| Sumario: | Manual evaluation of mitotic activity & nuclear atypia is laborious and prone to inter-observer variability; yet, accurate classification of breast cancer using histopathological pictures is essential for diagnosis and therapy planning. Automatic cancer of the breast grading using the MITOS-ATYPIA-14 dataset is presented in this research utilising an ambiguously guided dual-branch hybrid architecture. Hybrid nuclei segmentation with U-Net and watersheds algorithms follows stain normalisation and contrast enhancement in the suggested approach. Mitotic or nuclear atypia features are extracted separately using a dual-branch feature learning technique that combines profound learning, structural morphology, and texture-based descriptors. The most discriminative characteristics are selected using an attention-based dynamic feature selection technique and then merged to create a complete representation. An ensemble model that incorporates Support Vector Machine, Random Forest, & Light GBM is used for classification in order to increase accuracy and resilience. By integrating mitotic & atypia scores using a modified grading method, the ultimate cancer grade is determined. The experimental findings show that the suggested framework outperforms the current methods, with a 97.82% accuracy and a 96.77% F1-score. Based on these outcomes, it seems like the suggested approach is a solid bet for automatic breast cancer grading.
|
|---|