The Problem
3D CT volumes are rich in clinical signal but expensive to label and difficult to segment consistently. Teams need a repeatable pipeline that can train, optimize, and benchmark segmentation models without heavy manual setup.
What I Built
An end-to-end notebook workflow for CT segmentation using MONAI and PyTorch Lightning, covering data preparation, UNet training, quantization, and live inference. The project packages practical steps for evaluation, performance benchmarking, and deployment-ready experiments.
Key Results
- Reproducible notebook series for training and optimization
- Quantization path for faster inference on CT scan workloads
- Clear, documented workflow for clinical prototyping