Median nerve volume ultrasonic diagnosis device based on artificial intelligence
By constructing a median nerve disease prediction model based on convolutional neural networks and a three-dimensional ultrasound probe, the problem of significant impact from operator skill was solved, enabling portable and highly accurate detection of median nerve diseases and simplifying clinical diagnosis.
Patent Information
- Application Number
- CN202410795684.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-19
- Publication Date
- 2025-12-26
AI Technical Summary
Existing median nerve ultrasound examinations are greatly influenced by the operator's subjectivity, the diagnostic devices are inconvenient to carry, and the two-dimensional ultrasound images are difficult for clinicians to understand.
A median nerve disease prediction model was constructed using the U-Net segmentation algorithm and ResNet3D classification algorithm based on convolutional neural networks. Combined with an automatic scanning mobile device and a three-dimensional ultrasound probe, automatic detection and analysis were achieved.
It achieves stable detection unaffected by operator skill, boasts high diagnostic accuracy, is portable, and produces easily understandable images, providing a more practical detection method.
Smart Images
Figure HDA0004901400500000011 
Figure HDA0004901400500000012 
Figure HDA0004901400500000021
Abstract
Description
Technical Field
[0001] This invention belongs to the field of ultrasound detection technology and relates to an artificial intelligence-based median nerve volume ultrasound diagnostic device. Background Technology
[0002] Carpal tunnel syndrome and other median nerve disorders account for a large proportion of peripheral nerve diseases. Clinical assessment of peripheral neuropathy mainly relies on clinical symptoms, electrophysiological examination, and MRI. In 1991, Buchberger first described the ultrasound changes in median nerve carpal tunnel syndrome. In 2012, evidence-based medicine suggested that ultrasound has significant value in diagnosing median nerve diseases. Papers published in journals such as World Neurosurgery also confirmed that ultrasound examination is one of the most timely and effective imaging methods for diagnosing peripheral nerve diseases.
[0003] Ultrasound examination of nerves requires operators to quickly and accurately locate the nerve based on certain anatomical landmarks and perform continuous ultrasound scans following the nerve's anatomical course. This necessitates extensive anatomical knowledge and familiarity with the surrounding muscles, tendons, blood vessels, and other anatomical structures. Therefore, ultrasound examination of the median nerve demands a high level of operator skill. Operators lacking sufficient expertise may struggle to quickly and accurately locate the median nerve, leading to significant operator-related instability and a less stable examination process. Existing volumetric ultrasound diagnostic devices for median nerve diseases are bulky and inconvenient to carry, greatly limiting their application. Furthermore, while two-dimensional ultrasound images of the median nerve are clear, they remain relatively difficult for clinicians to interpret. Summary of the Invention
[0004] This invention provides an artificial intelligence-based median nerve volumetric ultrasound diagnostic device, which solves the technical problems of existing median nerve ultrasound examinations being greatly affected by the operator's subjectivity and the diagnostic devices being inconvenient to carry.
[0005] In a first aspect, the present invention claims protection for a method for constructing a median nerve disease prediction model, comprising the following steps:
[0006] S1: Median nerve segmentation
[0007] 1) Obtain continuous cross-sectional slices of three-dimensional reconstruction images of the wrist to mid-forearm from 200 subjects with median nerve lesions and 200 healthy subjects, and annotate the outline of the median nerve in each slice. The annotation includes the position and shape information of the median nerve on each slice. All subjects were required to have no acute trauma, fracture, congenital malformation or other lesions in their wrist and forearm.
[0008] 2) Use an appropriate segmentation algorithm to segment the sliced image;
[0009] 3) Overlay the segmented median nerve contour onto the original slice image and reprocess to obtain a three-dimensional reconstruction image of the wrist to mid-forearm with the median nerve contour attached.
[0010] S2: Classification of median nerve lesions
[0011] 1) Each 3D reconstruction image with the median nerve outline will be labeled with a category indicating whether the image contains carpal tunnel syndrome;
[0012] 2) Use an appropriate classification algorithm to classify the 3D reconstructed image. Based on the image features and median nerve contour information, automatically classify the image into different categories, namely, those with median nerve lesions and those without median nerve lesions.
