Description of Invention
Breast cancer remains the leading cause of death from female cancer, with over 2.3 million cases in 2022. Early diagnosis is crucial, but the manual biopsy examination is very time-consuming. This study explores a hybrid deep learning approach to ER-positive breast cancer nuclei segmentation on 220 ROI patches from 44 WSIs with ground truth. Specifically, a custom U-Net and Cellpose were integrated for segmentation accuracy and transferability. The method achieved a mean F1 Score of 0.7468 and IoU of 0.6018, and it shows its capability in the medical field.