Label-efficient transformer-based framework with self-supervised strategies for heterogeneous lung tumor segmentation

Zhenbing Liu, Weixing Li, Yanfen Cui, Xin Chen, Xipeng Pan, Guanchao Ye, Guangyao Wu, Yongde Liao, Leroy Volmer, Leonard Wee, Andre Dekker, Chu Han, Zaiyi Liu, Zhenwei Shi

Abstract: Highlights•Large datasets are used to build pre-training models by self-supervised learning.•Different self-supervised learning strategies are investigated.•A novel Surrounding Samples-based Contrastive Learning module is introduced.•MIM-based pre-training with SSCL shows superior performance on test data.