Artificial intelligence: competitor or best friend of breast cancer radiologists?

Abstract

Around the world, over half a million women die of breast cancer every year. Screening mammograms were introduced to aid in early breast cancer diagnosis more than 30 years ago. Traditionally, disease determination from mammograms has been dependent upon radiologist interpretation. However, technologies have been developed to improve the sensitivity and specificity of mammograms beyond the capabilities of humans. Artificial intelligence (AI) using deep learning and convolutional neural networks has been employed in the past few years to improve early detection of breast cancer in digital mammography and breast tomosynthesis. Here, we aggregated data from past studies comparing the performance of experienced breast cancer radiologists, AI, and AI-informed radiologists. Our review showed that the performance of the AI system was greater than that of radiologists alone and that the combination of AI and radiologist input showed the greatest promise in detecting breast cancer. Our review demonstrates the potential of AI while also acknowledging potential pitfalls of this technology. Prospective studies in hospital settings are needed to fully establish the potential benefits of utilizing AI in breast cancer screening.