A novel deep convolutional network for breast cancer detection in magnetic resonance images with explainable artificial intelligence

Breast cancer is the second leading cause of mortality among women worldwide. Although Magnetic Resonance Imaging (MRI) provides higher sensitivity than other screening techniques, it is prone to false positives, unnecessary biopsies, and challenges in detecting small tumors in dense breast tissue, making accurate interpretation difficult. To address this, the AttenBreast Cancer Deep Convolutional Network (AttenBC-Net) model is devised for detecting breast cancer through MRI images. Also, to enhance model transparency and interpretability, the Shapley Additive Explanations (SHAP) algorithm is applied. This approach aims to improve accuracy and ensure better interpretability in breast cancer detection. The process starts with Pixel Enhancement Transformation (PET) to improve image quality, followed by cancer region segmentation using Residual U-Net (ResU-Net) with a log loss function. Image augmentation approaches, like random erasing, flipping, and rotation, are performed to improve model generalization and minimize overfitting. Next, features including statistical, shape, Gabor, Local Vector Pattern (LVP), and Fuzzy Local Binary Pattern (FLBP) are extracted. AttenBC-Net, which integrates the Breast Cancer Deep Convolutional Neural Network (BCDCNN) and the Shuffle Attention Network (SA-Net), is used for breast cancer detection. Furthermore, AttenBC-Net attained a highest sensitivity of 97.468%, accuracy of 96.568%, and specificity of 96.568%.