[0013] Furthermore, the segmentation algorithm is the U-Net segmentation algorithm based on convolutional neural networks.
[0014] Furthermore, the classification algorithm is ResNet3D, a classification algorithm based on 3D convolutional neural networks.
[0015] In a second aspect of the invention, the invention claims protection for a median nerve disease prediction model obtained using the construction method of the first aspect of the invention.
[0016] In a third aspect, the present invention claims protection for a median nerve volume ultrasound diagnostic system, comprising:
[0017] The data acquisition module is used to acquire a continuous sequence of transverse images from the patient's wrist to the mid-forearm.
[0018] The processing module is used to perform artificial intelligence recognition on the acquired raw ultrasound images, automatically identify the median nerve portion in the raw ultrasound images, optimize the median nerve images, virtually generate a complete three-dimensional image of the median nerve, and perform perspective processing on the raw ultrasound images of the anatomical structures surrounding the nerve to obtain the processed image.
[0019] The model building module is used to build a predictive model for median nerve diseases;
[0020] The diagnostic module is used to diagnose whether median nerve lesions exist in processed images using a median nerve disease prediction model.
[0021] Furthermore, the median nerve disease prediction model is obtained by the following method:
[0022] S1: Median nerve segmentation
[0023] 1) Obtain continuous cross-sectional slices of three-dimensional reconstruction images of the wrist to mid-forearm from 200 subjects with median nerve lesions and 200 healthy subjects, and annotate the outline of the median nerve in each slice. The annotation includes the position and shape information of the median nerve on each slice. All subjects were required to have no acute trauma, fracture, congenital malformation or other lesions in their wrist and forearm.
[0024] 2) Use an appropriate segmentation algorithm to segment the sliced image;
[0025] 3) Overlay the segmented median nerve contour onto the original slice image and reprocess to obtain a three-dimensional reconstruction image of the wrist to mid-forearm with the median nerve contour attached.
[0026] S2: Classification of median nerve lesions
[0027] 1) Each 3D reconstruction image with the median nerve outline will be labeled with a category indicating whether the image contains carpal tunnel syndrome;
[0028] 2) Use an appropriate classification algorithm to classify the 3D reconstructed image. Based on the image features and median nerve contour information, automatically classify the image into different categories, namely, those with median nerve lesions and those without median nerve lesions.
[0029] Furthermore, the segmentation algorithm is the U-Net segmentation algorithm based on convolutional neural networks.
[0030] Furthermore, the classification algorithm is ResNet3D, a classification algorithm based on 3D convolutional neural networks.
[0031] In a fourth aspect, the present invention claims protection for an artificial intelligence-based median nerve volumetric ultrasound diagnostic device, comprising an automatic scanning and moving device 1, a coupling medium container 2, a portable ultrasound control device 3, and a three-dimensional ultrasound probe 4. The automatic scanning and moving device 1 is connected to the three-dimensional ultrasound probe 4 and can move the three-dimensional ultrasound probe 4. The coupling medium container 2 is provided with a groove 21 for holding the coupling medium. The front side of the coupling medium container 2 is covered with a transparent cover 5. The three-dimensional ultrasound probe 4 is disposed between the coupling medium container 2 and the transparent cover 5. Both the automatic scanning and moving device 2 and the ultrasound probe 4 are connected to the portable ultrasound control device 3.
[0032] Furthermore, the automatic scanning and moving device 1 is equipped with a drive motor and a threaded rod 11. One side of the three-dimensional ultrasonic probe 4 is connected to the threaded rod 11. The drive motor drives the threaded rod 11 to rotate, thereby driving the three-dimensional ultrasonic probe 4 to move.
[0033] Furthermore, the automatic scanning mobile device 1 is installed on the coupling medium container 2 and is integrally formed with the coupling medium container 2 and the transparent cover 5.
[0034] Furthermore, the portable ultrasonic control device 3 includes a screen area 31, a keyboard area 32, and a handle 33.
[0035] Furthermore, the coupling medium container 4 is provided with a power button 6, an on / off switch 7, and an emergency stop button 8.
