Semi-supervised segmentation of hyperspectral pathological imagery based on shape priors and contrastive learning

Hongmin Gao, Huaiyuan Wang, Lanxin Chen, Xueying Cao, Min Zhu, Peipei Xu

Abstract: Highlights•Our study proposed a semi-supervised segmentation of hyperspectral pathological images based on shape priors and contrastive learning.•Building upon existing uncertainty estimation techniques, we incorporate the shape characteristics of cholangiocarcinoma to refine model predictions, thereby enhancing the quality of the model's predictions.•Morphological operations are utilized to perturb input data, mitigating the detrimental effects of existing data perturbation methods on model predictions.•We propose a novel contrastive learning approach that constructs image-level category feature vectors for contrastive learning.•Extensive experiments demonstrated that our method outperforms fully supervised approaches and state-of-the-art semi-supervised methods, providing valuable insights for the field of histopathological image analysis.