Deep Temporal Learning Method for Amyotrophic Lateral Sclerosis Detection on EEG Signals

Received: 18 February 2026 | Revised: 21 March 2026, 4 April 2026, and 6 April 2026 | Accepted: 10 April 2026 | Online: 16 May 2026

Corresponding author: Suresh Annamalai

Abstract

Neurodegenerative disorders constitute a major global contributor to physical disability. In particular, Amyotrophic Lateral Sclerosis (ALS) is a type of neurodegenerative disease that greatly affects life expectancy because it damages nerve cell functions in the Central Nervous System (CNS). Deep Learning (DL) methods have achieved positive results, specifically in the diagnosis of diseases influenced by neurological disorders and in classifying and identifying neurological or psychiatric disorders. This study presents a Deep Temporal Learning Framework for Amyotrophic Lateral Sclerosis Detection (DTLF-ALSD) approach that uses EEG signals. The proposed workflow begins with comprehensive signal preprocessing, including bandpass filtering to remove noise and artifacts, normalization to standardize amplitude distributions, and visual inspection to ensure signal quality enhancement. After preprocessing, three temporal DL architectures, TemporalEEG, TemporalConvNeXtEEG, and TemporalConvNeXtAttEEG, are employed to model dynamic temporal dependencies in EEG signals. These models are designed to extract hierarchical and context-aware representations from multichannel EEG recordings. Subsequently, the learned representations are passed to a fully connected classification head for final decision-making. The models are optimized using RMSprop to ensure stable convergence and improved learning efficiency. The proposed DTLF-ALSD method was examined on the standard EEGET-ALS dataset. Comparative results demonstrate better performance than existing methods across diverse measures for the detection of ALS.