module 530 that calculates corresponding neural morphology information based on each segmentation mask and analyzes consecutive frame images in the ultrasound image sequence to obtain neural morphology change information, wherein the neural morphology information includes at least one of neural cross-sectional area and echo intensity;

[0151] The determination module 540 determines the morphological abnormality identification result of the peripheral nerve to be identified based on the nerve morphological change information.

[0152] In an optional embodiment, the acquisition module 510 is specifically configured to:

[0153] Dynamic scanning is performed along the anatomical path of the peripheral nerve to be identified by using an ultrasound probe to continuously acquire a sequence of ultrasound images of the peripheral nerve to be identified.

[0154] In an optional embodiment, the three-dimensional deep learning model includes an encoder, a decoder, and a skip connection structure; the segmentation module 520 is specifically used to:

[0155] The encoder extracts a multi-scale spatiotemporal feature map of the frame image in the ultrasound image sequence through multi-layer three-dimensional convolution and pooling operations;

[0156] The skip connection structure splices the multi-scale spatiotemporal feature maps of each layer of the encoder to the feature maps of the corresponding layer of the decoder;

[0157] The decoder generates a probability map through deconvolution and upsampling operations, and generates a segmentation mask for the frame image through activation and thresholding.

[0158] In an optional embodiment, the segmentation module 520 is specifically configured to:

[0159] When set conditions are met, the frame images in the ultrasound image sequence are segmented based on the three-dimensional deep learning model to obtain a segmentation mask for each frame image, and the corresponding frame image is used as the previous frame image, and the next frame image of the previous frame image is used as the current frame image;

[0160] For the current frame image, based on the segmentation mask of the previous frame image, determining a peripheral nerve region in the previous frame image as a region of interest;

[0161] estimating a displacement vector between a previous frame image and the current frame image by using a motion compensation technique;

[0162] Mapping the region of interest of the previous frame image to the current frame image based on the displacement vector to obtain the region of interest of the peripheral nerve region of the current frame image;

[0163] Using the region of interest of the current frame image as a segmentation mask of the current frame image, then using the current frame image as the previous frame image and the next frame image of the current frame as the current frame image, returning the current frame image, and determining the peripheral nerve region in the previous frame image as the region of interest based on the segmentation mask of the previous frame image;

[0164] Until the set conditions are met.

[0165] In an optional embodiment, the analysis module 530 is specifically configured to:

[0166] The calculating corresponding neuromorphic information based on each segmentation mask includes:

[0167] Counting the number of pixels in the peripheral nerve region in the segmentation mask using a pixel counting method to calculate the nerve cross-sectional area;

[0168] The average gray value is calculated as the echo intensity by statistically analyzing the pixel value distribution of the surrounding nerve area in the segmentation mask.

[0169] In an optional embodiment, the analysis module 530 is specifically configured to:

[0170] The frame images in the ultrasound image sequence are processed using a sliding window technique, and the nerve cross-sectional area change and the echo intensity change are generated by comparing the nerve cross-sectional area and the echo intensity at upstream and downstream positions along the anatomical path.

[0171] In an optional embodiment, the determination module 540 is specifically configured to:

[0172] Based on the nerve cross-sectional area data of the historical frames corresponding to the frame images in the ultrasound image sequence, a dynamic threshold algorithm is used to calculate a determination threshold, wherein the determination threshold is a weighted moving standard deviation of a moving average of the nerve cross-sectional areas in the historical frames;

[0173] When the change in the nerve cross-sectional area exceeds the determination threshold, it is determined to be morphologically abnormal.

[0174] In an optional embodiment, the determination module 540 is further configured to:

[0175] When the morphological abnormality recognition result shows morphological abnormality, the morphological abnormality recognition result is displayed through a display interface.

[0176] In an optional embodiment, the determination module 540 is further configured to:

[0177] In the ultrasound image sequence, the abnormal nerve segments in the morphological abnormality identification result are visually marked.

[0178] In an optional embodiment, when the morphological abnormality recognition results of the continuous multiple frames of images show morphological abnormality, a tactile feedback signal is sent to the ultrasound device to prompt the operator.

