AI-powered diabetic foot ulcer detection and classification for faster, more accurate wound assessment.
First place, HMI Hackathon 2026 (Kennesaw State University).
The backend runs on Render's free tier and spins down when idle. The first classification after a quiet period can take up to a minute while the service wakes; after that, results come back in a second or two.
Every 20 seconds, someone loses a limb to diabetes worldwide, and up to 80% of these amputations are preventable with early detection. Diabetic foot ulcers (DFUs) are often misclassified by clinicians 30-40% of the time, and many rural or underserved clinics lack access to wound care specialists altogether.
StepSafe is a web-based clinical decision support tool that lets clinicians capture a photo of a diabetic foot ulcer and receive instant AI-driven wound classification. By turning any camera-equipped device into a virtual wound specialist, StepSafe helps standardize assessments and support earlier, more accurate interventions.
- Capture or upload wound images directly from a phone or desktop browser.
- AI classification of infection, ischemia (poor blood flow), gangrene (tissue death), and normal tissue.
- Confidence scores for each prediction to support clinician judgment, not replace it.
- Severity assessment with suggested next-step clinical actions.
- Exportable clinical report (copy to clipboard or print to PDF) for EHR documentation.
- Mobile-responsive UI that runs in any modern browser, with no app install required.
- A nurse or clinician captures or uploads a DFU image in the React frontend.
- The image is sent via HTTP POST to a FastAPI backend (
/api/v1/classify). - The backend preprocesses the image and passes it into a TensorFlow classification model trained on DFUC 2020/2021 and Kaggle DFU datasets (~20,000 expert-labeled images).
- The model outputs class probabilities, which the frontend displays with a severity indicator and recommended action.
| Piece | Stack | Hosted on |
|---|---|---|
Frontend (StepSafe/frontend) |
React 19, Vite, Tailwind CSS 4 | Vercel (static, /api/* proxied to the backend) |
Backend (backend) |
Python, FastAPI, TensorFlow (MobileNetV2 transfer learning) | Render |
The current production model is a binary classifier (DFU vs normal); ischemia and gangrene probabilities are reported as 0 until the full 4-class model is trained.
# Backend (Python 3.11)
cd backend
pip install -r requirements.txt
# place model weights at backend/ml_model/stepsafe_model.h5
uvicorn app.main:app --port 8000
# Frontend
cd StepSafe/frontend
npm install
npm run dev # dev server proxies /api to localhost:8000Or with Docker: docker-compose up from the repo root.
- Stephen Sookra: Frontend & pitch.
- Tylin: Backend/API & model integration.
- Ryann: Data & model training.
- Benjamin Mettler: Research, literature review, and presentation support (PowerPoint).
- Emma Pittman: Research, article sourcing, and presentation design (PowerPoint).
For educational and demonstration purposes only. Not medical advice. Results should be validated by a qualified healthcare professional; StepSafe is intended to support, not replace, clinical judgment.