CoAtNet-Lite: Advancing Mammogram Mass Detection Through Lightweight CNN - Transformer Fusion with Attention Mapping

Mammograms are a main diagnostic technique for breast cancer screening, but their interpretation can be difficult and prone to human error. In the field of medical imaging, with a specific focus on mammogram analysis, this research paper introduces an innovative approach to mammogram mass detection. The study leverages an optimized CoAtNet (a hybrid model that combines depthwise convolution and self-attention via relative attention while stacking convolution and attention layers selectively) applied to a publicly available mammography dataset. The proposed model exhibits encouraging performance metrics, boasting precision and recall rates of 0.77 and 0.78, respectively, along with an F1-score of 0.77. This notable improvement distinguishes it from other existing deep learning (DL) and machine learning (ML) models that we have tested before-hand. Additionally, we attempt to enhance the interpretability and transparency of the model's decision-making process, the research integrates explainable AI techniques such as LIME (Local Interpretable Model-agnostic Explanations) and Grad-CAM (Gradient-weighted Class Activation Mapping). These tools play a pivotal role in providing insightful explanations for the model's predictions, thereby increasing the trustworthiness and reliability of AI in medical diagnostics, especially in the context of breast cancer detection. Moreover, the research underscores the significance of combining deep learning architectures with explainable AI methodologies, highlighting their potential to advance the accuracy and transparency of medical imaging technologies.