[0179] In an optional embodiment, the determination module 540 is further configured to:

[0180] A clinical report is generated based on the morphological abnormality recognition result. The clinical report includes a key frame image, a nerve cross-sectional area change curve, and a schematic diagram of the anatomical position of the abnormal nerve segment. The key frame image is a frame image showing the morphological abnormality.

[0181] In an optional embodiment, the segmentation module 520 is further configured to:

[0182] Before segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model, the peripheral nerve morphological abnormality identification method also includes: preprocessing the ultrasound image sequence, wherein the preprocessing includes Gaussian filtering and / or normalization operations to segment the preprocessed frame images based on the three-dimensional deep learning model.

[0183] This system achieves high-precision segmentation through a three-dimensional deep learning model, addressing the problem of traditional two-dimensional deep learning methods' inability to capture spatiotemporal information. The three-dimensional deep learning model can simultaneously process spatial and temporal dimensions, extracting multi-scale spatiotemporal features from ultrasound image sequences and capturing dynamic changes in neural morphology between frames, thereby more accurately segmenting neural regions and adapting to complex anatomical structures and blurred boundaries. Compared to traditional two-dimensional deep learning methods, the disclosed embodiments significantly improve segmentation accuracy, maintaining high accuracy and robustness in identifying morphological abnormalities, particularly in the presence of pathological conditions. Furthermore, the disclosed embodiments address the problem of traditional methods requiring large amounts of annotated data. By extracting multi-scale spatiotemporal features, the three-dimensional deep learning model can learn richer features from limited annotated data, leveraging inter-frame information to reduce reliance on single-frame annotations, thereby reducing the need for annotated data. Furthermore, because the three-dimensional deep learning model can better capture dynamic changes and adapt to complex scenarios, it is more robust to noise interference and image quality variations, enabling the model to adapt to ultrasound images from different devices, scanning parameters, and image quality, improving generalization capabilities.

[0184] This system consists of an acquisition module 510, a segmentation module 520, an analysis module 530, and a determination module 540. These modules can be interconnected via data interfaces to collaboratively implement a solution for identifying peripheral nerve morphological abnormalities. The system can be embedded into front-end development tools as a standalone software module or provided as a cloud service interface for remote access by developers.

[0185] An embodiment of the present invention further provides a peripheral nerve morphological abnormality identification device, comprising a processor and a memory storing executable instructions for the processor. The processor is configured to execute the executable instructions to complete the steps of the peripheral nerve morphological abnormality identification method.

[0186] As demonstrated above, the peripheral nerve morphological abnormality identification device of the present invention achieves high-precision segmentation through a three-dimensional deep learning model, overcoming the inability of traditional two-dimensional deep learning methods to capture spatiotemporal information. The three-dimensional deep learning model simultaneously processes both spatial and temporal dimensions, extracting multi-scale spatiotemporal features from ultrasound image sequences and capturing dynamic changes in neural morphology between frames. This allows for more accurate segmentation of neural regions and adapts to complex anatomical structures and blurred boundaries.

[0187] Those skilled in the art will appreciate that various aspects of the present invention may be implemented as systems, methods, or program products. Accordingly, various aspects of the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, microcode, etc.), or in a combination of hardware and software, collectively referred to herein as "circuits," "modules," or "platforms."

[0188] Figure 6 This is a schematic diagram of the structure of the peripheral nerve morphological abnormality identification device of the present invention. Figure 6 An electronic device 600 according to this embodiment of the present invention will be described. Figure 6 The electronic device 600 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0189] like Figure 6 As shown, electronic device 600 is implemented as a general-purpose computing device. Components of electronic device 600 may include, but are not limited to, at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), and a display unit 640.

[0190] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 performs the steps of various exemplary embodiments of the present invention described in the above-mentioned peripheral nerve morphological abnormality identification method section of this specification. For example, the processing unit 610 can perform the following steps: Figure 2 or Figure 4 Follow the steps shown in .

[0191] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 621 and / or a cache memory unit 622 , and may further include a read-only memory unit (ROM) 623 .

