Twins2-Sat: A Taylor Bird Swarm Optimized Twin-SpatiallyA separable Self-Attention Transformer Model for Highly Accurate Breast Cancer Classification

In light of the growing need for sophisticated diagnostic systems in medical imaging, this work presents a novel framework for the ‎diagnosis of breast cancer that focuses on accurate tissue characterization using texture and attention-based learning. Mammography images ‎of the breast are first preprocessed using an Adaptive Savitzky-Golay Filter (ASGF) to reduce noise and improve important structures. The ‎next step is the precise segmentation of problematic areas using the Markov Random Field (MRF) model, which efficiently defines tumor ‎borders by utilizing spatial dependencies. Following the isolation of the region of interest, texture features are extracted using the Gray-Level ‎Co-occurrence Matrix (GLCM), which uses statistical metrics like energy, contrast, and correlation to capture crucial spatial correlations in ‎pixel intensities. After that, these descriptive characteristics are sent to a TwinS²SAT model, which is an improved Twins-Spatially ‎Separable Self-Attention (SSSA) Transformer that has been tuned using the Taylor Bird Swarm Algorithm (TBSA) to enhance ‎classification performance and learning efficiency. The combination of a lightweight, attention-driven deep architecture with GLCM features ‎allows for a reliable differentiation between benign and malignant tissue patterns. Python software is used to evaluate the proposed model, and ‎the comparison is made with the Self-Attention Transformer (SAT) model. The results show that compared to SAT with 94.99% of accuracy, ‎the TBSA- TwinS²SAT model ranks with excellent accuracy of 97.94%, precision of 98%, and 99% recall, confirming its potential as a ‎useful tool for automated breast cancer diagnosis‎.