AI-Driven Early Detection of Amyotrophic Lateral Sclerosis (ALS): A Machine Learning Approach Using Acoustic Biomarkers

Abstract

Introduction: Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disease that affects the motor nervous system, specifically the motor neurons responsible for voluntary muscle movement. One of the main affected areas in the brain is the bulbar region which controls speech. The motor neuron degeneration in the bulbar region causes speech impairment dysarthria (slurred speech) which is also linked to shorter survival. Speech impairment is often one of the earliest manifestations of ALS; however, the articulatory movements involved are so subtle, rapid, variable, and minute that identifying abnormalities demands substantial clinical expertise. Consequently, accurate diagnosis is frequently delayed until the symptoms become more pronounced. While early detection is critical, it remains challenging due to the inherent limitations of conventional diagnostic tools. In this context the advancements in machine learning and AI coupled with high quality audio systems are playing a pivotal role in finding subtle patterns in patient speech data to help neurologists in early detection of ALS symptoms. Aim: The aim is to demonstrate the application of artificial intelligence and machine learning in developing a prototype tool for the early and accurate detection of ALS. Method: The proposed tool analyzes patients’ acoustic data, identifies characteristic speech patterns associated with dysarthria, and predicts the likelihood of current or potential ALS manifestation with high accuracy. In this paper I present acoustic analysis on the VOC-ALS dataset containing voice recordings of ALS patients and healthy individuals. From the collected vocal features, a machine learning model was developed to classify ALS patients and correlate them to specific speech disorder voice characteristics. The study then led to the building of a model that was able to classify healthy subjects from ALS patients. Results: Considering a small number of the patients in the data set, the model achieved good results with a classification accuracy of 81% and demonstrated strong sensitivity (recall = 0.67) and precision (0.93) for ALS detection. Conclusion: This study demonstrates the potential of machine learning in supporting early ALS diagnosis through speech analysis by identifying subtle articulatory changes that often go unnoticed in clinical settings.