Deeply supervised network for white matter hyperintensities segmentation with transfer learning

Yilei Wu, Fang Ji, Yao Feng Chong, Li-Hsian Christopher Chen, Juan Helen Zhou

Keywords: white matter hyperintensities, transfer learning, deep supervision

TL;DR: We proposed a deeply supervised 3D network with transfer learning from UKBiobank dataset for white matter hyperintensities (WMH) segmentation.

Abstract: White matter hyperintensities (WMH) are brain white matter lesions commonly found in the elderly. Due to its association with cerebrovascular and neurodegenerative diseases, quantifying WMH volume is critical for many neurological applications. Previous segmentation approaches using 2D U-Net potentially omit the learning of 3D spatial contextual information. This paper proposes a deeply supervised 3D U-Net-like network with transfer learning to perform WMH segmentation in fluid attenuation inversion recovery (FLAIR) magnetic resonance images (MRI). We leveraged a pretrained network constructed by predicting brain age from structural MRIs. The proposed method achieved a Dice score of 82.3 on the MICCAI WMH Challenge training dataset and 75.3 on another independent testing dataset, outperforming other state-of-the-art methods.

Registration: I acknowledge that acceptance of this work at MIDL requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: novel methodological ideas without extensive validation

Primary Subject Area: Segmentation

Secondary Subject Area: Transfer Learning and Domain Adaptation

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.