Intelligent hand neuropathy detection method based on machine learning
By employing a machine learning-based intelligent hand neuropathy detection method, which utilizes gesture image data and key point detection to construct a feature set and train a model, the method solves the subjectivity problem of traditional diagnostic methods and achieves automated, accurate diagnosis and assessment of hand neuropathies.
Patent Information
- Application Number
- CN202510916847.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-03
- Publication Date
- 2025-10-28
AI Technical Summary
Clinically, the diagnosis of peripheral nerve lesions of the hand mainly relies on the doctor's experience and manual observation, which cannot achieve precise quantification and automatic analysis of complex features.
A machine learning-based intelligent hand neuropathy detection method is adopted. By acquiring the subject's gesture image data, key point detection and posture estimation are performed, the three-dimensional coordinate information of key points in the hand is extracted, a feature set is constructed, and logistic regression and support vector machine models are used for training, the hyperparameter configuration is optimized, and it is deployed on a mobile terminal for remote assisted diagnosis.
It enables automatic identification and diagnosis of hand neuropathy, providing more objective and accurate assessment results, reducing reliance on large amounts of clinical data, and is suitable for application scenarios where clinical data is limited.
Smart Images
Figure CN120837014A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an intelligent hand nerve lesion detection method based on machine learning, belonging to the field of computer image processing. Background Technology
[0002] Peripheral neuropathy of the hand refers to pathological conditions affecting the function of the nerves in the hand, mainly including damage to the radial, ulnar, and median nerves. These nerve injuries can manifest as decreased sensation, muscle weakness, or even motor dysfunction. For example, radial nerve injury usually presents as loss of wrist and finger extension function, ulnar nerve injury presents as limited sensory and motor function in the little and ring fingers, and median nerve injury often presents as sensory and motor dysfunction in the thumb, index, and middle fingers.
[0003] Currently, the clinical diagnosis of peripheral neuropathy of the hand relies primarily on the experience and manual observation of physicians. This traditional method is inherently subjective, with inconsistent assessment standards among different doctors, resulting in a lack of objectivity and consistency in diagnostic results, making it difficult to ensure accuracy and stability. Hand nerve function is closely related to hand movement patterns, especially the degree of movement of the fingers and hand joints, as well as the relative angles between joints, which can reflect the functional state of specific nerves. However, traditional assessment methods cannot achieve precise quantification and automated analysis of these complex characteristics, thus limiting their application in early diagnosis, disease monitoring, and prognostic assessment. Summary of the Invention
[0004] The technical problem that this invention aims to solve is that the clinical diagnosis of peripheral nerve lesions of the hand mainly relies on the doctor's experience and manual observation, which cannot achieve accurate quantification and automatic analysis of complex features.
[0005] To address the aforementioned technical problems, the present invention discloses an intelligent hand nerve lesion detection method based on machine learning, characterized by comprising the following steps:
[0006] Step 1: Obtain the hand gesture image data of the subjects. Each subject performs N standard hand gestures in turn, and M images of each hand gesture are collected repeatedly, N≥1, M≥1;
[0007] Step 2: Perform key point detection and pose estimation on the image data to extract the three-dimensional coordinate information of the key points of the hand;
[0008] Step 3: Based on the obtained hand key points, accurately calculate the geometric features closely related to hand movement function, construct a feature set, and perform feature filtering on the features in the feature set;
[0009] Step 4: Based on the feature set, build a machine learning model: train the model using logistic regression and support vector machine, and optimize the hyperparameter configuration using the Scikit-learn framework through grid search and 5-fold cross-validation.
[0010] Step 5: Divide the dataset into training and test sets. Calculate the accuracy, sensitivity, and specificity of each model on the test set, and plot the confusion matrix to evaluate the actual performance of the model.
[0011] Step 6: Deploy the model on a mobile terminal and integrate it directly into the remote assisted diagnosis platform for use in primary healthcare institutions or remote health services.
[0012] Preferably, in step 1, the subjects include patients with unilateral radial nerve, ulnar nerve or median nerve injury and healthy volunteers.
[0013] Preferably, in step 1, all images are captured by a smartphone, with the shooting angle located slightly above the hand on the radial or ulnar side, and the shooting distance is 40–50 cm.
[0014] Preferably, in step 2, if the key points of the complete gesture cannot be identified in the gesture image, the gesture image is removed.
[0015] Preferably, in step 3, the geometric features include radial nerve injury-related features, ulnar nerve injury-related features, and median nerve injury-related features.
