The scenario
ScanSure trains a transfer-learning model on public MRI datasets, adds explainability with Grad-CAM so every prediction comes with a heatmap, tracks experiments, and deploys the model as an API on GCP. A research and learning project, not a diagnostic device.
System architecture
01 MRI scans Public dataset
02 DVC pipeline Versioned data
03 CNN + Grad-CAM Prediction + heatmap
04 Cloud Run API Upload and view
High-level flow of ScanSure, from input to output.
What you will learn
- Transfer learning for medical images
- Explainable AI with Grad-CAM
- Experiment tracking and data versioning
What you will build
- Training pipeline
- Explainability layer
- Cloud API
Build plan
- Data pipeline
- Train
- Explainability
- Tracking
- Deploy