Efficient Breast Cancer Detection and Classification Using Rotation-Invariant Progressive Feedback Cosine Convolutional Neural Network with Tyrannosaurus Optimization Algorithm in Ultrasound Image

Breast cancer is the second most common cause of cancer-related deaths among women globally. Improved survival depends on a prompt and precise diagnosis, yet traditional ultrasound interpretation is time-consuming and error-prone. This study presents a new Rotation-Invariant Progressive Feedback Cosine Convolutional Neural Network (RiPFC-CNN-TOA) for the automatic and very effective diagnosis and classification of breast cancer in sonograms using the Tyrannosaurus Optimization Algorithm. First, UDIAT and BUSI dataset images are preprocessed with Fast Gradient Domain-Guided Image Filtering to improve contrast and eliminate noise while maintaining vital tumor features. Precise lesion segmentation is performed with a Machine learning-based Hybrid Mamba-Transformer model, capturing both spatial and sequential patterns of the images. For categorization, the RiPFC-CNN combines a Rotation-Invariant Attention Network and a Progressive Feedback Cosine CNN for feature-rich and orientation-invariant feature extraction. By modeling intelligent predator–prey behavior and fine-tuning weights, the Tyrannosaurus Optimization Algorithm also enhances the model's performance. The suggested approach obtains excellent performance: accuracy of 99.53% (BUSI) and 99.6% (UDIAT), precision over 99.3%, and large F1-scores and specificity, greatly enhancing diagnostic reliability with a significant reduction in false positives. This framework provides a robust, clinician-friendly algorithm for early breast cancer identification with excellent potential for real-world application in diagnostic pipelines.