[0016] Preferably, in step 3, the radial nerve injury-related features include:
[0017] The angle of thumb metacarpophalangeal joint flexion: the angle formed by the wrist joint, the thumb carpal metacarpophalangeal joint, and the thumb metacarpophalangeal joint.
[0018] Flexion angles of the metacarpophalangeal joints of each finger: with the wrist joint as the reference point, the angles formed by the lines connecting the metacarpophalangeal joints of the index finger, middle finger, and ring finger are respectively.
[0019] Distance from fingertip to wrist joint: Calculate the Euclidean distance between the fingertip of the index finger, the fingertip of the middle finger, and the fingertip of the ring finger and the wrist joint respectively;
[0020] The ulnar nerve injury-related features include:
[0021] The angle between the middle finger and the ring and little fingers: the angle formed by the tips of the middle finger, the ring finger, and the little finger;
[0022] The flexion angle of the interphalangeal joint of the little finger further includes:
[0023] Proximal interphalangeal joint angle: the angle formed by the metacarpophalangeal joint of the little finger, the proximal interphalangeal joint of the little finger, and the distal interphalangeal joint of the little finger;
[0024] Distal interphalangeal joint angle: the angle formed by the proximal interphalangeal joint of the little finger, the distal interphalangeal joint of the little finger, and the tip of the little finger;
[0025] The median nerve injury-related features include:
[0026] Thumb-to-palm angle: the spatial angle formed by the three points: the tip of the index finger, the metacarpophalangeal joint of the thumb, and the carpometacarpophalangeal joint of the thumb.
[0027] Distance between the tips of the thumb and index finger: Calculate the distance between the tips of the thumb and index finger.
[0028] Preferably, in step 3, when performing feature screening, all features are evaluated using the ROC curve analysis method, the area under the curve of the feature is calculated, and features with strong discriminative ability and an area under the curve of not less than 0.8 are retained.
[0029] Preferably, in step 4, for the detection of radial nerve injury and median nerve injury, an SVM model is selected as the classifier. The model hyperparameters are tuned by constructing a parameter network and using 5-fold cross-validation to determine the optimal parameter combination as follows:
[0030] C=1,γ=scale,kernel=linear,degree=1
[0031] Where: C is the penalty coefficient, used to control the model's tolerance for misclassification; γ is the kernel function coefficient, used to adjust the influence of RBF or polynomial kernel; kernel indicates the kernel function type, which is a linear kernel here; degree indicates the order of the polynomial kernel function.
[0032] The SVM classification model trained with optimal parameters has the following linear decision function form:
[0033] f(x) = w T x+b
[0034] Where w is the feature weight vector and b is the bias term, both of which are automatically learned during the training process.
[0035] During the testing phase, this invention uses a confusion matrix to evaluate classification performance and calculates accuracy, sensitivity, specificity, and false negative rate (FNR), defined as follows:
[0036]
[0037] Where TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative;
[0038] The model demonstrated high accuracy and stable sensitivity and specificity in the radial nerve injury and median nerve injury identification tasks.
[0039] For ulnar nerve injury detection, a logistic regression model with L1 regularization was used to construct the classifier. The penalty coefficient C, L1 ratio l1_ratio, and penalty type were jointly tuned through cross-validation and grid search. The optimal parameter combination was determined as follows:
[0040] C=0.1, penalty=l1_ratio, l1_ratio=0.01, solver=liblinear
[0041] Where: C is the penalty coefficient, penalty represents the regularization method, l1_ratio represents the ratio between L1 and L2 regularization terms, and solver represents the solver type;
[0042] Under optimal configuration, the classification function of logistic regression takes the following form:
[0043] f(x)=σ(w T x+b)
[0044] Where w is the weight coefficient vector and b is the bias term. This represents the sigmoid activation function;
[0045] The model output is compared with the true labels to construct a binary classification confusion matrix:
[0046]
[0047] Further calculations were performed on accuracy, sensitivity, specificity, and false negative rate (FNR).
[0048] This invention proposes a machine learning-based intelligent method for detecting hand neuropathy. This method efficiently extracts the location information of 21 key points related to peripheral nerves in the hand using Python's MediaPipe library. These key points not only reflect the structural features of the hand but also its movement patterns. This invention not only extracts the spatial coordinates of these key points but also calculates comprehensive motion features such as relative angles between joints and distances between fingers, providing a more comprehensive description of the neurological function of the hand. For example, specific movement patterns and abnormal changes in key points can indirectly reflect damage to the radial, ulnar, and median nerves: radial nerve damage may lead to impaired wrist extension, causing abnormalities in the position and angle of related key points; ulnar nerve damage may cause limited movement of the little finger, affecting the position and angle of the little finger joint; and median nerve damage manifests as limited movement of the thumb and other fingers, leading to abnormal distances between fingers and joint angles.
