Experts' cognition-driven safe noisy labels learning for precise segmentation of residual tumor in breast cancer

Residual tumor in breast cancer (RTBC) indicates the tumor that still remains in breast cancer tissue after neoadjuvant chemotherapy, which is an important iatrotechnique in the breast cancer treatment process (Asaoka et al., 2020; Charfare et al., 2005; Mieog et al., 2007; Schott & Hayes, 2012). Commonly, RTBC is associated with invasive ductal carcinoma in which tumor has spread into surrounding breast tissue. Quantitative evaluation of RTBC can provide clues important to prognosis and subsequent therapy of breast cancer (Pu et al., 2020; Yau et al., 2022). The key point of quantitative evaluation of RTBC is to achieve precise segmentation of RTBC (PSRTBC), which is a fundamental key technique in the treatment process of breast cancer, such as be leveraged to calculate the tumor-stroma ratio that has been proven to be a prognostic factor in breast cancer (de Kruijf et al., 2011). Whole sliding imaging (WSI) (Hanna et al., 2020), which was previously referred to as virtual microscopy, involves scanning a pathology glass slide into digital image at high resolution and displaying the digitalized image on a computer screen (Gilbertson & Yagi, 2005; Pantanowitz et al., 2018).