Interpreting Stroke-Impaired Electromyography Patterns through Explainable Artificial Intelligence