Improved Breast Cancer Localization by using Hybrid Approach with Swarm Intelligence and Deep Learning

Abstract

One of the most prevalent illnesses affecting women globally is breast cancer, necessitating enhanced detection approaches for better clinical results. For accurate breast cancer localization, this study suggests a hybrid deep learning architecture that combines massive language models, swarm intelligence, and transfer learning. The suggested method utilizes the ResNet50 network as the feature extractor. In order to further boost model performance, the Honeybee Algorithm was utilized to automatically tune important hyper parameters such as optimization as well as generalization. Performance reached an accuracy of 90% for the Ultrasound dataset, 86.15% for the MRI data, and 95.7% for both datasets combined, showing the advantages of multimodal learning and swarm optimization. Beyond image analysis, a module for interpreting clinical reports was implemented with a BERT based large language model that was able to extract useful diagnostic information such as type of tumor, malignancy level, and treatment recommendations. Testing based on parameters like accuracy, precision, recall, F1-score, sensitivity, specificity, and AUC validated the strength of the suggested approach. The combination of HBA-tuned ResNet50 and LLM based Bert pre-trained model clinical report comprehension creates an integrated and smart diagnostic paradigm that improves not only the accuracy but also the explainability of breast cancer diagnosis systems. This research helps create dependable, explainable, and AI-based healthcare solutions that can assist radiologists in real-time clinical decision-making.