[0036] Beneficial effects
[0037] This invention establishes a median nerve disease prediction model and a median nerve disease diagnosis system. Based on the established model and artificial intelligence diagnosis system, the detection and analysis of the median nerve can be completed automatically, and test results can be obtained. The entire testing process is not affected by the operator's skill and has high stability.
[0038] The diagnostic device of this invention is simple to operate, portable, and applicable in various scenarios. It employs a three-dimensional ultrasound probe, facilitating the interpretation of ultrasound images by clinicians. Furthermore, this invention provides a more practical detection method for the clinical diagnosis and preoperative examination of median nerve diseases.
[0039] Ultrasound examination of peripheral nerves typically employs high-frequency linear array probes. In longitudinal sections, the nerve appears as a long, parallel linear structure containing strip-shaped hypoechoic nerve bundles and hyperechoic perineural membranes. In transverse sections, the nerve appears as a honeycomb-like, oval structure containing several small, round or oval hypoechoic nerve bundles, surrounded by epithelial ganglia and neurofatty tissue. Because transverse images are most valuable for describing nerve anatomy and detecting lesions, the model and diagnostic device developed in this invention use the cross-sectional area of the nerve as a crucial parameter for diagnosing neurological diseases, resulting in high diagnostic accuracy. Attached Figure Description
[0040] Figure 1 This is a front view of the median nerve volume ultrasound diagnostic device of this application.
[0041] Figure 2 This is a cross-sectional view of the median nerve volume ultrasound diagnostic device of this application.
[0042] Figure 3 This is a front view of the portable ultrasonic control device of this application.
[0043] 1-Automatic scanning moving device; 2-Coupled medium container; 3-Portable ultrasonic control device; 4-Three-dimensional ultrasonic probe; 5-Transparent cover; 6-Power button; 7-On / off switch; 8-Emergency stop button; 11-Threaded rod; 21-Groove; 31-Screen area; 32-Keyboard area; 33-Handle Detailed Implementation
[0044] The present application is described in detail below with reference to embodiments, but this does not imply any adverse limitations on the present application. The present application has been described in detail herein, and specific embodiments thereof have been disclosed. It will be apparent to those skilled in the art that various changes and modifications can be made to the specific implementations of the present application without departing from the spirit and scope thereof.
[0045] Example 1
[0046] This invention provides a method for constructing a median nerve disease prediction model, comprising the following steps:
[0047] S1: Median nerve segmentation
[0048] 1) Obtain continuous cross-sectional slices of three-dimensional reconstruction images of the wrist to mid-forearm from 200 subjects with median nerve lesions and 200 healthy subjects, and annotate the outline of the median nerve in each slice. The annotation includes the position and shape information of the median nerve on each slice. All subjects were required to have no acute trauma, fracture, congenital malformation or other lesions in their wrist and forearm.
[0049] 2) Use an appropriate segmentation algorithm to segment the sliced image;
[0050] 3) Overlay the segmented median nerve contour onto the original slice image and reprocess to obtain a three-dimensional reconstruction image of the wrist to mid-forearm with the median nerve contour attached.
[0051] S2: Classification of median nerve lesions
[0052] 1) Each 3D reconstruction image with an attached median nerve outline will be labeled with a category indicating whether the image contains a median nerve lesion;
[0053] 2) Use an appropriate classification algorithm to classify the 3D reconstructed image. Based on the image features and median nerve contour information, automatically classify the image into different categories, namely, those with median nerve lesions and those without median nerve lesions.
[0054] The segmentation algorithm is the U-Net segmentation algorithm based on convolutional neural networks.
[0055] The U-Net algorithm operates in the following steps: (1) Shrinking path: Through repeated convolution and pooling layer sequences, the network gradually captures and reduces the dimensions of the image, thereby extracting key features in the image; (2) Expanding path: Through the above sampling and convolution operations, the size and details of the image are gradually restored. In this process, skip connections are used to fuse the feature map of the shrinking path with the corresponding layer of the expanding path to ensure that important location information is not lost; (3) Final segmentation map: Through a 1x1 convolutional layer, the deep features are mapped to the category prediction of each pixel, and a segmentation map with the same size as the original image is output, where each pixel is labeled as the median nerve or the background.
