Abstract
Amyotrophic lateral sclerosis (ALS) is a rare, fatal, and irreversible disease that shares some key clinical features with radiculopathy, including muscle atrophy, muscle cramps, and fasciculation. The aim of this study was to find a reliable method to differentiate these two diseases. Machine learning was used to discover new clinical biomarkers for the differential diagnosis of ALS from radiculopathy using nerve conduction study (NCS) data from patients. Data preparation and feature selection were performed by a random forest classifier algorithm, as well as a confusion matrix tool for model selection. After selecting the minimum number of features and the best algorithm, grid search cross-validation was used to optimize the hyperparameters of the chosen algorithm. 77 features were ranked according to their importance. The results of 20 algorithms acting on 8 different groups of features showed that the best performance (accuracy, precision, recall, f-1 score) was obtained using 35 important features and the XGB algorithm, particularly for the recall parameter. Using the XGB algorithm, ALS patients could be identified with accuracy = 0.871, precision = 0.923, recall = 0.850, and f-1 score = 0.857. The XGB algorithm using 35 NCS features could differentiate radiculopathy from ALS in patients with high accuracy.