SynthSegNet

This repository provides code for training a joint segmentation and synthesis network for MS lesion segmentation and synthetic DIR (Double Inversion Recovery). Our framework is rather generic and can be adapted to different diseases, body parts and modalities.

Data format

For our example code to work, image data are supposed to be

• Co-registered per scan into MNI152 space

• Skull-stripped

• File names should contain "_flair", "_t1", "_dir" and "_seg", respectively:

  patient_01

  patient_01_flair.nii.gz

  patient_01_t1.nii.gz

  patient_01_dir.nii.gz

  patient_01_seg.nii.gz