[0056] Furthermore, the classification algorithm is ResNet3D, a classification algorithm based on 3D convolutional neural networks.
[0057] Furthermore, the ResNet3D classification algorithm includes the following steps: (1) 3D convolutional layer: using 3D convolution to process the input image and capture features in depth, height and width; (2) residual connection: maintaining the depth of the network through residual modules, while avoiding the gradient vanishing problem during training. These modules ensure that features are not destroyed when they are passed from one layer to another; (3) feature pooling: using global average pooling to reduce overfitting and aggregating features to prepare for the final classification decision; (4) output layer: finally outputting the probability of each category through a fully connected layer and a softmax activation function, i.e. determining whether there is a median nerve lesion in the image.
[0058] Example 2
[0059] This invention provides a median nerve disease prediction model, which is obtained using the construction method in Example 1.
[0060] Example 3
[0061] This invention provides a median nerve volume ultrasound diagnostic system, comprising:
[0062] The data acquisition module is used to acquire a continuous sequence of transverse images from the patient's wrist to the mid-forearm.
[0063] The processing module is used to perform artificial intelligence recognition on the acquired raw ultrasound images, automatically identify the median nerve portion in the raw ultrasound images, optimize the median nerve images, virtually generate a complete three-dimensional image of the median nerve, and perform perspective processing on the raw ultrasound images of the anatomical structures surrounding the nerve to obtain the processed image.
[0064] The model building module is used to build a predictive model for median nerve diseases;
[0065] The diagnostic module is used to diagnose whether median nerve lesions exist in processed images using a median nerve disease prediction model.
[0066] Furthermore, the median nerve disease prediction model is obtained by the following method:
[0067] S1: Median nerve segmentation
[0068] 1) Obtain continuous cross-sectional slices of three-dimensional reconstruction images of the wrist to mid-forearm from 200 subjects with median nerve lesions and 200 healthy subjects, and annotate the outline of the median nerve in each slice. The annotation includes the position and shape information of the median nerve on each slice. All subjects were required to have no acute trauma, fracture, congenital malformation or other lesions in their wrist and forearm.
[0069] 2) Use an appropriate segmentation algorithm to segment the sliced image;
[0070] 3) Overlay the segmented median nerve contour onto the original slice image and reprocess to obtain a three-dimensional reconstruction image of the wrist to mid-forearm with the median nerve contour attached.
[0071] S2: Classification of median nerve lesions
[0072] 1) Each 3D reconstruction image with an attached median nerve outline will be labeled with a category indicating whether the image contains a median nerve lesion;
[0073] 2) Use an appropriate classification algorithm to classify the 3D reconstructed image. Based on the image features and median nerve contour information, automatically classify the image into different categories, namely, those with median nerve lesions and those without median nerve lesions.
[0074] Furthermore, the segmentation algorithm is the U-Net segmentation algorithm based on convolutional neural networks.
[0075] Furthermore, the U-Net algorithm operates in the following steps: (1) Shrinking path: Through repeated convolution and pooling layer sequences, the network gradually captures and reduces the dimensions of the image, thereby extracting key features in the image; (2) Expanding path: Through the above sampling and convolution operations, the size and details of the image are gradually restored. In this process, skip connections are used to fuse the feature map of the shrinking path with the corresponding layer of the expanding path to ensure that important location information is not lost; (3) Final segmentation map: Through a 1x1 convolutional layer, the deep features are mapped to the category prediction of each pixel, and a segmentation map with the same size as the original image is output, where each pixel is labeled as the median nerve or the background.
[0076] Furthermore, the classification algorithm is ResNet3D, a classification algorithm based on 3D convolutional neural networks.
[0077] Furthermore, the ResNet3D classification algorithm includes the following steps: (1) 3D convolutional layer: using 3D convolution to process the input image and capture features in depth, height and width; (2) residual connection: maintaining the depth of the network through residual modules, while avoiding the gradient vanishing problem during training. These modules ensure that features are not destroyed when they are passed from one layer to another; (3) feature pooling: using global average pooling to reduce overfitting and aggregating features to prepare for the final classification decision; (4) output layer: finally outputting the probability of each category through a fully connected layer and a softmax activation function, i.e. determining whether there is a median nerve lesion in the image.
