A Weighted Average Ensemble Approach for Breast Cancer detection using Deep Learning with Genetic Optimization:GeneOptBM-Net

Breast cancer poses a significant global health challenge, especially due to its asymptomatic early stages, which complicates detection. While Machine Learning (ML) and Deep Learning (DL) techniques are commonly employed, achieving a comprehensive diagnosis remains difficult. Medical experts stress early treatment as crucial for improving patients’ chances of recovery and mitigating associated risks. Many ComputerAided Diagnosis (CAD) systems have been proposed, often integrating various imaging modalities. However, most rely on either attention-based deep features or handcrafted features with a single classifier based models, lacking the ability to provide vital local information for precise tumor detection and robustness. Furthermore, existing breast cancer datasets suffer from class imbalance problem. Hence, this study introduces a novel Genetic Optimized Breast Mass Network (GeneOptBMNet) for early-stage breast cancer detection, leveraging ensemble techniques for improved robustness and accuracy over singleclassifier models. The proposed model integrates decisions from base learners and employs weighted averaging for final decisionmaking. The optimal combination of weights is determined using the Genetic Algorithm (GA). Furthermore, it enhances overall system performance by combining attention-based deep features and handcrafted features extracted using HOG, thus offering precise local information. To tackle class imbalance, the model incorporates the Borderline Synthetic Minority Over-sampling Technique (BSMOTE). Evaluation on BUSI and UDIAT datasets shows promising results, with average accuracies of 99.08% and $\mathbf{9 7. 0 2 \%}$, respectively.