The impact of feature extraction for the classification of amyotrophic lateral sclerosis among neurodegenerative diseases and healthy subjects

• Published in: Biomedical signal processing and control Vol. 31; pp. 288 - 294
• Main Author: Bilgin, Suleyman
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2017
• Subjects: Amyotrophic lateral sclerosis; Discrete wavelet transform; Linear discriminant analysis; Naïve bayes classification; Neurodegenerative diseases; Amyotrophic lateral sclerosis; Discrete wavelet transform; Naïve bayes classification; Linear discriminant analysis; Neurodegenerative diseases
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2016.08.016

•The Compound Force Signal (CFS) is defined at first in the literature. So, one minute records of CFS will be enough for the discrimination of ALS rather than five-minutes records.•Especially ALS vs. Control discrimination has good performance as 90.93%.•The best features are obtained in ALS disease using DWT. The D5 (4.6875–9.375Hz) is a significant feature for all classifications.•The dual features extracted from CFS based on portable foot-switch system in a clinic can provide an important contribution with other clinical indexes for discrimination of ALS. Neurodegenerative diseases (NDD) including Amyotrophic Lateral Sclerosis (ALS), Parkinson’s disease (PD) and Huntington disease (HD) can be defined as the degeneration in the structure of neurons in human body. It is mentioned in the related literature that NDD may cause various clinical symptoms disrupting gait dynamics. The characterization of gait analysis is crucial for early diagnosis, efficient treatment planning and monitoring of ALS progression and other NDD. The database consisting of 64 one-minute recordings of Compound Force Signal (CFS) obtained from 13 ALS, 15 PD, 20 HD and 16 healthy subjects was used in the study. CFS is the composition of force signals for both left and right feet of each subject during the gait. CFS was decomposed for determination of features using 6-level Discrete Wavelet Transform (DWT) with different wavelets in the study. The obtained features were evaluated using the means of 20-trials for five-fold cross-validation (FFCV) in Linear Discriminant Analysis (LDA) and Naïve Bayesian Classifier (NBC). As a result, D5 (4.6875–9.375Hz) in all classifications, D4 (9.375–18.75Hz) in ALS vs. PD, ALS vs. PD+HD and ALS vs. Co+PD+HD classifications while D2 (37.5–75Hz) and D6 (2.3438-4.6875Hz) in ALS vs. Co. and ALS vs. HD classifications were determined as the most significant frequency bands in CFS for discrimination of ALS among healthy and other NDD subjects in the end of the study.