Reconstructing digital phenotyping of psychological distress: integrating visual complexity and linguistic sentiment

Objective Identifying objective behavioral signatures of suicide risk remains a critical challenge. While digital phenotyping tracks emotional dynamics, validating multimodal frameworks requires exceptionally dense, chronological data. Methods We utilized the comprehensive historical legacy of Vincent van Gogh as a longitudinal paradigm, integrating fractal dimension (FD) analysis of 388 paintings with VADER-based sentiment analysis of his correspondence across his final five clinical phases (1888–1890). Results Results demonstrated a significant progressive escalation of FD during periods of severe psychological instability. Concurrently, negative language sentiment peaked during hospitalization but markedly declined in the final weeks preceding his death. Clinically, this distinct divergence—normalization of linguistic affect despite persistently high creative complexity— may serve as a quantifiable behavioral correlate of “terminal resolve” preceding suicide. Conclusions This integration of quantitative visual analysis and natural language processing establishes a reproducible framework for retrospective digital phenotyping and hypothesis-generation for future crisis monitoring models.