Evidence-grounded PCI vs. CABG recommendations from a cath report.
Upload a coronary angiogram (ICA) or CCTA report and get a revascularization recommendation with citations to the 2021 ACC/AHA/SCAI Coronary Revascularization Guidelines, in seconds.
How it works
Every recommendation is traceable to a specific Class of Recommendation and Level of Evidence statement in the guideline text, not to the model's general knowledge.
Extract
Gemini reads the PDF and fills a 60+ field JSON schema: segment-level stenosis, dominance, LV function, comorbidities, clinical presentation.
Retrieve
Three queries (case summary, disease-specific, generic PCI-vs-CABG) hit a ChromaDB index of the guideline embedded with PubMedBERT, and the results are merged.
Recommend
The model reasons over the extracted findings and the retrieved sections, returning a strategy, confidence, factors favoring each approach, and verbatim citations.
Tech stack
| Extraction & reasoning LLM | Google Gemini 2.5 Flash |
|---|---|
| Embeddings | PubMedBERT NeuML/pubmedbert-base-embeddings |
| Guideline source | 2021 ACC/AHA/SCAI Coronary Revascularization Guidelines |
Clinical background
Choosing between PCI and CABG for multi-vessel coronary artery disease is one of the most consequential decisions in cardiology.
NSTEMI
A heart attack where one or more coronary arteries are partially blocked. Many patients have multi-vessel disease that needs a revascularization strategy.
PCI
Catheter-based balloon and stent procedure. Less invasive, faster recovery, favored for simpler anatomy.
CABG
Open-heart bypass surgery that reroutes flow around blockages. Generally preferred for complex multi-vessel or left main disease.
The guidelines rate specific anatomical and clinical scenarios by class of recommendation and level of evidence. Applying them correctly means integrating many variables at once, which is exactly the kind of task AI-assisted decision support is suited for.