Machine Learning and Brain-Computer Interface Approaches in Prognosis and Individualized Care Strategies for Individuals with Amyotrophic Lateral Sclerosis: A Systematic Review

Abstract

Amyotrophic lateral sclerosis (ALS) characterized by progressive degeneration of motor neurons is a debilitating disease, posing substantial challenges in both prognosis and daily life assistance. However, with the advancement of machine learning (ML) which is renowned for tackling many real-world settings, it can offer unprecedented opportunities in prognostic studies and facilitate individuals with ALS in motor-imagery tasks. ML models, such as random forests (RF), have emerged as the most common and effective algorithms for predicting disease progression and survival time in ALS. The findings revealed that RF models had an excellent predictive performance for ALS, with a testing R2 of 0.524 and minimal treatment effects of 0.0717 for patient survival time. Despite significant limitations in sample size, with a maximum of 18 participants, which may not adequately reflect the population diversity being studied, ML approaches have been effectively applied to ALS datasets, and numerous prognostic models have been tested using neuroimaging data, longitudinal datasets, and core clinical variables. In many literatures, the constraints of ML models are seldom explicitly enunciated. Therefore, the main objective of this research is to provide a review of the most significant studies on the usage of ML models for analyzing ALS. This review covers a variation of ML algorithms involved in applications in ALS prognosis besides, leveraging ML to improve the efficacy of brain-computer interfaces (BCIs) for ALS individuals in later stages with restricted voluntary muscular control. The key future advances in individualized care and ALS prognosis may include the advancement of more personalized care aids that enable real-time input and ongoing validation of ML in diverse healthcare contexts.