| """Gradio app - Diabetic Retinopathy Clinical Report Generator. |
| |
| Pipeline: upload fundus image -> trained DR model grades severity (0-4) -> |
| Gemini drafts a structured clinical report around that grade. |
| |
| Run locally: python app.py |
| This file is also the HuggingFace Spaces entry point. |
| """ |
| from __future__ import annotations |
|
|
| import gradio as gr |
|
|
| from src.dr_grader import grade_fundus |
| from src.report_generator import generate_report |
| from src.config import CLASS_NAMES |
|
|
| DISCLAIMER = ( |
| "⚠️ **Research/education demo only - not for clinical use.** " |
| "The grade is produced by an automated model and the report is AI-generated; " |
| "both must be reviewed by a qualified ophthalmologist." |
| ) |
|
|
|
|
| def _per_class_from_ordinal(ordinal): |
| """Convert CORN ordinal probabilities P(>0..>3) into a per-class distribution.""" |
| ge = [1.0] + list(ordinal) |
| pc = [] |
| for k in range(5): |
| nxt = ge[k + 1] if k + 1 < len(ge) else 0.0 |
| pc.append(max(0.0, ge[k] - nxt)) |
| total = sum(pc) or 1.0 |
| return {CLASS_NAMES[k]: pc[k] / total for k in range(5)} |
|
|
|
|
| def analyze(image): |
| if image is None: |
| return "Please upload a fundus image.", None, {}, "" |
|
|
| result = grade_fundus(image) |
| grade, label = result["grade"], result["label"] |
| grade_md = f"## Grade {grade}: {label}" |
| confidences = _per_class_from_ordinal(result["ordinal_probabilities"]) |
|
|
| try: |
| report = generate_report(grade, result["processed_image"]) |
| except Exception as exc: |
| report = f"_Report generation unavailable: {exc}_" |
|
|
| return grade_md, result["processed_image"], confidences, report |
|
|
|
|
| with gr.Blocks(title="DR Clinical Report Generator") as demo: |
| gr.Markdown("# 🩺 Diabetic Retinopathy - Clinical Report Generator") |
| gr.Markdown( |
| "Upload a retinal fundus image. A trained EfficientNet-B4 + Swin Transformer " |
| "model grades diabetic-retinopathy severity (0-4), and Gemini drafts a " |
| "structured clinical report around that grade." |
| ) |
| gr.Markdown(DISCLAIMER) |
|
|
| with gr.Row(): |
| with gr.Column(): |
| inp = gr.Image(type="pil", label="Upload fundus image") |
| btn = gr.Button("Analyze", variant="primary") |
| with gr.Column(): |
| grade_out = gr.Markdown() |
| conf_out = gr.Label(num_top_classes=5, label="Severity probabilities") |
| proc_out = gr.Image(label="Preprocessed (RFOV Crop + CLAHE)", type="pil") |
|
|
| gr.Markdown("### Clinical report") |
| report_out = gr.Markdown() |
|
|
| btn.click(analyze, inputs=inp, outputs=[grade_out, proc_out, conf_out, report_out]) |
|
|
|
|
| if __name__ == "__main__": |
| demo.launch() |
|
|