[0049] By analyzing the dynamic changes of key points in the hand, this invention effectively extracts features related to various nerve injuries, enabling automatic identification and diagnosis of radial, ulnar, and median nerve injuries. This intelligent detection method not only reduces reliance on large amounts of clinical data but also provides more objective and accurate assessment results, exhibiting high stability and reliability, making it particularly suitable for applications where clinical data is limited. Attached Figure Description
[0050] Figure 1 This is an overall overview diagram of the intelligent hand nerve lesion detection method based on machine learning described in this invention;
[0051] Figure 2 This illustration shows the key point information related to 21 peripheral nerves of the hand extracted using the MediaPipe library in this invention;
[0052] Figure 3 The sample examples illustrate normal and abnormal hand gestures;
[0053] Figure 4 It is a full-process architecture diagram that integrates the model into the actual mini-program. Detailed Implementation
[0054] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.
[0055] This invention discloses a machine learning-based intelligent method for detecting hand neuropathy, such as... Figure 1As shown, it includes the following steps.
[0056] Step 1: Acquiring Hand Gesture Image Data from Subjects. During the image data acquisition phase, subjects included patients with unilateral radial, ulnar, or median nerve injuries, as well as healthy volunteers. All images were acquired using smartphones at a resolution of 1980×1080, with the camera positioned slightly above the hand on the radial or ulnar side, at a distance of 40–50 cm. Each subject sequentially performed three standard hand gestures (full finger abduction, finger adduction, little finger lateralization, and the "OK" gesture), with four images captured for each gesture.
[0057] Step 2: Image Data Preprocessing and Keypoint Extraction. The MediaPipe Hands library provided by Google is used to perform keypoint detection and pose estimation on the image data, automatically extracting the 3D coordinates of 21 keypoints. If a complete hand gesture keypoint cannot be identified in an image, the image will be discarded to ensure the quality of the training data. The coordinates are used for subsequent geometric feature calculations, including joint angles, fingertip distances, and palm-finger angles.
[0058] Step 3: Constructing the feature set and feature selection. Based on clinical anatomical knowledge and nerve innervation areas, this invention uses the MediaPipe library to automatically extract the coordinates of 21 key points on the hand. The locations and numbers of these key points have been... Figure 2 Detailed annotations are provided.
[0059] Based on the above key points, the geometric features closely related to hand motor function were accurately calculated, including:
[0060] (1) Characteristics related to radial nerve injury:
[0061] The angle of thumb metacarpophalangeal joint flexion: the angle formed by the wrist joint (key point 0), the thumb wrist metacarpophalangeal joint (key point 1), and the thumb metacarpophalangeal joint (key point 2), which is used to reflect the extension function of the thumb.
[0062] The flexion angle of each finger's metacarpophalangeal joint: With the wrist joint (key point 0) as the reference point, the angles formed by the lines connecting the index finger's metacarpophalangeal joint (key point 5), the middle finger's metacarpophalangeal joint (key point 9), and the ring finger's metacarpophalangeal joint (key point 13) are used to measure the degree of finger extension.
[0063] Distance from fingertip to wrist joint: The Euclidean distance between the fingertip of the index finger (key point 8), the fingertip of the middle finger (key point 12), and the fingertip of the ring finger (key point 16) and the wrist joint (key point 0) is calculated to quantitatively assess whether the fingers can be fully extended.
[0064] (2) Characteristics related to ulnar nerve injury:
[0065] The angle between the middle finger and the ring and little fingers: The angle formed by the three points of the middle finger tip (key point 12), the ring finger tip (key point 16), and the little finger tip (key point 20) is used to assess the finger adduction ability and reflect the function of the palmar interosseous muscles.
[0066] The flexion angle of the interphalangeal joint of the little finger includes:
[0067] Proximal interphalangeal joint angle: the angle formed by the three points: the metacarpophalangeal joint of the little finger (key point 17), the proximal interphalangeal joint of the little finger (key point 18), and the distal interphalangeal joint of the little finger (key point 19).
