Outlook into the future of AI in mammography reading


Utilizing artificial intelligence (AI) and deep learning, computer-aided detection (CAD) systems are becoming increasingly intelligent. The basis for this intelligence: The algorithms are fed with thousands upon thousands of clinical images, teaching them how abnormalities can look like. As a result, CAD systems have become highly reliable – and have the power to change the reading workflow in breast cancer screening.

CAD image with lesion score

Already, new CAD systems show not only the position of a lesion or microcalcifications. Some can also provide a lesion score, reflecting the probability of cancer. And certain solutions can provide a case score combining the results of the different views of a mammography study. The advantage is obvious: Radiologists can sort the cases following the case scores – so they can read their cases in a chosen order or directly set up subsequent examinations like ultrasound or a biopsy right away. These types of solutions are evolving from classic CAD system to provide clinical decision support.

There are even considerations that the new CAD systems can entirely replace the second reader. In the future, some even hypothesize that these systems could even replace both readers in very straight-forward cases. Radiologists would then only have to read the cases where the CAD system found something suspicious.

In some years, CAD or clinical decision support systems might even be able to provide more information on a suspicious finding than they do now. They might, for example, define the tumor and make therapy recommendations – which would mean biopsies could no longer be needed. This would clearly revolutionize workflow and procedures in mammography screening.