Masked Video Modeling with Correlation-Aware Contrastive Learning for Breast Cancer Diagnosis in Ultrasound

Breast cancer is one of the leading causes deaths in women. As primary output breast screening, ultrasound (US) video contains exclusive dynamic information for diagnosis. However, training models analysis non-trivial as it requires a voluminous dataset which also expensive to annotate. Furthermore, diagnosis lesion faces unique challenges such inter-class similarity and intra-class variation. In this paper, we propose pioneering approach that directly utilizes US videos computer-aided It leverages masked modeling pretraning reduce reliance on size detailed annotations. Moreover, correlation-aware contrastive loss developed facilitate identifying internal external relationship between benign malignant lesions. Experimental results show our proposed achieved promising classification performance can outperform other state-of-the-art methods.