AI Helps General Radiologists Achieve Specialist-Level Performance in Interpreting Mammograms – Radiography

Image: Saige-DX is a custom-built categorical AI system designed to assist in breast cancer screening (Photo courtesy of DeepHealth)

Breast cancer, affecting one in eight women during their lifetimes, becomes far more treatable when detected early. The five-year relative survival rate for stage 1 breast cancer is an encouraging 99%, highlighting the importance of early detection. Now, groundbreaking research reveals that an advanced artificial intelligence (AI) technology, designed for mammography, can significantly enhance the early detection and diagnosis of breast cancer by enabling general radiologists to perform at the level of specialists.

DeepHealth’s (Los Angeles, CA, USA) Saige-Dx is a custom-built categorical AI system that automatically spots suspicious lesions in mammograms, assigning a level of suspicion to each finding and the entire case. In a pivotal study conducted by DeepHealth, the interpretative skills of 18 physicians, including both breast specialists and general radiologists, were evaluated. The researchers analyzed 240 retrospectively gathered digital breast tomosynthesis scans for cancer indicators. Utilizing Saige-DX, each radiologist demonstrated an enhanced ability to interpret mammograms. Their average diagnostic accuracy, measured by the area under the receiver operating characteristic curve, improved from 0.87 to 0.93 with the aid of AI.

The improvement was significant in both groups of radiologists, with generalists adding 0.08 to their accuracy score and specialists improving by 0.05. This enhanced performance was consistent across various cancer characteristics, such as lesion type and size, and across patient subgroups, including different races and ethnicities, ages, and breast densities. The researchers attribute this success to the AI system’s design, which specifically addresses challenging cases. The algorithm was trained using a dataset that included cancers previously missed in clinical settings by radiologists. Furthermore, Saige-Dx’s minimal use of “bounding boxes” for marking breast images likely prevented radiologists from being overwhelmed by excessive markings, a common issue with less precise computer-aided diagnostic tools.

“The mean performance of general radiologists with AI exceeded that of breast imaging specialists unaided by AI, suggesting that the AI software could help patients receive specialist-level interpretations for their screening mammogram even if interpreted by a general radiologist,” the researchers noted. “The benefits of using AI are not limited to generalists, as specialists also showed improved performance.”

“In conclusion, our results show that general radiologists may achieve specialist-level performance when interpreting screening DBT mammograms with the aid of AI and that specialists can achieve even higher performance (increased sensitivity and specificity) in a diverse population across multiple cancer types,” stated Jiye G. Kim, PhD, DeepHealth’s director of clinical studies.