Comprehensive Summary
This study evaluates the use of AI-assisted quantitative muscle ultrasound for diagnosing carpal tunnel syndrome (CTS). Researchers analyzed ultrasonographic images from 74 hands, including 47 with CTS and 27 controls, to measure echo intensity (EI) in thenar and hypothenar muscles. Conventional grayscale analysis calculated mean EI and standard deviation, while machine learning (ML) utilized 176 radiomic features extracted via the Pyradiomics platform. Four ML models were tested: Random Forest, AdaBoost, Support Vector Classifier (SVC), and Extreme Gradient Boosting (XGB), with and without Recursive Feature Elimination (RFE). The best-performing model, SVC with RFE, achieved an AUC of 0.89, precision of 0.75, recall of 0.88, and F1-score of 0.81, outperforming traditional methods. Key features like robust mean absolute deviation, interquartile range, and small-area emphasis were influential in AI-based classification. Limitations include manual ROI selection and lack of subgroup analysis by age or sex.
Outcomes and Implications
AI-enhanced ultrasound offers significant advancements in diagnosing CTS, particularly in distinguishing between normal and affected hands using textural muscle features. The SVC with RFE model delivered the highest diagnostic performance, improving accuracy by 0.13 over traditional grayscale analysis. This model's strong recall and F1-score indicate balanced performance in detection and classification, potentially reducing reliance on invasive tests like electromyography. However, manual ROI selection may introduce bias, and the study's binary classification limits finer severity grading. Future improvements could include deep neural networks for automated ROI detection and expanded datasets for nuanced classification.