Evidence-Grounded AI for Musculoskeletal Care
OrthoPilot is the first clinical LLM system purpose-built for musculoskeletal care, demonstrating end-to-end evidence-grounded decision-making—from admission diagnosis to rehabilitation planning—by autonomously integrating real-time multimodal hospital data streams (imaging, lab, pathology, orders) with authoritative external medical knowledge, validated on a specialist-annotated benchmark spanning 1,000 ICD disease codes derived from real-world EHRs.
Core Contribution
OrthoPilot is a clinical artificial intelligence system designed to address critical bottlenecks in longitudinal musculoskeletal management—including fragmented evidence across visits/departments/systems, disjointed knowledge sources, and delayed clinical reasoning. Built upon a large language model (LLM), it is the first system to dynamically co-model heterogeneous hospital data streams with structured external medical knowledge.
Technical Implementation
- Real-time hospital integration: autonomously retrieves and parses imaging reports (e.g., MRI/CT), laboratory results, pathology reports, and clinical orders;
- Longitudinal patient representation: converts multi-month/year clinical trajectories (recovery, remodelling, degeneration) into structured, inferable temporal evidence states;
- External knowledge fusion: integrates evidence-based clinical guidelines, anatomical-functional ontologies, and stage-specific functional goals to ground clinical decisions;
- Full-care-pathway coverage: supports end-to-end reasoning—from admission diagnosis and intervention selection (surgical/non-surgical), to complication risk forecasting and personalized rehabilitation planning.
Validation Methodology & Metrics
- Specialist-validated benchmark: constructed from real-world electronic health records (EHRs) covering 1,000 ICD-coded musculoskeletal conditions;
- Comprehensive reader study: blinded evaluation by orthopaedic specialists assessing OrthoPilot’s diagnostic suggestions, treatment pathways, and functional goal proposals across the entire care continuum;
- Preprint published on arXiv:2607.12527v1 (Announce Type: new).