[0068] Distal interphalangeal joint angle: The angle formed by the three points: the proximal interphalangeal joint of the little finger (key point 18), the distal interphalangeal joint of the little finger (key point 19), and the fingertip of the little finger (key point 20).
[0069] (3) Characteristics related to median nerve injury:
[0070] Thumb opposition angle: The spatial angle formed by the three points of the index finger tip (key point 8), thumb metacarpophalangeal joint (key point 2), and thumb carpometacarpophalangeal joint (key point 1), used to assess thumb opposition ability.
[0071] Distance between the tips of the thumb and index finger: Calculate the distance between the tip of the thumb (key point 4) and the tip of the index finger (key point 8) to determine whether the two fingers can form a fine pinching motion (i.e., the "OK gesture").
[0072] During feature selection, the above features were evaluated using ROC curve analysis to calculate the area under the curve (AUC) of each feature. Features with strong discriminative power (AUC ≥ 0.8) were retained to ensure that the selected features have high diagnostic value for radial, ulnar, and median nerve injuries and will be used as the basis for classification in subsequent models.
[0073] Step 4: Build a classification model, train it, and evaluate it. Based on the feature set, build a machine learning model: train it using logistic regression (LR) and support vector machine (SVM) models, and optimize the hyperparameter configuration using the Scikit-learn framework through grid search and 5-fold cross-validation.
[0074] For the detection of radial nerve injury and median nerve injury, this invention uses a linear kernel Support Vector Machine (SVM) model as the classifier. The model hyperparameters were fine-tuned by constructing a parameter network and using 5-fold cross-validation to determine the optimal parameter combination as follows:
[0075] C=1,γ=scale,kernel=linear,degree=1
[0076] Where: C is the penalty coefficient, used to control the model's tolerance for misclassification; γ is the kernel function coefficient, used to adjust the influence of RBF or polynomial kernel; kernel indicates the kernel function type, which is a linear kernel here; degree indicates the order of the polynomial kernel function.
[0077] The SVM classification model trained with optimal parameters has the following linear decision function form:
[0078] f(x) = w T x+b
[0079] Where w is the feature weight vector and b is the bias term, both of which are automatically learned during the training process.
[0080] During the testing phase, this invention uses a confusion matrix to evaluate classification performance and calculates accuracy, sensitivity, specificity, and false negative rate (FNR), defined as follows:
[0081]
[0082] Where TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative.
[0083] The model demonstrated high accuracy and stable sensitivity and specificity in the radial nerve injury and median nerve injury identification tasks.
[0084] For ulnar nerve injury detection, this invention further employs a logistic regression model with L1 regularization to construct a classifier. This method has advantages such as strong model sparsity and good feature interpretability, making it particularly suitable for small-sample classification scenarios in medical settings. This model uses cross-validation and grid search to jointly fine-tune hyperparameters such as the penalty coefficient C, L1 ratio l1_ratio, and penalty type penalty. The final optimal parameter combination is as follows:
[0085] C=0.1, penalty=l1_ratio, l1_ratio=0.01, solver=liblinear
[0086] Where: C is the penalty coefficient, used to control the model's tolerance for misclassification; penalty represents the regularization method, here L1 regularization is chosen to achieve feature sparsity; l1_ratio represents the ratio between L1 and L2 regularization terms, set to 0.01, indicating that L1 regularization is used almost entirely; solver represents the solver type, here the liblinear optimizer is used, suitable for small datasets with sparse features. Under the optimal configuration, the classification function of logistic regression takes the following form:
[0087] f(x)=σ(w T x+b)
[0088] Where w is the weight coefficient vector and b is the bias term. This represents the sigmoid activation function. After the model is trained, explicit coefficients and biases can be obtained, which facilitates subsequent interpretation of medical features and formulation of screening rules.
[0089] The model output is compared with the true labels to construct a binary classification confusion matrix:
[0090]
[0091] Further calculations of classification performance metrics, including:
[0092]
[0093] The logistic regression model constructed in this invention achieves high accuracy, stable sensitivity and specificity in the ulnar nerve injury identification task, while also having good model sparsity and medical interpretability, making it suitable for clinical application.
[0094] Step 5: Model Performance Evaluation. The dataset was divided into training and test sets in a 3:1 ratio. Accuracy, sensitivity, and specificity were calculated for each model on the test set, and a confusion matrix was plotted to evaluate the actual performance of the models. Experimental results show that the rule-based model achieved high accuracy (up to 98.2%) in the recognition of the three types of neural injuries, while the ML model exhibited higher sensitivity (up to 96%) in some tasks.
