Gray matter segmentation algorithm: sct_deepseg_gm

Results of the GM Challenge

For segmenting the gray matter, SCT features the function sct_deepseg_gm, which is based on a deep learning architecture trained from 232 subjects (~4000 slices).

  • Algorithm: Deep learning with dilated convolutions [Perone et al., Sci Report 2018]

  • Pros: High accuracy, robust to pathologies

  • Cons: Restricted to T2*-like contrasts (GM bright, WM dark)

sct_deepseg_gm obtained the best Dice score amongst all other methods that participated in the GM challenge [Prados et al., Neuroimage 2017].