AI-optimized breast cancer prognostics: Robust prediction with transformer-based lesion localization and hard voting classifier

Abstract

Breast cancer remains a significant threat to global health, driving the demand for more effective and prompt detection methods. AI-optimized deep learning models are revolutionizing mammography screening, by substantially improving breast lesion recognition and diagnosis accuracy. This research introduces a groundbreaking approach that leverages advanced deep-learning architectures to enable early and precise breast cancer prognostics. The study proposes a novel combination of scaling techniques, depth-wise convolution, and max pooling layers to enhance feature extraction from mammographic images, facilitating the detailed prognosis of intricate lesion patterns in both benign and malignant cases. The method efficiently eliminates redundant features, identifies the most important ones, and improves detection efficiency while reducing computational complexity compared to advanced models. To combat overfitting and integrate outputs from multiple models, a hard voting classifier is employed, ensuring fine-grained lesion detection and addressing the challenges of limited training data in medical imaging. The robust voting process leverages diverse augmentation techniques across large mammography datasets to provide comprehensive outcomes. Additionally, the Swin Transformer’s performance is evaluated against nonparametric statistical tests, validating its suitability for mammography image classification. The proposed model was evaluated using three public datasets, accurately detecting breast lesions with a sensitivity score of 99.31%. The approach achieved an impressive accuracy of 98.5%, with a standard deviation of 0.085 using 10-fold cross-validation, and an optimal AUC of 0.98. These results underscore the model’s effectiveness and robustness, particularly in data-constrained scenarios, making it a cost-effective solution for early breast cancer detection. Our findings highlight the transformative potential of AI-driven solutions in advancing breast cancer diagnostics and emphasize their importance in medical imaging applications.