A IA consegue detetar doença de Parkinson em fase inicial a partir de tremores subtis na voz em chamadas telefónicas ?

A doença de Parkinson muitas vezes manifesta-se em alterações precoces e quase impercetíveis na voz — tremores subtis ou padrões irregulares na fala. Sistemas de IA treinados com gravações de voz poderiam, teoricamente, detetar estas micro-mudanças antes de os sintomas clínicos surgirem. Estas ferramentas poderiam ser implementadas através de aplicações de telemedicina ou centros de chamadas como uma primeira triagem. O desafio reside em distinguir os tremores relacionados com a doença do ruído de fundo, do stresse emocional ou dos sotaques.

Background

Parkinson’s disease often manifests in early, barely perceptible voice changes—subtle tremors or irregular patterns in speech. AI systems trained on voice recordings could theoretically pick up these micro-changes before clinical symptoms appear. Such tools might be deployed via telehealth apps or call centers as a first-pass screening tool. The challenge lies in distinguishing disease-related tremors from background noise, emotional stress, or accents.

Research teams have demonstrated that subtle voice tremors and other dysphonic features can be extracted from brief phone-call recordings and used to flag early-stage Parkinson’s disease with moderate accuracy, typically achieving area-under-the-curve values between 0.75 and 0.88 in proof-of-concept studies. Because these voice changes often precede clinically obvious motor symptoms, researchers are exploring lightweight smartphone apps that run near–real time analysis on encrypted voice snippets while preserving speaker privacy. Current systems remain investigational: they need larger, more diverse datasets and rigorous external validation before regulatory approval or public deployment.

— Enriched May 12, 2026 · Source: npj Digital Medicine