CancerLens AI — Cancer Detection AI

CancerLens AI is an oncological triage platform built for the IBM Bob Hackathon 2026. IBM Bob was our core development partner for the entire project. Every component was generated, debugged, and refined through Bob — from the Flask backend architecture to the AMD MI300X GPU integration and Google Cloud Run deployment. What Bob built for us: → Complete Flask backend with 3 REST endpoints → 3-agent AI pipeline from scratch → All oncological analyst prompts → AMD MI300X vLLM endpoint integration → Google Cloud Vertex AI authentication → Frontend to backend API connection → Google Cloud Run deployment configuration → AMD toggle system for switching between GPU and fallback mode instantly The 3-Agent Pipeline: Agent 1 — Extractor: Qwen2-VL running on AMD MI300X with 192GB HBM3 analyzes the medical scan and extracts detailed visual observations at full precision with zero quantization. Agent 2 — Analyst: Gemini 2.0 Flash processes imaging findings alongside patient context and blood report values to generate a structured clinical report including cancer type, TNM staging, Early vs Late Stage classification, and risk scoring. Agent 3 — Validator: An independent AI pass cross-checks the entire report for logical consistency and assigns a reliability score before results reach the user. Additional features include stage-specific survival statistics pulled dynamically for any cancer type, a nearest oncologist finder via Google Maps, direct links to specialist hospitals, and a context-aware AI health assistant chatbot. Tested on real clinical teaching cases — correctly identified Stage III Osteosarcoma from a knee X-ray, cross-referenced elevated ALP and LDH from a blood report, and detected Glioblastoma from a brain MRI with 9-10 out of 10 reliability scores. CancerLens AI makes radiologist-level cancer detection accessible to anyone, anywhere — in seconds, not weeks. Built by Team Noxis.