Domain Contrast Network for cross-muscle ALS disease identification with EMG signal. (April 2023)

• Record Type: Journal Article
• Authors: Zhang, Haoyu; Liu, Yan; Qing, Zhongfei; He, Ji; Teng, Shenghua; Wang, Xujian; Hao, Chenxu; Zhang, Shuo; Fan, Dongsheng; Su, Guiping
• Is Part Of: Biomedical signal processing and control. Volume 82(2023)
• Publication Date: 2023-04
• Subjects: Amyotrophic Lateral Sclerosis (ALS) -- Electromyography (EMG) -- Deep learning -- Domain contrast -- Loss function; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2023.104582
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: As an efficient means of Amyotrophic Lateral Sclerosis (ALS) diagnosis in clinical practice, needle Electromyography (EMG) is often used to sample data from different muscle parts for ALS diagnosis. Although EMG signals from different muscle parts have different effects on the diagnosis of ALS, there are common features of neurogenic injury for cross-muscle parts. It can reduce the patient's pain and improve the accuracy and efficiency of ALS disease diagnosis for physicians based on more sensitive muscle parts. In this paper, we propose a novel Domain Contrast Network (DCN) to extract common features of neurogenic injury for cross-muscle ALS disease identification. First, a domain contrast pre-training framework (DCP) is proposed to reduce the differences in the distribution of data from different muscle parts to learn more domain-invariant embeddings. Second, two loss functions are introduced to simultaneously reduce the difference between two domain distributions and increase the distance between ALS samples and normal samples. Finally, a new classifier is presented to classify the obtained embeddings. Experimental results demonstrate the efficiency and robustness of the proposed method on the cross-muscle ALS disease identification with EMG data from different individuals, different devices, and different human races. The proposed method will be useful in exploring more sensitive muscle parts for early ALS disease identification in clinical applications. Highlights: Differences in the distribution of cross-muscle EMG data are reduced. Common features of neurogenic injury are extracted. Performance of cross-individual ALS identification is further improved.