3D Masked Autoencoding and Pseudo-labeling for Domain Adaptive Segmentation of Heterogeneous Infant Brain MRI

Xuzhe ZhangYuhao WuJia GuoJerod M. RasmussenThomas G. O'ConnorHyagriv N. SimhanSonja EntringerPathik D. WadhwaClaudia BussCristiane S. DuarteAndrea JackowskiHai LiJonathan PosnerAndrew F. LaineYun Wang

Xuzhe ZhangYuhao WuJia Guo ...+11 Yun Wang

Mar 2023

摘要

Robust segmentation of infant brain MRI across multiple ages, modalities, andsites remains challenging due to the intrinsic heterogeneity caused bydifferent MRI scanners, vendors, or acquisition sequences, as well as varyingstages of neurodevelopment. To address this challenge, previous studies haveexplored domain adaptation (DA) algorithms from various perspectives, includingfeature alignment, entropy minimization, contrast synthesis (style transfer),and pseudo-labeling. This paper introduces a novel framework called MAPSeg(Masked Autoencoding and Pseudo-labelling Segmentation) to address thechallenges of cross-age, cross-modality, and cross-site segmentation ofsubcortical regions in infant brain MRI. Utilizing 3D masked autoencoding aswell as masked pseudo-labeling, the model is able to jointly learn from labeledsource domain data and unlabeled target domain data. We evaluated our frameworkon expert-annotated datasets acquired from different ages and sites. MAPSegconsistently outperformed other methods, including previous state-of-the-artsupervised baselines, domain generalization, and domain adaptation frameworksin segmenting subcortical regions regardless of age, modality, or acquisitionsite. The code and pretrained encoder will be publicly available athttps://github.com/XuzheZ/MAPSeg

展开全部

图表提取