Omics feature learning for cross individual ALS disease identification with EMG signal

Electromyography (EMG) analysis is an important means to assist the diagnosis of amyotrophic lateral sclerosis (ALS) and other neuromuscular diseases. Following the idea of omics analysis, this paper adopts omics feature extraction and hybrid feature selection strategies to identify ALS with EMG signal in cross-individual scenarios. Specifically, multiple features from time domain, frequency domain, wavelet domain and nonlinear dynamics are extracted to capture the intrinsic characteristics of the EMG signal to the greatest extent. And then, a hybrid feature selection strategy is designed which combines certain basic feature selection methods in two rounds to screen the discriminative features for classification. Finally, the EMG signals of ALS and the normal control are classified by linear discriminant analysis. Experiments are carried out on the public and private datasets to verify the effectiveness of the proposed method.