Enhanced U-Net Framework for Nuclei Segmentation in Triple Negative Breast Cancer

Breast cancer histopathology dataset contains images of varying complexities including different types of lesions and tissue densities. Breast cancer hiistopathology images are known for its molecular and morphological heterogeneity. Existing segmentation methods struggles to adapt to heterogeneity of breast cancer data. One-size fits all techniques fails to generalize across diverse datasets which leads to suboptimal segmentation performance. In order to minimize the staining differences that impede correct interpretation of stained histopathology images, stain normalization (SN) is applied to image samples. Various SN methods and their importance in enhancing the consistency and dependability of image analysis is presented in this paper. The aim of SN is to reduce staining variances preserving biological information. SN produces a uniform appearance which improves the precision and reproducibility of image analysis processes used in histopathology such as segmentation, feature extraction and classification. A modified U-Net method for Triple Negative Breast Cancer (TNBC) nuclei segmentation is presented in this paper. This paper highlights the potential of stain normalization techniques, image enhancement technique and enhanced U-Net architecture to support predictions by providing rigorous and consistent interpretation of histopathology images. The findings are shown experimentally on publicly available TNBC dataset. The nuclei segmentation accuracy for proposed method is 91.54%(SN+U-Net) and 93.86% (SN+EPLT+U-Net).