SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining

Despite advances in data augmentation and transfer learning, convolutional neural networks (CNNs) difficultly generalise to unseen domains. When segmenting brain scans, CNNs are highly sensitive to changes in resolution and contrast: even within the same MRI modality, performance can decrease across datasets. Here we introduce SynthSeg, the first segmentation CNN robust against changes in contrast and resolution. SynthSeg is trained with synthetic data sampled from a generative model conditioned on segmentations. Crucially, we adopt a domain randomisation strategy where we fully randomise the contrast and resolution of the synthetic training data. Consequently, SynthSeg can segment real scans from a wide range…

• NI
  National Institute for Health and Care Research
• NI
  National Institutes of Health

  Awards: P41EB015896, R21EB018907, RF1MH123195, 1R01EB023281, R01AG016495, 1R01AG070988, 1R01AG064027, R01EB019956, U01MH117023, R01EB006758
• EA
  Engineering and Physical Sciences Research Council

  Award: EP/ L016478/1