Building TrialBridge: An AI clinical trial matcher built in a single hack day

说明与用途

TrialBridge 是在一次黑客松中构建的 AI 临床试验匹配工具。它从上传的医学报告中提取患者信息，与印度临床试验数据库进行匹配，并生成供肿瘤科医生使用的检查清单，同时提醒用户需医生确认资格。

问题

Cancer treatments push Indian families into debt while free, cutting‑edge clinical trials remain unfilled because eligibility rules are dense medical jargon that patients and busy oncologists cannot easily interpret.

结果

TrialBridge was built in a single hack day, enabling patients and oncologists to quickly assess trial eligibility for Indian clinical trials, reducing manual effort and improving access to free treatments.

工作流

• 1. User uploads photos of medical reports, scans, and prescriptions in the browser.
• 2. Images are shrunk client‑side to stay within Vercel request‑size limits.
• 3. Gemma‑4 (via Gemini API) performs multimodal extraction to build a structured patient profile and extracts exact quotes as evidence.
• 4. The profile is manually verified by the user before matching.
• 5. TrialBridge pulls active recruiting trials with Indian sites from ClinicalTrials.gov v2 API.
• 6. A custom text splitter breaks registry text into atomic inclusion/exclusion rules.
• 7. Gemma judges each rule against the patient profile, returning boolean results.
• 8. Plain, unit‑tested code aggregates rule outcomes into trial match categories (likely match, possible match, not eligible).
• 9. The app generates a printable summary with a ✓/✗/? checklist and specific questions for the oncologist.
• 10. Safety notice reminds users to confirm eligibility with a doctor.

证据

We built a custom text splitter (lib/criteria.ts) to break down the massive blocks of registry text into atomic inclusion and exclusion rules.

来源

Gemma performs multimodal extraction, reading the uploaded images to build a structured patient profile (cancer type, stage, biomarkers, ECOG, labs, current medicines) alongside the exact quotes it used as evidence.

来源

We fixed latency by switching Gemma to minimal thinking mode and writing medical hints directly into the prompt, dropping a 4‑trial batch down to 42 seconds with no thinking tokens used.

来源

The app never claims true medical eligibility—it only says "may qualify, confirm with your oncologist".

来源