Answers are grounded in this paper and cite their sources.
Neuromuscular disorders (NMDs) are clinically heterogeneous conditions which often require interpretation of muscle ultrasound by an expert for accurate diagnosis. In our study, we proposed a robust and sample efficient framework for the classification of neuromuscular disorders using muscle ultrasound images by combining Siamese neural network with traditional machine learning classifiers. The Siamese network is used as a deep feature extractor which learns discriminative embeddings based on similarity learning of after which embeddings are classified using multiple classical classifiers such as Support Vector Machine(SVM), Logistic Regression(LR),Random Forest(RF), K-Nearest Neighbors(KNN) and Multi Layer Perceptron(MLP).The proposed framework was evaluated on the ultrasound images of three types of muscle: Gastrocnemius Medialis (GM),Biceps Brachii (BB), and Tibialis Anterior (TA). A baseline CNN and transfer learning based transfer learning based EfficientNetB0 were trained independently on each muscle which was used in comparative analysis. All the models were evaluated by a patient level, class stratified train/test split, preventing data leakage. The results of experiments showed that the proposed Siamese network based framework achieved comparable output against the CNN baseline and outperformed transfer learning based baseline EfficientNetB0 model over the three muscles. Among all the classifiers the best performing model was Siamese + KNN (accuracy 0.940, 95% CI [0.899, 0.973]; F1 0.83; AUC 0.93). The results compare favorably with the SOTA (state-of-the-art) models mentioned int the literature. The findings demonstrated the effectiveness of Siamese framework based classifiers for neuromuscular ultrasound image classification and offers a practical and reliable solution for clinical decision support. This study provides a strong foundation for development of AI assisted tool for neuromuscular disorders with limited labeled data.
No takes yet. Share an insight, caveat, or question.
Khamparia et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: