Dense-RBF Grad-CAM: A Clinically Trustworthy AI Framework for early Breast Cancer Detection using mammography images

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Breast cancer continues to be the most common cause of cancer-related deaths in women globally. The earlier breast cancer is identified, the higher the chance of survival. In screening dense breast tissue, the standard diagnostic approach has essential limitations, including a relatively high false-negative rate and poor interpretability. This research introduces a Dense Radial Bias Functional Gradient-weighted Class Activation Mapping method optimized using the Revolution Optimization Algorithm (DRBF-GradCAM_ROA) for early breast cancer detection at advanced stages in clinical settings. The raw data is preprocessed using BM3D filtering to reduce noise, and then it is augmented by rotating and flipping the medical images. Region-growing segmentation is then applied to mammogram images using the U-Net model. The most relevant features are then selected from the mammography images using the Histogram of Oriented Gradients (HOG). The DRBF-GradCAM_ROA model is designed to efficiently improve early breast cancer detection and improve the performance of mammogram images. To boost interpretability, the Grad-CAM visualization is incorporated into the framework, enabling clinicians to easily identify the discriminative regions of the mammogram that drive classification decisions. The inclusion of Explainable AI (XAI) enhances both trustworthiness and transparency of the model, making it more applicable for clinical decision making. The DRBF-GradCAM model uses the ROA algorithm during training to further improve mammogram image analysis. The method's accuracy, precision, recall, and F1-score were 99.87%, 99.79%, and 99.85%, respectively. This supports prompt clinical diagnosis and therapy by greatly enhancing early breast cancer detection in medical images.