Feasibility study on a noninvasive assessment of ALS patient emotional state.

Introduction: Objective assessment of emotional responsiveness in Amyotrophic Lateral Sclerosis (ALS) remains limited despite its importance for personalized care. This study investigated whether non-invasive speech analysis could identify digital biomarkers associated with emotional coping behaviors in ALS patients. Methods: We analyzed 28 ALS patient visits using clinician-rated emotional concern scores, ALS Functional Rating Scale-Revised (ALS-FRS-R), forced vital capacity (FVC), and speech acoustic features. We also evaluated the reliability of large language models (LLMs), including ChatGPT, Gemini, and Claude, for automated concern assessment. Patients were classified relative to functional impairment as congruent, muted, or excessive responders. Results: LLMs failed to provide reliable or reproducible assessments of patient concern without expert clinical supervision. Subjective concern levels also showed discordance with objective respiratory measures such as FVC. Excessive responders were predominantly male and required significantly greater clinician interaction time, while muted responders were predominantly female. Acoustic analysis revealed distinct vocal profiles between groups. Muted responders demonstrated high loudness and sharpness with low roughness and fluctuation, whereas excessive responders showed the opposite profile. These findings suggest that dysarthria may function as an acoustic filter modulating emotional expression. Discussion: Speech-derived acoustic biomarkers may enable objective identification of emotional coping phenotypes in ALS and support earlier, personalized psychosocial interventions. The findings also highlight important limitations of current LLM-based clinical interpretation tools in unsupervised settings.