Multifrequency Electrical Impedance Myography Enhanced with Machine Learning for Screening Patients with Neuromuscular Disorders

Abstract Purpose We evaluated surface electrical impedance myography (EIM) enhanced with machine learning to serve as a new office-based screening tool for neuromuscular disease Methods EIM of nine muscles was successfully performed in 119 adults and 111 children (approximately half healthy and half diseased) for a total of 3158 individual muscle measurements. Multifrequency data were processed with feature engineering and classification. Nested cross-validation assessed performance, with muscle-based predictions aggregated to participant via majority voting Results Single-feature analyses showed moderate-to-good discrimination (area-under-ROC curve of 0.62–0.73). When multifrequency features were used, participant-based logistic regression and extra trees ensemble models achieved 84% accuracy with 88.1% sensitivity in adults and 93% accuracy and 94.6% sensitivity in children. Beyond classification, regression using EIM predicted muscle strength with R 2 = 0.49, outperforming single-frequency correlations Conclusion These results demonstrate that machine learning-enhanced EIM can successfully distinguish individuals with neuromuscular disease from healthy individuals. Nevertheless, further studies in larger populations of people would help advance this technology to the point that it could serve as a convenient, office-based screening tool for neuromuscular disease.