SCIseg: Automatic Segmentation of T2-weighted Intramedullary Lesions in Spinal Cord Injury

This study introduces SCIseg, a deep learning tool designed for automatic segmentation of T2-weighted intramedullary lesions in spinal cord injury (SCI). The goal is to automate lesion identification and measurement on MRI scans. The SCIseg model was trained using MRI data from 191 SCI patients across three different sites, accounting for variations in MRI scanners, image resolutions, and lesion types. The training included a three-phase process, using active learning to improve the model's performance. The model's performance was compared with manual lesion segmentations and other open-source methods. Results showed that SCIseg accurately segments spinal cord lesions, providing reliable measurements of lesion characteristics.

• 1
  SCIseg, an open-source automatic method, was trained and evaluated on a dataset of 191 spinal cord injury patients from three sites for segmenting spinal cord and T2-weighted lesions.
• 2
  SCIseg generalizes across traumatic and non-traumatic lesions, scanner manufacturers, and heterogeneous image resolutions, enabling automatic extraction of lesion morphometrics in large multi-site cohorts.
• 3
  Quantitative MRI biomarkers, such as lesion length and maximal axial damage ratio, derived from SCIseg's automatic predictions, showed no statistically significant difference compared to manual ground truth.