[0095] Step 6: System Adaptation and Deployment Recommendations. This invention is applicable to mobile terminals. Combined with the lightweight nature of MediaPipe Hands, it can be directly integrated into remote assisted diagnostic platforms such as WeChat mini programs and mobile apps for use in primary healthcare institutions or remote health services, possessing good deployment flexibility and promising prospects for widespread adoption.
[0096] This invention proposes an intelligent detection framework for peripheral neuropathy of the hand based on machine learning. Unlike general gesture recognition tasks, hand neuropathy has distinct medical characteristics and clinical manifestations, particularly in finger joint movement disorders and abnormal interfinite finger distances caused by specific nerve damage. These manifestations have clear medical significance; therefore, in the design process, this invention fully integrates professional knowledge in the medical field and defines several diagnostically valuable features for abnormal gestures that may be caused by damage to the radial, ulnar, and median nerves.
[0097] Using Python's MediaPipe library, this invention automatically and efficiently extracts the coordinates of 21 key points related to peripheral nerves in the hand. These key points correspond to areas such as hand joints, fingertips, and finger roots, and are detailed in the provided text. Figure 2 These key points reflect the structure and movement patterns of the hand. This invention accurately captures abnormal manifestations caused by injuries to the radial, ulnar, and median nerves by analyzing the distance between fingers, changes in the angle between joints, and hand movement patterns. For example, radial nerve injury typically manifests as loss of wrist and finger extension function, and abnormal positioning of related key points; ulnar nerve injury manifests as abnormal movement of the little and ring fingers, affecting the positional changes of corresponding key points; median nerve injury manifests as dysfunction of the thumb, index, and middle fingers, which can be accurately captured through changes in the distance between fingers and joint angles.
[0098] In the feature extraction stage, this invention accurately calculates medically relevant features such as finger joint angles and inter-finger distances, which are highly correlated with injuries to the radial, ulnar, and median nerves. Through in-depth data analysis and optimization, using ROC curve analysis, the contribution of each feature to diagnosis is clarified, and the core features with the highest diagnostic value are effectively selected by ranking feature importance.
[0099] In terms of model construction, this invention employs classic machine learning algorithms such as Support Vector Machine (SVM) and Logistic Regression (LR) to build an efficient and stable automatic detection model. Experiments show that this model can accurately identify abnormal gestures caused by injuries to the radial, ulnar, and median nerves without requiring a large amount of labeled data, achieving a classification accuracy of up to 96%, a sensitivity of 91.7%, and a specificity of 99%.
[0100] This invention significantly reduces the cost of medical data annotation and acquisition, and improves the model's generalization ability and diagnostic accuracy. It is particularly suitable for primary healthcare, telemedicine, and areas with scarce medical resources. The framework is clear and easy to understand, facilitating widespread application. It can effectively assist non-medical personnel in conducting preliminary screening and diagnosis of hand nerve disorders, significantly improving the diagnostic efficiency and service quality of primary healthcare institutions. It has significant clinical application value and broad prospects for promotion.
Claims
1. A machine learning-based intelligent method for detecting hand nerve lesions, characterized in that, Includes the following steps: Step 1: Obtain the hand gesture image data of the subjects. Each subject performs N standard hand gestures in turn, and M images of each hand gesture are collected repeatedly, N≥1, M≥1; Step 2: Perform key point detection and pose estimation on the image data to extract the three-dimensional coordinate information of the key points of the hand; Step 3: Based on the obtained hand key points, accurately calculate the geometric features closely related to hand movement function, construct a feature set, and perform feature filtering on the features in the feature set; Step 4: Based on the feature set, build a machine learning model: train the model using logistic regression and support vector machine, and optimize the hyperparameter configuration using the Scikit-learn framework through grid search and 5-fold cross-validation. Step 5: Divide the dataset into training and test sets. Calculate the accuracy, sensitivity, and specificity of each model on the test set, and plot the confusion matrix to evaluate the actual performance of the model. Step 6: Deploy the model on a mobile terminal and integrate it directly into the remote assisted diagnosis platform for use in primary healthcare institutions or remote health services.
2. The intelligent hand nerve lesion detection method based on machine learning as described in claim 1, characterized in that, In step 1, the subjects included patients with unilateral radial nerve, ulnar nerve or median nerve injury and healthy volunteers.
