02 — Case Study
MedLens
Computer-vision triage for medical imaging, built as a graduation project at AIU.
- Year
- 2025
- Role
- ML Engineer & Full Stack Developer
- Status
- Shipped
01
The Problem
Radiology departments in under-resourced hospitals face long queues. Scans that need urgent review sit behind routine ones because there is no automated pre-screening step.
02
The Solution
A web platform where clinicians upload scans and a fine-tuned CNN flags likely-critical cases for priority review. The model serves through a FastAPI microservice; the clinician-facing app is a React/Node monorepo with a DICOM-aware viewer.
03
The Result
Reached 94% recall on the critical class in validation. Selected as a distinguished graduation project at Alamein International University.
Engineering highlights
- Transfer learning on EfficientNet with class-imbalance-aware training
- FastAPI inference microservice with request batching
- In-browser scan viewer with window/level controls
- Explainability overlays (Grad-CAM) so clinicians see *why* a scan was flagged
Stack
- React
- Node.js
- Express
- MongoDB
- Python
- PyTorch
- FastAPI
- Docker
