Brain Tumor MRI Classifier - Neuro-Symbolic AI on MCP

9-class classification, FHIR R4, SHARP-compliant, 6 tools.

Comment

##Inspiration##

There are over 150 types of brain tumors. Doctors must identify which type from a scan — each needs different treatment and urgency. Every other submission handles text workflows. Nobody tackled imaging. So we did.

##What it does##

Analyzes brain MRI scans using neuro-symbolic AI:

  • FFT preprocessing for frequency-domain analysis
  • MobileNetV2 neural network (9-class classification) Lightweight enough for rural clinics with basic hardware
  • Temperature scaling for calibrated confidence
  • Symbolic reasoning engine (3 clinical rules)
  • FHIR R4 DiagnosticReport output — plug into any EHR
  • 6 MCP tools for complete clinical workflow
  • SHARP Extension Specs for multi-agent integration
  • Live demo: vendhal.github.io/Brain-Tumor-Classifier

## How we built it##

PyTorch + MobileNetV2 for lightweight classification. Custom FFT preprocessing pipeline. CGAN for training data augmentation. Temperature scaling for honest confidence scores. FastMCP for MCP protocol integration. FHIR R4 DiagnosticReport standard output. SHARP Extension Specs for patient context propagation. FastAPI + Render for deployment. GitHub Pages for public interface.

## Challenges we ran into##

Getting FastMCP working with Prompt Opinion's Streamable HTTP transport required significant debugging. Ensuring all PyTorch numpy types were JSON-serializable. Designing symbolic reasoning rules with real clinical value beyond what the neural network provides alone.

## Accomplishments that we're proud of##

Only MRI scan analysis AI in Prompt Opinion marketplace. Complete pipeline from raw MRI to FHIR DiagnosticReport. 6 clinical tools covering full diagnostic workflow. Edge-optimized with MobileNetV2 — runs on basic hardware, accessible to rural clinics worldwide. SHARP-compliant, publicly accessible, zero setup required.

##What we learned##

MCP protocol, FHIR R4 standards, SHARP Extension Specs, neuro-symbolic AI architecture, clinical AI safety requirements, production deployment on Render.

##What's next for Brain Tumor MRI Classifier - Neuro-Symbolic AI on MCP##

DICOM file support for real radiology workflows. Radiologist feedback loop for continuous improvement. Direct Epic/Cerner EHR integration. Multi-modal CT + MRI fusion for better accuracy.

Built With

Share this project:

Updates