Music-induced physiological markers for detecting Alzheimer's disease using machine learning.

Introduction: Alzheimer's disease (AD) is characterized by progressive cognitive and emotional decline, highlighting the need for novel, non-invasive biomarkers to aid in early detection, monitoring, and stage-specific interventions. This study investigates music-evoked physiological responses as potential biomarkers of AD and evaluates their translational value using machine learning (ML). Materials and methods: A total of 36 AD patients, spanning different severity levels, listened to emotionally evocative musical excerpts while electrodermal activity and facial electromyography (corrugator and zygomaticus muscles) were recorded. Machine learning models were then trained on these signals to classify the presence and severity of AD and to detect residual emotion-specific physiological responses elicited by music. Results: Physiological reactivity to music declined with disease progression, with positive emotions eliciting more distinct responses than negative ones. The Random Forest classifier distinguished AD patients from healthy controls with 70.5% accuracy, while the Naïve Bayes model predicted severity with 65.6% accuracy, demonstrating that ML models can detect subtle music-evoked physiological differences even in individuals with AD. Discussion: Music-evoked physiological signals reflect the hierarchical disruption of emotion-related neural circuits in AD and hold promise as complementary biomarkers for disease presence and stage. When combined with machine learning (ML), these measures provide a non-invasive, ecologically valid approach to support early detection, monitoring, and the development of stage-specific interventions.