Optimized attentional transfer learning based digital twin convolutional network for breast cancer detection with mammography images

Breast cancer detection involves identifying malignant tumors in breast tissue using diagnostic techniques like mammograms and deep learning models. Key challenges in existing models include high false positive and negative rates, difficulty detecting small or dense masses, variability in image quality, and limited generalization across populations. Hence, to address the prevailing limitation of the existing methods, the research proposes the Optimized Attentional Transfer Learning-enabled Digital Twin Convolutional Network (OAL-DTCNet) for breast cancer detection, which unites the advanced techniques with a robust classification system. This model leverages dual convolutional pathways to extract and refine features from mammographic images, enhancing its ability to accurately identify cancerous tissues. The adaptive learning mechanism in the OAL-DTCNet model refers to its ability to continuously refine and improve its predictions based on past data. The MIVIC is proposed as the optimization mimicking the foraging characteristics to further strengthen the model by enabling effective exploration of complex search spaces while efficiently focusing on optimal regions. This hybrid approach leads to enhanced feature selection and classification performance, ultimately representing a significant advancement in automated breast cancer-detecting exposure. The designed OAL-DTCNet model attains a high accuracy, specificity, and sensitivity of 96.40%, 95.99%, and 96.80% using the Breast Micro-Calcifications Dataset (BMCD).