Ohabm-net: an enhanced attention-driven hybrid network for improved breast mass detection

Breast cancer begins in the breast tissues and can progressively spread to other parts of the body. Early detection is crucial, as it allows for timely treatment, potentially saving lives. Researchers have devised methods to detect cancer in its early stages. However, the majority of the approaches primarily utilize either attention-based deep models or handcrafted features-based models for providing local information. However, both these approaches lack the ability to provide crucial local information for precise tumor detection. Additionally, available breast cancer datasets are inherently imbalanced. To address these challenges, this paper presents the Optimized Hybrid Attention Breast Mass Network (OHABM-Net) for breast cancer detection. The proposed OHABM-Net uses a newly developed hybrid attention-based feature extraction network that combines attention-based deep features and handcrafted features using HOG which provides precise local information thereby enhancing overall performance of the system. Moreover, the proposed model incorporates the Borderline Synthetic Minority Over-sampling Technique (BSMOTE) to resolve the class imbalance problem. Furthermore, to improve the performance of the system, the proposed model incorporates BM3D denoising filter and hill climbing-based optimization method to fine-tune the feature extraction network, culminating in classification through SVM. The proposed OHABM-Net evaluates with the BUSI and UDIAT datasets, achieving average accuracies of 98.11% and 96.78%, respectively, which surpass the performance of existing models.