[0078] Example 4
[0079] This invention provides an artificial intelligence-based median nerve volumetric ultrasound diagnostic device, comprising an automatic scanning and moving device 1, a coupling medium container 2, a portable ultrasound control device 3, and a three-dimensional ultrasound probe 4. The automatic scanning and moving device 1 is connected to the three-dimensional ultrasound probe 4 and can drive the three-dimensional ultrasound probe 4 to move. The coupling medium container 2 is provided with a groove 21 for holding the coupling medium. The front side of the coupling medium container 2 is covered with a transparent cover 5. The three-dimensional ultrasound probe 4 is disposed between the coupling medium container 2 and the transparent cover 5. Both the automatic scanning and moving device 2 and the three-dimensional ultrasound probe 4 are connected to the portable ultrasound control device 3.
[0080] The automatic scanning and moving device 1 is equipped with a drive motor and a threaded rod 11. One side of the three-dimensional ultrasonic probe 4 is connected to the threaded rod 11. The drive motor drives the threaded rod 11 to rotate, thereby driving the ultrasonic probe 4 to move.
[0081] The automatic scanning mobile device 1 is installed on the coupling medium container 2 and is integrally formed with the coupling medium container 2 and the transparent cover 5.
[0082] The portable ultrasonic control device 3 includes a screen area 31, a keyboard area 32, and a handle 33.
[0083] The coupling medium container 4 is equipped with a power button 6, an on / off switch 7, and an emergency stop button 8.
[0084] The automatic scanning mobile device 1, coupling medium container 2, and transparent cover 5 are all made of ABS plastic. In the absence of water, the total weight of the one-piece automatic scanning mobile device 1, coupling medium container 2, and transparent cover 5 can be as low as about 11kg. The portable ultrasound control device 3 is equipped with a portable handle, making the entire diagnostic device easy to carry.
[0085] The wrist to mid-forearm section of the human body is placed in the groove 21 of the coupling medium container 2 and immersed in the coupling medium. The automatic scanning and moving device 1 is activated by the switch 7. The drive motor and the threaded rod 11 drive the three-dimensional ultrasound probe 4 to perform vertical scanning and moving of the wrist to mid-forearm section of the human body to form a continuous sequence of transverse images. The volume data obtained from the scanning is then transmitted to the handheld portable ultrasound control device 3 via DICOM. The handheld portable ultrasound control device 4 performs artificial intelligence recognition on the ultrasound images, automatically identifies the median nerve portion in the original ultrasound image, optimizes the median nerve image, and virtually creates a complete three-dimensional image of the median nerve. The original ultrasound image of the anatomical structures around the nerve is processed by perspective. Based on the median nerve volume ultrasound diagnostic system as described in Example 3 on the handheld portable ultrasound control device 3, a diagnosis of whether there is a lesion in the median nerve can be obtained with the assistance of artificial intelligence.
[0086] It should be understood that the present invention is not limited to the specific embodiments described above, as variations can be made to the specific embodiments and they still fall within the scope of the appended claims.
Claims
1. A method for constructing a median nerve disease prediction model, comprising the following steps: S1: Median nerve segmentation 1) Obtain continuous cross-sectional slices of three-dimensional reconstruction images of the wrist to mid-forearm from 200 subjects with median nerve lesions and 200 healthy subjects, and annotate the outline of the median nerve in each slice. The annotation includes the position and shape information of the median nerve on each slice. All subjects were required to have no acute trauma, fracture, congenital malformation or other lesions in their wrist and forearm. 2) Use an appropriate segmentation algorithm to segment the sliced image; 3) Overlay the segmented median nerve contour onto the original slice image and reprocess to obtain a three-dimensional reconstruction image of the wrist to mid-forearm with the median nerve contour attached. S2: Classification of median nerve lesions 1) Each 3D reconstruction image with the median nerve outline will be labeled with a category indicating whether the image contains carpal tunnel syndrome; 2) Use an appropriate classification algorithm to classify the 3D reconstructed image. Based on the image features and median nerve contour information, automatically classify the image into different categories, namely, those with median nerve lesions and those without median nerve lesions.
