LungIQ

Inspiration

Across U.S. cancers, the median time from diagnosis to treatment initiation is nearly 4 weeks, with much of that time spent on treatment planning, evidence review, and clinical trial matching. Yet in that narrow window, clinicians still need to synthesize prior patient outcomes, current literature, and relevant clinical trials to make evidence-grounded decisions.

The obvious 2026 instinct is to ask ChatGPT for a treatment plan, but oncology decisions need to be grounded in real evidence, real cases, and real trial availability.

What it does

LungIQ turns a CT scan and patient records into an evidence-grounded starting point for lung cancer treatment planning. It matches similar historical cases, summarizes likely treatment patterns, and surfaces relevant clinical trials and literature instead of generating a plan from scratch.

How we built it

The architecture was intentionally designed as a single inspectable pipeline rather than a black-box chatbot. Every major output in LungIQ, from similar case retrieval to treatment guidance to trial matching, is tied back to a specific stage in the system.

Challenges we ran into

The hardest part was building something trustworthy under hackathon constraints, especially with a small synthetic case base of about 150 patients rather than a real hospital-scale dataset. We also had to balance speed, reliability, and explainability across imaging, retrieval, and evidence-grounded generation.

Accomplishments that we're proud of

We built a working prototype that goes beyond “ask an LLM and hope” by grounding outputs in tumor data, similar cases, and supporting evidence. We are especially proud that the system connects imaging, retrieval, treatment guidance, and trial/literature search into one coherent workflow.

What we learned

We learned that in clinical AI, retrieval, provenance, and traceability matter just as much as generation quality. A useful treatment-planning system is not just about producing an answer — it is about showing why that answer is defensible.

What's next for LungIQ

Next, we want to integrate LungIQ with real FHIR/EHR hospital systems so it can learn from much larger and more representative patient data than the ~150-case demo database. That would improve the quality of our vector search, strengthen the statistical grounding of recommendations, and make the platform much more clinically useful.

Built With

• chromadb
• fastapi
• pydicom
• python
• scikit-image