Role of artificial intelligence in the diagnosis of amyotrophic lateral sclerosis via biomarkers, neuroimaging, and wearable health technologies: a narrative review

Abstract

Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disorder characterised by upper and lower motor neuron loss. Diagnosis is frequently delayed because of phenotypic heterogeneity, overlap with ALS mimics, and the absence of a single definitive biomarker. Artificial intelligence (AI) offers complementary analytical tools across neuroimaging, biomarker panels, clinical records, and wearable sensor streams. To synthesise the evidence on AI applications for ALS diagnosis, prognosis, and longitudinal monitoring, and to critically appraise the methodological quality, generalisability, and translational readiness of this literature. We conducted a narrative review based on a PRISMA 2020-guided search of PubMed, Scopus, Web of Science, IEEE Xplore, and Google Scholar for studies published between January 2012 and March 2026. Eligible studies applied any AI method (machine learning, deep learning, natural language processing, multimodal fusion) to ALS diagnosis, progression prediction, or biomarker discovery, and reported at least one quantitative performance metric. No minimum accuracy threshold was imposed. After de-duplication, title/abstract screening, and full-text review, 103 studies informed the narrative synthesis. Risk of bias in studies contributing quantitative performance estimates was appraised with an adapted QUADAS-2 framework covering patient selection, index test, reference standard, and flow/timing. Across modalities, AI models reported a wide performance range — from modest discrimination in small single-centre imaging studies to reproducible accuracy in a prospectively validated blood-based miRNA classifier (area under the curve [AUC] 98.3% in discovery, > 98% across external validation cohorts). Multimodal fusion consistently outperformed unimodal approaches, but most studies relied on small convenience samples, lacked prospective external validation, and did not report calibration, subgroup performance, or fairness metrics. Risk of bias was judged high or unclear in patient selection and flow/timing domains for the majority of included studies. AI-enabled approaches show promise for earlier and more objective ALS detection, but the current literature is dominated by small, retrospective, single-site studies. Translation into clinical practice will require prospective multicentre external validation, transparent reporting under TRIPOD + AI and CONSORT-AI, regulatory-grade explainability, and equitable deployment across diverse populations.