pipeline-ms-lesion

Pre-processing pipeline for multi-contrast spinal MS lesion segmentation.

Goals

The goals of this project are to:

• Provide a single multi-contrast model that generalizes well across different sites/vendors/sequence parameters;
• Establish an analysis framework that accommodates multiple sessions (over time) to estimate lesion growth/reduction;

Methods

The segmentation algorithm uses the deep learning framework SoftSeg, providing outputs that encode partial volume effect and provide uncertainty measures.

Data

The data come from the following sites:

• Brigham and Women's Hospital 🇺🇸
• Massachusetts General Hospital 🇺🇸
• New York University 🇺🇸
• Vanderbilt University 🇺🇸
• Zuckerberg San Francisco General Hospital 🇺🇸
• National Institute of Health 🇺🇸
• U. Mass (in data.neuro) 🇺🇸
• CanProCo (in process) 🇨🇦
• University College London 🇬🇧
• Inserm 🇫🇷
• Aix Marseille Université 🇫🇷
• OFSEP 🇫🇷
• Ospedale San Raffaele 🇮🇹
• Karolinska Insitutet 🇸🇪