2. The construction method according to claim 1, wherein the segmentation algorithm is a U-Net segmentation algorithm based on a convolutional neural network.
3. The construction method according to claim 1, wherein the classification algorithm is ResNet3D, a classification algorithm based on a 3D convolutional neural network.
4. A median nerve disease prediction model, which is obtained by the construction method described in any one of claims 1-3.
5. A median nerve volume ultrasound diagnostic system, comprising: The data acquisition module is used to acquire a continuous sequence of transverse images from the patient's wrist to the mid-forearm. The processing module is used to perform artificial intelligence recognition on the acquired raw ultrasound images, automatically identify the median nerve portion in the raw ultrasound images, optimize the median nerve images, virtually generate a complete three-dimensional image of the median nerve, and perform perspective processing on the raw ultrasound images of the anatomical structures surrounding the nerve to obtain the processed image. The model building module is used to build a predictive model for median nerve diseases; The diagnostic module is used to diagnose whether median nerve lesions exist in processed images using a median nerve disease prediction model.
6. The diagnostic system according to claim 5, wherein the median nerve disease prediction model is obtained by the following method: S1: Median nerve segmentation 1) Obtain continuous cross-sectional slices of three-dimensional reconstruction images of the wrist to mid-forearm from 200 subjects with median nerve lesions and 200 healthy subjects, and annotate the outline of the median nerve in each slice. The annotation includes the position and shape information of the median nerve on each slice. All subjects were required to have no acute trauma, fracture, congenital malformation or other lesions in their wrist and forearm. 2) Use an appropriate segmentation algorithm to segment the sliced image; 3) Overlay the segmented median nerve contour onto the original slice image and reprocess to obtain a three-dimensional reconstruction image of the wrist to mid-forearm with the median nerve contour attached. S2: Classification of median nerve lesions 1) Each 3D reconstruction image with the median nerve outline will be labeled with a category indicating whether the image contains carpal tunnel syndrome; 2) Use an appropriate classification algorithm to classify the 3D reconstructed image. Based on the image features and median nerve contour information, automatically classify the image into different categories, namely, those with median nerve lesions and those without median nerve lesions.
7. The diagnostic system according to claim 6, wherein the segmentation algorithm is a U-Net segmentation algorithm based on a convolutional neural network.
8. The diagnostic system according to claim 6, wherein the classification algorithm is a ResNet3D classification algorithm based on a 3D convolutional neural network.
9. A median nerve volumetric ultrasound diagnostic device based on artificial intelligence, comprising an automatic scanning and moving device (1), a coupling medium container (2), a handheld portable ultrasound control device (3), and a three-dimensional ultrasound probe (4). The automatic scanning and moving device (1) is connected to the three-dimensional ultrasound probe (4) and can drive the three-dimensional ultrasound probe (4) to move. The coupling medium container (2) is provided with a groove (21) for holding the coupling medium. The front side of the coupling medium container (2) is covered with a transparent cover (5). The three-dimensional ultrasound probe (4) is disposed between the coupling medium container (2) and the transparent cover (5). The automatic scanning and moving device (2) and the ultrasound probe (4) are both connected to the handheld portable ultrasound control device (3).
10. The diagnostic device according to claim 9, wherein the automatic scanning moving device 1 is provided with a drive motor and a threaded rod (11), one side of the three-dimensional ultrasound probe (4) is connected to the threaded rod (11), and the drive motor drives the threaded rod (11) to rotate, thereby driving the three-dimensional ultrasound probe (4) to move.
11. The diagnostic device according to claim 9 or 10, wherein the automatic scanning moving device 1 is mounted on the coupling medium container (2) and integrally formed with the coupling medium container (2) and the transparent cover (5).
12. The diagnostic device according to claim 9 or 10, wherein the portable ultrasound control device 3 includes a screen area (31), a keyboard area (32), and a handle (33).
13. The diagnostic device according to claim 9 or 10, wherein the coupling medium container (4) is provided with a power button (6), an on / off switch (7) and an emergency stop button (8).