Simulation-Grounded Deep Learning Approach for Discrimination of Sodium- and Chloride-Channel Myotonias from Electromyography

This is a preprint / author-submitted version of a manuscript submitted to the IEEE Journal of Biomedical and Health Informatics on 20 March 2026. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. Skeletal muscle channelopathies caused by variants in SCN4A and CLCN1 generate myotonic discharges detectable in intramuscular electromyography (iEMG). It remains unclear whether routine iEMG contains reliable subtype-specific signatures independent of the selected muscle and examination protocols as well as expert annotation. We propose an automated, simulation-grounded pipeline that classifies sodium- versus chloride-channel myotonia directly from routine iEMG, outputting calibrated subtype probabilities without manual event segmentation. The method converts iEMG to the time-frequency domain using a wavelet transform and classifies these representations with an ensemble of pretrained deep neural networks; training and evaluation are grouped at the patient level to prevent leakage. A complementary biophysical "digital twin" produces synthetic discharges to validate feature selection and saliency behavior. On held-out clinical recordings, the ensemble achieved a balanced accuracy of 0.82 and an area under the receiver operating characteristic curve of 0.89, with a Brier score of 0.14. Under selective prediction, higher confidence thresholds increased accuracy on retained cases at the expense of coverage. Saliency maps on synthetic data highlighted class-specific spectral signatures consistent with the model’s decision basis. These results suggest that routine iEMG contains discriminative subtype-specific spectral patterns and that calibrated deep ensembles can exploit them to support diagnostic workflows, for example, by prioritizing targeted genetic testing. To our knowledge, this is the first automated method to distinguish sodium- and chloride-channel non-dystrophic myotonias from routine clinical iEMG.