Machine learning for supporting diagnosis of amyotrophic lateral sclerosis using surface electromyogram

Needle electromyogram (EMG) is routinely used in clinical neurophysiology for examination of neuromuscular diseases. This study presents a noninvasive surface EMG method for supporting diagnosis of amyotrophic lateral sclerosis (ALS). Three diagnostic markers including the clustering index, the kurtosis of EMG amplitude histogram, and the kurtosis of EMG crossing-rate expansion, were used respectively to characterize surface EMG patterns recorded during different levels of voluntary muscle contraction. We then applied a linear discriminant analysis classifier to discriminate the ALS subjects from the neurologically intact subjects, using the statistics derived from all the three markers as input feature sets to the classifier. The method was tested in 10 ALS subjects and 11 neurologically intact subjects. Combination of the three surface EMG markers achieved 90% diagnostic sensitivity and 100% diagnostic specificity, which were higher than solely using a single surface EMG marker. Given the high diagnostic yield, the proposed surface EMG analysis can be used as a supplement to needle EMG examination in supporting the diagnosis of ALS.