copy's
NeuroScan AI is a deep learning diagnostic assistant trained to locate and classify cerebral tumors from clinical MRI DICOM imagery.
"By deploying lightweight PyTorch models to an edge-serving API, NeuroScan delivers classification markers in under 200ms."
Neural Architecture
We designed a customized DenseNet121 architecture, pre-trained on ImageNet and fine-tuned on clinical brain scans. The network isolates region anomalies using Grad-CAM heatmaps, outlining the suspected tumor coordinate grids directly on the web application's layout.
Technical Stack
Built with PyTorch, FastAPI, Docker, and React. Grad-CAM layers highlight target regions dynamically on the canvas frontend.
System Architecture
[IMAGE DICOM SUBMISSION CLIENT]
React App with Canvas overlay drawing for MRI boundaries
▼
[GPU INFERENCE WORKER]
FastAPI Router / PyTorch C++ wrapper / DenseNet121 + Grad-CAM layers
▼
[DIAGNOSTIC LOG CLUSTER]
PostgreSQL for anonymized diagnostic metrics & scan references