¿Puede la IA detectar depresión a partir de cambios sutiles en las microexpresiones faciales en videollamadas ?

El reconocimiento emocional a partir de videos ha avanzado rápidamente gracias a los modelos de aprendizaje profundo. Estos sistemas analizan movimientos faciales mínimos que los humanos suelen pasar por alto. Se correlacionan con escalas de depresión clínica y el seguimiento sostenido del estado de ánimo. La tecnología plantea preguntas éticas sobre el consentimiento y la vigilancia en las interacciones digitales.

Background

Emotional recognition from video has advanced rapidly due to deep learning models, which analyze minute facial movements often missed by humans. These systems correlate with clinical depression scales and sustained mood tracking, though they raise ethical questions about consent and surveillance in digital interactions. Current systems can reliably recognize basic facial action units and coarse emotions, but detecting depression from subtle, real-time micro-expressions in ordinary video calls remains unreliable in clinical settings. Research prototypes using 3D facial meshes, frame-level attention, and multimodal signals (voice, typing cadence) show modest correlations with PHQ-9 scores in controlled studies, yet generalization to diverse lighting, angles, and backgrounds is poor. Privacy, consent, and algorithmic fairness concerns further limit large-scale deployment, and no certified device is approved for diagnosis via video alone. (Enriched May 12, 2026; Source: National Academies of Sciences, Engineering, and Medicine)