Digital mammogram based robust feature extraction and selection for effective breast cancer classification in earlier stage

Abstract

Breast cancer poses a significant health risk for women, demanding early detection to mitigate its mortality impact. Leveraging the power of Deep Learning (DL) in medical imaging, this paper introduces a hybrid model that integrates YOLOv7 and Half UNet for feature extraction. YOLOv7 identifies and localizes potential cancerous regions, while Half UNet focuses on extracting pertinent features with its encoder-decoder structure. The fusion of these discriminative features, coupled with feature selection via Coati Optimization, ensures a comprehensive and optimized dataset. The selected features then feed into the CatBoost classification algorithm, refining parameters iteratively for precise predictions and minimizing the loss function. Evaluation metrics, including precision, recall, specificity, and accuracy, demonstrate the model’s superior performance. Notably, the proposed model surpasses existing methods in early-stage breast cancer detection. Beyond numerical metrics, its significance lies in the potential to positively impact patient outcomes and increase survival rates. By amalgamating cutting-edge DL techniques, the model excels in identifying intricate patterns crucial for early cancer detection. The efficient fusion of YOLOv7 and Half UNet, coupled with feature optimization through Coati Optimization, sets this model apart. This research contributes to the evolving landscape of medical imaging and DL applications, emphasizing the potential for enhanced breast cancer diagnosis and improved patient prognoses.