[0192] The storage unit 620 may also include a program / utility 624 having a set (at least one) of program modules 625, such program modules 625 including but not limited to: a processing system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0193] Bus 630 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0194] The electronic device 600 can also communicate with one or more external devices 601 (e.g., a keyboard, a pointing device, a Bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device 600, and / or any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., a router, a modem, etc.). Such communication can occur via an input / output (I / O) interface 650. Furthermore, the electronic device 600 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 660. The network adapter 660 can communicate with other modules of the electronic device 600 via a bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with the electronic device 600, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0195] The present disclosure also provides a computer-readable storage medium for storing a program that, when executed, implements the steps of the method for identifying peripheral nerve morphological abnormalities. In some possible implementations, various aspects of the present invention may also be implemented in the form of a program product, which includes program code. When the program product is executed on a terminal device, the program code is configured to cause the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the aforementioned section of this specification regarding the method for identifying peripheral nerve morphological abnormalities.

[0196] like Figure 7 As shown, a computer program product 700 for implementing the above method according to an embodiment of the present invention can be implemented in a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the computer program product of the present invention is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0197] The computer program product can employ any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0198] Computer-readable storage media may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination thereof.

[0199] The program code for performing the processes of the present invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0200] In summary, the purpose of the present invention is to provide a method, system, device and storage medium for identifying peripheral nerve morphological abnormalities, which realizes the segmentation of frame images in ultrasound image sequences through a three-dimensional deep learning model. The three-dimensional deep learning model can simultaneously process spatial and temporal dimensions, extract multi-scale spatiotemporal features in ultrasound image sequences, and capture the dynamic changes of neural morphology between frames, thereby more accurately segmenting nerve regions, adapting to complex anatomical structures and blurred boundaries, significantly improving segmentation accuracy, and maintaining high accuracy and robustness of morphological abnormality recognition, especially in the case of lesions. At the same time, the embodiments of the present disclosure also solve the problem that traditional methods require a large amount of labeled data. The three-dimensional deep learning model can learn richer features from limited labeled data through multi-scale spatiotemporal feature extraction, and use inter-frame information to reduce dependence on single-frame annotation, thereby reducing the demand for labeled data. At the same time, because the three-dimensional deep learning model can better capture dynamic changes and adapt to complex scenes, it is more robust to noise interference and image quality changes, enabling the model to adapt to ultrasound images of different devices, scanning parameters and image quality, thereby improving generalization ability.

[0201] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for identifying peripheral nerve morphological abnormalities, characterized in that: include: Acquiring an ultrasound image sequence of the peripheral nerve to be identified; Segmenting the frame images in the ultrasound image sequence based on a three-dimensional deep learning model to obtain a segmentation mask for each frame image, wherein the segmentation mask marks a peripheral nerve area; Calculating corresponding neural morphology information based on each segmentation mask, and analyzing consecutive frame images in the ultrasound image sequence to obtain neural morphology change information, wherein the neural morphology information includes at least one of a neural cross-sectional area and an echo intensity; Based on the nerve morphology change information, a morphological abnormality recognition result of the peripheral nerve to be identified is determined.

2. The method for identifying peripheral nerve morphological abnormalities according to claim 1, characterized in that: The step of acquiring an ultrasound image sequence of the peripheral nerve to be identified comprises: The ultrasound image sequence is obtained by dynamically scanning along the anatomical path of the peripheral nerve to be identified using an ultrasound probe.

3. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: The three-dimensional deep learning model includes an encoder, a decoder and a skip connection structure; The segmenting of the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask for each frame image includes: The encoder extracts a multi-scale spatiotemporal feature map of the frame image in the ultrasound image sequence through multi-layer three-dimensional convolution and pooling operations; The skip connection structure splices the multi-scale spatiotemporal feature maps of each layer of the encoder to the feature maps of the corresponding layer of the decoder; The decoder generates a probability map through deconvolution and upsampling operations, and generates a segmentation mask of the frame image through activation and thresholding processing.

4. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: The segmenting of the frame images in the ultrasound image sequence based on the three-dimensional deep learning model to obtain a segmentation mask for each frame image includes: When set conditions are met, the frame images in the ultrasound image sequence are segmented based on the three-dimensional deep learning model to obtain a segmentation mask for each frame image, and the corresponding frame image is used as the previous frame image, and the next frame image of the previous frame image is used as the current frame image; For the current frame image, based on the segmentation mask of the previous frame image, determining a peripheral nerve region in the previous frame image as a region of interest; estimating a displacement vector between the previous frame image and the current frame image by using a motion compensation technique; Mapping the region of interest of the previous frame image to the current frame image based on the displacement vector to obtain the region of interest of the peripheral nerve region of the current frame image; Using the region of interest of the current frame image as a segmentation mask of the current frame image, then using the current frame image as the previous frame image and the next frame image of the current frame as the current frame image, returning the current frame image, and determining the peripheral nerve region in the previous frame image as the region of interest based on the segmentation mask of the previous frame image; Until the set conditions are met.

5. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: The calculating corresponding neuromorphic information based on each segmentation mask includes: Counting the number of pixels in the peripheral nerve region in the segmentation mask using a pixel counting method to calculate the nerve cross-sectional area; The average grayscale value is calculated as the echo intensity by counting the pixel value distribution of the peripheral nerve area in the segmentation mask.

6. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: The analyzing the continuous frame images in the ultrasound image sequence to obtain the nerve morphology change information includes: The frame images in the ultrasound image sequence are processed using a sliding window technique, and the nerve cross-sectional area change and the echo intensity change are generated by comparing the nerve cross-sectional area and the echo intensity at upstream and downstream positions along the anatomical path.

7. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: The determining of the morphological abnormality recognition result of the peripheral nerve to be identified based on the nerve morphological change information includes: Based on the nerve cross-sectional area data of the historical frames corresponding to the frame images in the ultrasound image sequence, a dynamic threshold algorithm is used to calculate a determination threshold, where the determination threshold is a weighted moving standard deviation of the moving average of the nerve cross-sectional areas in the historical frames; When the change in the nerve cross-sectional area exceeds the determination threshold, it is determined to be morphologically abnormal.

8. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: The method for identifying abnormal peripheral nerve morphology further comprises: When the morphological abnormality recognition result shows morphological abnormality, the morphological abnormality recognition result is displayed through a display interface.

9. The method for identifying peripheral nerve morphological abnormalities according to claim 8, characterized in that: The displaying of the morphological abnormality recognition result through a display interface includes: In the ultrasound image sequence, the abnormal nerve segments in the morphological abnormality identification result are visually marked.

10. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: The peripheral nerve morphological abnormality identification method further includes: when the morphological abnormality identification results of the continuous multiple frames of images show morphological abnormality, sending a tactile feedback signal to the ultrasound device to prompt the operator.

11. The method for identifying peripheral nerve morphological abnormalities according to claim 1, characterized in that: The method for identifying abnormal peripheral nerve morphology further comprises: A clinical report is generated based on the morphological abnormality recognition result. The clinical report includes a key frame image, a nerve cross-sectional area change curve, and a schematic diagram of the anatomical position of the abnormal nerve segment. The key frame image is a frame image showing the morphological abnormality.

12. The method for identifying peripheral nerve morphological abnormalities according to claim 1, wherein: Before segmenting the frame images in the ultrasound image sequence based on the three-dimensional deep learning model, the peripheral nerve morphological abnormality identification method also includes: preprocessing the ultrasound image sequence, wherein the preprocessing includes Gaussian filtering and / or normalization operations to segment the preprocessed frame images based on the three-dimensional deep learning model.

13. A peripheral nerve morphological abnormality identification system, characterized in that: include: an acquisition module, for acquiring an ultrasound image sequence of the peripheral nerve to be identified; a segmentation module, which segments the frame images in the ultrasound image sequence based on a three-dimensional deep learning model to obtain a segmentation mask for each frame image, wherein the segmentation mask marks the peripheral nerve area; an analysis module, calculating corresponding neural morphological information based on each segmentation mask, and analyzing consecutive frame images in the ultrasound image sequence to obtain neural morphological change information, wherein the neural morphological information includes at least one of neural cross-sectional area and echo intensity; The determination module determines the morphological abnormality identification result of the peripheral nerve to be identified based on the nerve morphological change information.

14. An electronic device, characterized in that: include: processor; a memory storing executable instructions for the processor; The processor is configured to execute the steps of the peripheral nerve morphological abnormality identification method according to any one of claims 1 to 12 by executing the executable instructions.

15. A computer-readable storage medium for storing a program, characterized in that: When the program is executed, the steps of the method for identifying peripheral nerve morphological abnormalities according to any one of claims 1 to 12 are implemented.

Images