Tencent Health AI Medical Report Interpreter

Problem

Clinical reports often produce many arrows and units, making it hard to triage severity, specialty, and follow-up timing. Item-by-item translation can miss critical values or present non-diagnostic findings as diagnoses.

How It Works

• Type routing: distinguishes lab, health-check, imaging, pathology, and genetic tests; pathology findings are treated as the diagnosis of record.
• Structured normalization: extracts item, value, unit, reference range, and method, with built-in ranges labeled when missing.
• Critical-value gate: when a critical value is hit, the first sentence gives a care-seeking instruction.
• Problem-level grouping: groups abnormal markers into issue names and avoids naming diseases below diagnostic thresholds.
• Pattern and trap checks: evaluates combinations, causal chains, sampling, physiologic, drug, and individual differences.
• Delivery: outputs fixed Markdown by default; exported HTML uses inline SVG trend charts and passes compliance checks.

Boundaries

It supports report text and structured data, not imaging interpretation, tumor benign/malignant decisions, medication plans, or psychological scale interpretation. Unclear fields must be flagged for verification instead of guessed.