Apppublic

Rf33d/DR-CKD

sourceHugging Faceupdated 5mo agoView on Hugging Face

0likes

App README

DR–CKD Early Detection Web App

Diabetic Retinopathy Based Early Detection of Chronic Kidney Disease Using Deep Learning

Project Structure

dr_ckd_app/
├── app.py               ← Flask backend (model, inference, GradCAM, PDF)
├── model.pth            ← YOUR Phase 2 checkpoint (place here)
├── requirements.txt     ← Python dependencies
└── templates/
    └── index.html       ← Full frontend (upload, results, chatbot)

Setup (one time)

bash

# 1. Install dependencies
pip install -r requirements.txt

# 2. Place your Phase 2 .pth file in this folder, named model.pth
#    (this is the BEST_CKPT from Step 7 of your notebook)
cp /path/to/your/phase2_best.pth model.pth

Run

bash

python app.py
# or: python app.py path/to/your_model.pth

Then open → http://localhost:5000

What the Website Does

Feature	Description
Upload	Drag & drop or click to upload a fundus image
Validation	Rejects non-fundus images with an error message
Preprocessing	Runs the full cv2 pipeline: crop border → Ben Graham → CLAHE → normalize
Inference	ConvNeXtDR Phase 2 model predicts DR grade 0–4 + confidence
GradCAM	Shows original + heatmap overlay side by side
CKD Risk	Lookup table maps DR grade → risk category, eGFR, CKD stage, clinical action
Probability bars	Confidence for all 5 DR classes
PDF Report	Downloadable report with all results + GradCAM images
Chatbot	Keyword-based Q&A about DR, CKD, GradCAM, model — no API needed

Notes

—The model runs on CPU by default (no GPU needed for inference)
—Model loads once at startup — predictions are fast
—The chatbot is pure JavaScript with ~17 Q&A topics, no external API
—PDF is generated server-side using ReportLab