3. The intelligent hand nerve lesion detection method based on machine learning as described in claim 1, characterized in that, In step 1, all images were captured using a smartphone, with the shooting angle positioned slightly above the hand on the radial or ulnar side, at a distance of 40–50 cm.
4. The intelligent hand neuropathy detection method based on machine learning as described in claim 1, characterized in that, In step 2, if the key points of the complete gesture cannot be identified in the gesture image, the gesture image will be removed.
5. The intelligent hand neuropathy detection method based on machine learning as described in claim 1, characterized in that, In step 3, the geometric features include radial nerve injury-related features, ulnar nerve injury-related features, and median nerve injury-related features.
6. The intelligent hand nerve lesion detection method based on machine learning as described in claim 5, characterized in that, In step 3, the radial nerve injury-related features include: The angle of thumb metacarpophalangeal joint flexion: the angle formed by the wrist joint, the thumb carpal metacarpophalangeal joint, and the thumb metacarpophalangeal joint. Flexion angles of the metacarpophalangeal joints of each finger: with the wrist joint as the reference point, the angles formed by the lines connecting the metacarpophalangeal joints of the index finger, middle finger, and ring finger are respectively. Distance from fingertip to wrist joint: Calculate the Euclidean distance between the fingertip of the index finger, the fingertip of the middle finger, and the fingertip of the ring finger and the wrist joint respectively; The ulnar nerve injury-related features include: The angle between the middle finger and the ring and little fingers: the angle formed by the tips of the middle finger, the ring finger, and the little finger; The flexion angle of the interphalangeal joint of the little finger further includes: Proximal interphalangeal joint angle: the angle formed by the metacarpophalangeal joint of the little finger, the proximal interphalangeal joint of the little finger, and the distal interphalangeal joint of the little finger; Distal interphalangeal joint angle: the angle formed by the proximal interphalangeal joint of the little finger, the distal interphalangeal joint of the little finger, and the tip of the little finger; The median nerve injury-related features include: Thumb-to-palm angle: the spatial angle formed by the three points: the tip of the index finger, the metacarpophalangeal joint of the thumb, and the carpometacarpophalangeal joint of the thumb. Distance between the tips of the thumb and index finger: Calculate the distance between the tips of the thumb and index finger.
7. The intelligent hand nerve lesion detection method based on machine learning as described in claim 5, characterized in that, In step 3, when performing feature screening, all features are evaluated using the ROC curve analysis method, the area under the curve of the feature is calculated, and features with strong discriminative ability and an area under the curve of not less than 0.8 are retained.
8. The intelligent hand nerve lesion detection method based on machine learning as described in claim 5, characterized in that, In step 4, for the detection of radial nerve injury and median nerve injury, an SVM model was selected as the classifier. The model hyperparameters were tuned by constructing a parameter network and using 5-fold cross-validation to determine the optimal parameter combination as follows: C=1,γ=scale,kernel=linear,degree=1 Where: C is the penalty coefficient, used to control the model's tolerance for misclassification; γ is the kernel function coefficient, used to adjust the influence of RBF or polynomial kernel; kernel indicates the kernel function type, which is a linear kernel here; degree indicates the order of the polynomial kernel function. The SVM classification model trained with optimal parameters has the following linear decision function form: f(x)=w T x+b Where w is the feature weight vector and b is the bias term, both of which are automatically learned during the training process. During the testing phase, this invention uses a confusion matrix to evaluate classification performance and calculates accuracy, sensitivity, specificity, and false negative rate (FNR), defined as follows: Where TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative; The model demonstrated high accuracy and stable sensitivity and specificity in the radial nerve injury and median nerve injury identification tasks. For ulnar nerve injury detection, a logistic regression model with L1 regularization was used to construct the classifier. The penalty coefficient C, L1 ratio l1_ratio, and penalty type were jointly tuned through cross-validation and grid search. The optimal parameter combination was determined as follows: C=0.1, penalty=l1_ratio, l1_ratio=0.01, solver=liblinear Where: C is the penalty coefficient, penalty represents the regularization method, l1_ratio represents the ratio between L1 and L2 regularization terms, and solver represents the solver type; Under optimal configuration, the classification function of logistic regression takes the following form: f(x)=σ(w T x+b) Where w is the weight coefficient vector and b is the bias term. This represents the sigmoid activation function; The model output is compared with the true labels to construct a binary classification confusion matrix: Further calculations were performed on accuracy, sensitivity, specificity, and false negative rate (FNR).