Artificial Intelligence in Medicine: The Lead is Narrowing in the Radiologist vs. AI Race

No radiologist was born knowing how to interpret MRI scans of the prostate gland. They must be trained to interpret images, and the rate at which they improve is called their learning curve. According to Rosencrantz et al. (2017),

At a basic level, by seeing many cases, radiologists experience innate self-directed visual learning, even without any active teaching or feedback—that is, through spatial pattern recognition and visual memory, radiologists gain an appreciation for what constitutes normal versus abnormal appearances simply by reflecting on a high volume of studies, as occurs with growing experience.^[i]

The authors caution that self-learning can leave gaps, so ongoing education via courses/presentations and feedback from experienced radiologists may accelerate the learning curve and foster improved accuracy. All of this requires time, even years, to become proficient at interpreting prostate MRI scans. The PI-RADS scoring system was developed to compensate for inconsistent radiologist performance due to inexperience. In many ways, it’s like a detailed user’s manual for capturing and interpreting prostate MRI, and it has set the interpretation standard thanks to its specific details. Still, there’s no substitute for experience. Or is there?

Will Artificial Intelligence top radiologists?

There is no doubt that MRI before biopsy has improved prostate cancer (PCa) diagnosis. “However, the use of MRI is impeded by its dependence on experienced radiologists and the inter-radiologist variability in [PI-RADS] scoring of prostate MRI exams,” writes Martin Eklund,^[ii] but AI may be able to come to the rescue. A consortium of multinational radiologists and AI developers developed and trained an AI system for detecting clinically significant PCa tumors (grade group 2 or higher).^[iii] They then compared the diagnostic power of their program against the average performance of 62 radiologists who were using the PI-RADS (version 2.1) system of scoring.

Both the AI model and the radiologists read MRI scans of 400 patients for whom pre-prostatectomy images were available, and whose surgical specimens had confirmed grade group 2 or higher PCa. These surgical results were the standard against which the AI detection and reader interpretations were evaluated. The authors found that their AI program’s results were “…superior to radiologists using PI-RADS (2.1), on average, at detecting clinically significant prostate cancer and comparable to the standard of care.”^[iv] Of course, more testing is needed to confirm these results. Even so, they are a clear promise that AI can benefit busy radiologists in several ways, including easing their workload by rapidly flagging suspected significant PCa for further human inspection, more accurately rule out insignificant disease which would eliminate unnecessary biopsies, and ultimately lessen the rate of over-diagnosing and over-treating PCa patients. If we think of AI accuracy as racing to get ahead of radiologists who need years of training and experience, it appears that the human lead is narrowing—but AI will never replace humans as the final decision-makers.

At the Sperling Prostate Center, we are already integrating the benefits of AI to aid in detection, and we are excited about the future possibilities that can only be in patients’ best interests.

NOTE: This content is solely for purposes of information and does not substitute for diagnostic or medical advice. Talk to your doctor if you have health concerns or questions of a personal medical nature.

[i] Rosenkrantz AB, Ayoola A, Hoffman D, Khasgiwala A et al. The Learning Curve in Prostate MRI Interpretation: Self-Directed Learning Versus Continual Reader Feedback. AJR Am J Roentgenol. 2017 Mar;208(3):W92-W100.
[ii] Eklund, M. Artificial intelligence for scoring prostate MRI: ready for prospective evaluation. Lancet Oncology, 2024 Jul; 25(7):827-28.
[iii] Saha A, Bosma JS, Twilt JJ, van Ginneken B et al. Artificial intelligence and radiologists in prostate cancer detection on MRI (PI-CAI): an international, paired, non-inferiority, confirmatory study. Lancet Oncol. 2024 Jun 11:S1470-2045(24)00220-1.
[iv] Ibid.