Explainable AI in breast cancer ultrasound imaging: current developments and challenges

Breast cancer has been one of the most common causes of cancer mortality in the world, and thus, early and correct diagnosis is crucial in enhancing patient outcomes. The use of ultrasound imaging as a complementary diagnostic tool is very common because it is safe, accessible, and cost-effective, especially in resource-strained environments. The deep learning methods have achieved spectacular success in the past few years in their effort to automate the process of detecting and classifying breast cancer based on ultrasound images. The transparency of these models, however, is not always clear, and this problem is commonly known as the black-box problem, which is a serious obstacle to the implementation of these models in clinical practice. Explainable Artificial Intelligence (XAI) is an emerging technology that opens business opportunities to improve the interpretability and reliability of deep learning models by offering insights into the decision-making process of these models. This mini review will be a summary of the current developments in XAI applied to the ultrasound image of breast cancer, such as saliency-based approaches, model-agnostic methods, and attention mechanisms. Moreover, it outlines some of the most significant obstacles, which include a lack of standardized evaluation metrics, clinical validation, and the fact that it is hard to interpret an explanation of noisy imaging conditions. Lastly, possible future pathways are outlined to close the gap between the currently successful AI systems and their successful application to clinical practice.