Self-Supervised Learning Method for Breast Cancer Detection with Image Feature Set and Modified U-Net Segmentation using Whole Slide Image

Among women, breast cancer has been one of the most frequent causes of death across the globe and early diagnosis is very crucial in enhancing the survival outcomes of such patients. The conventional methods of histopathological image analysis rely considerably on huge quantities of marked-up information which can be costly and time-consuming to generate. In order to overcome these shortcomings, we present a self-supervised learning framework, based on whole slide images (WSIs) to detect breast cancer. It consists of a contrastive learning based feature extractor, that is pre-trained on unlabeled WSIs, and a modified U-Net architecture with additional attention mechanisms and residual connections to segment the tumor precisely. On the CAMELYON16 dataset, experimental findings indicate that the suggested approach attains a Dice coefficient of 0.873 and an Intersection over Union (IoU) of 0.812, which is 6.5 and 7.8 percentage points higher than the standard U-Net baselines, respectively. Also, the use of unlabeled data cut the training time by about 25 percent. Such results confirm the effectiveness and efficiency of such a solution as the combination of self-supervised feature learning and state-of-the-art segmentation methods in automated breast cancer detection in digital pathology.