MRI Brain Tumor Detection & Segmentation

Overview

Built a production-ready multimodal MRI classification and segmentation pipeline trained on the BraTS dataset. The model uses a shared encoder architecture for both tumor classification and pixel-level segmentation, achieving strong performance across both tasks simultaneously.

Results

Metric	Score
Classification Accuracy	91.3%
Sensitivity	97.1%
Dice Score (Segmentation)	76.5%
Model Parameters	31.7M
Inference Speedup	~40% faster vs separate models

Key Features

• Shared Encoder Architecture: Single backbone for both classification and segmentation reduces parameter count and inference time
• Conditional Segmentation: Segmentation head activates only when tumor is detected, optimizing inference
• One-Command Pipeline: Productized training-to-demo pipeline with automation across 6 stages, 4 smart prompts, and 25/25 passing tests
• API Endpoints: 11 FastAPI endpoints for model inference, health checks, and data management

Tech Stack

• ML: PyTorch, torchvision, BraTS dataset
• API: FastAPI, Uvicorn, Pydantic
• Testing: pytest with 25/25 test coverage
• Optimization: Conditional inference, shared encoder, mixed precision