OGA-AID: Clinician-in-the-loop AI Report Drafting Assistant for Multimodal Observational Gait Analysis in Post-Stroke Rehabilitation
A new multi-agent AI system synthesizes video, motion data, and clinical notes into structured assessments.
A research team from Nanyang Technological University and Tan Tock Seng Hospital has introduced OGA-AID, a novel AI assistant designed to tackle the time-intensive and cognitively demanding task of clinical gait analysis for post-stroke patients. The system is built as a multi-agent large language model (LLM) framework, coordinating three specialized AI agents. Each agent is responsible for synthesizing a different data modality: patient movement videos, detailed kinematic motion-capture trajectories, and existing clinical profiles. By integrating these disparate data streams, OGA-AID generates a comprehensive, structured draft assessment report, a process that traditionally requires significant manual effort from physiotherapists.
In rigorous evaluations with expert clinicians using real patient data, OGA-AID demonstrated superior performance compared to simpler, single-pass multimodal AI baselines, achieving its results with notably low error rates. Crucially, the system is designed for a 'clinician-in-the-loop' workflow. When experts provided brief preliminary notes, the AI's output error was further reduced compared to gold-standard reference assessments. This highlights the system's core philosophy: AI and human expertise are complementary. The findings, presented for the 2026 CV4Clinic CVPR Workshop, validate the feasibility of using such agentic AI systems for structured clinical tasks and point toward a future where AI handles data synthesis and drafting, freeing clinicians to focus on higher-level judgment and patient interaction.
- Uses a multi-agent LLM framework with three specialized agents to process video, motion data, and clinical notes.
- Evaluated with expert physiotherapists and outperformed standard multimodal AI baselines on real patient data.
- Designed for 'clinician-in-the-loop' use, where brief expert input further improves the AI's report accuracy.
Why It Matters
It could drastically reduce administrative burden in rehab, allowing clinicians to spend more time on direct patient care and complex decision-making.