It was 24 years ago that a new website called AuntMinnie.com was launched, providing interesting news on medical imaging. Apparently, the provenance for Aunt Minnie dates to the 1940s when a radiologist, Ben Felson, described “a case with radiologic findings so specific and compelling that no realistic differential diagnosis exists. Thus, if the image looks like your Aunt Minnie, then it must indeed be your Aunt Minnie!” Ever since 2000, the editors at AuntMinnie.com have identified the “best of the best in radiology,” and have awarded the Minnies to recognise people and technologies that have made a meaningful impact.
Not surprisingly, winner of the Most Significant News Event in Radiology in 2023 was the use of artificial intelligence (AI). This exciting technology has now moved into the mainstream, improving clinical decision-making, streamlining workflow and patient scheduling, and enhancing image interpretation. For example, the application of AI has led to an improvement in the interpretation of CT exams of the head, and the ability to detect breast cancer at an earlier stage.
Runner up in the category of Most Influential Radiology Researcher was Dr Emily Conant (seen left), Chief of Breast Imaging at the University of Pennsylvania. As we highlighted earlier this year, she and her colleagues provided evidence that digital breast tomosynthesis (DBT) was superior to full-field digital mammography (FFDM) when screening for early detection of breast cancer. Conant has recently published papers on AI, one on its potential use to reduce the harms of screening, and a second on AI applied to DBT images. In 2023 alone, she has published over 20 original articles in high quality journals.
Winner of Scientific Paper of the Year was Dr Jung Hyun Yoon (seen below right) from Seoul, Korea whose paper, published in Radiology, was entitled “Standalone AI for breast cancer detection at screening digital mammography and digital breast tomosynthesis: a systematic review and meta-analysis.” Her review encompassed 16 studies published between 2017 and 2022 and included over one million examinations of 500,000 women. Yoon concluded that standalone AI algorithms performed as well or better than radiologists for detecting breast cancer.
Finally, winner of the Hottest Clinical Procedure was the use of AI for predictions. Among the applications are mortality risk from lung cancer, detecting pneumonia by analysing chest X-rays, and the early diagnosis of breast cancer, including the use of AI algorithms and ultrasound images to detect breast lumps and to diagnose invasive lobular carcinoma. As Dr Yoon commented, “The radiology field, especially breast imaging, should have a leading role in implementing AI […] for the right intentions, to benefit our patients.”
The biggest problem with AI, certainly for breast evaluation, is the fact that the computer (neural network) cannot tell you why it is concerned about an area in the breast. We are heading into an era where we will perform interventions and treatment without having a clear reason for why we are doing it. I expect that computers should be better than humans at finding early cancer, but there will be major risks that will evolve in all uses of AI that come from our reliance on computers with no way to actually question “their reasoning”.
I agree with your comment, Dan, that one of the shortcomings of AI algorithms — at least for now — is that they are unable to tell us exactly _why_ a particular diagnosis has been suggested, or lesion identified. However, I am optimistic that the day will come when they are able to explain their decisions.