MRI-SEG Lite - Lightweight Spine MRI Segmentation

I developed MRI SEG Lite, a lightweight version of the original MRI-SEG software package, specifically tailored for resource-constrained environments such as the Raspberry Pi. This version is designed to perform automatic classification and segmentation of Spine MRI images directly on edge devices without reliance on high-end GPUs or cloud infrastructure.

To make this possible, the original deep learning model was converted and optimized using techniques such as model pruning, quantization, and format conversion (e.g., TensorFlow Lite), significantly reducing memory and compute requirements while preserving model accuracy to a similar level of the original one.

MRI SEG Lite supports the same core functionalities as the desktop version, including real-time visualization, but is optimized for environments with limited hardware resources and power constraints. This makes it suitable for use in remote clinics, field hospitals, or mobile diagnostic units.

Ongoing Work: The project is currently in its research and development phase, with ongoing efforts to improve speed, accuracy, and power efficiency for clinical deployment.

Note: More information about this project will be added soon.