NeuroScan: Explainable MRI Brain Lesion Classification

### Problem Statement Medical imaging research in academic settings frequently struggles with model interpretability. Black-box deep learning models predict classifications without exposing visual justification, limiting their utility in educational and clinical research settings. ### Proposed Solution NeuroScan provides an interactive web interface where researchers can upload DICOM/NIfTI brain scan slices, run inference across fine-tuned ResNet and EfficientNet backbones, and immediately observe pixel-level attention heatmaps through Grad-CAM. ### Technology Stack & Architecture - ML Core: PyTorch, TorchVision, MONAI (Medical Open Network for AI) - Backend: FastAPI, Celery with Redis for asynchronous inference queues - Client: React, Three.js slice renderer, Tailwind CSS - Data Handling: Anonymized BraTS benchmark dataset ### Team Needs Looking for an ML engineer interested in model quantizing (ONNX) and a frontend engineer experienced with Canvas/WebGL renderers.