In a study reported in The Lancet Oncology (PI-CAI), Saha et al found that an artificial intelligence (AI) system’s readings of magnetic resonance imaging (MRI) outperformed study radiologist readings using Prostate Imaging—Reporting and Data System (PI-RADS) version 2.1 in detecting clinically significant prostate cancer. That said, AI performance did not show noninferiority to that of radiology readings made during multidisciplinary routine practice.
Study Details
The international multicenter study included 10,207 MRI examinations performed between January 2012 and December 2021 at sites in the Netherlands and Norway; of these, 9,207 cases were used to develop the AI system. A separate cohort of matched cases was used compare AI system readings with study radiologist readings. In a second cohort, performance of AI readings was compared with historical radiology readings made during multidisciplinary routine practice at sites in 20 countries.
Key Findings
Among the total of 10,207 cases, 2,440 cases had histologically confirmed Gleason grade group ≥ 2 prostate cancer. Among 400 cases in which the AI system was compared with the study radiologist readings, the area under the receiver operating characteristic curve (AUC) value was 0.91 (95% confidence interval [CI] = 0.87–0.94) for the AI system, significantly better (P < .0001) and noninferior to the AUC value of 0.86 (95% CI = 0.83–0.89) for radiologist readings. At a mean PI-RADS ≥ 3 threshold, the AI system detected 6.8% more cases with Gleason grade group ≥ 2 cancers at the same specificity (57.7%, 95% CI = 51.6%–63.3%) or had 50.4% fewer false-positive results and 20.0% fewer cases with Gleason grade group 1 cancers at the same sensitivity (89.4%, 95% CI = 85.3%–92.9%).
Among 1,000 cases in which the AI system was compared with the radiology readings made during multidisciplinary practice, noninferiority was not confirmed; the AI system showed lower specificity (68.9%, 95% CI = 65.3%–72.4%) vs 69.0% (95% CI = 65.5%–72.5%) at the same sensitivity (96.1%, 95% CI = 94.0%–98.2%) at the PI-RADS ≥ 3 threshold.
The investigators concluded, “An AI system was superior to radiologists using PI-RADS (2.1), on average, at detecting clinically significant prostate cancer and comparable to the standard of care. Such a system shows the potential to be a supportive tool within a primary diagnostic setting, with several associated benefits for patients and radiologists. Prospective validation is needed to test clinical applicability of this system.”
Anindo Saha, MSc, of the Diagnostic Image Analysis Group, Department of Medical Imaging, Radboud University Medical Center, Nijmegen, Netherlands, is the corresponding author for The Lancet Oncology article.
Disclosure: The study was funded by Health~Holland and EU Horizon 2020. For full disclosures of the study authors, visit thelancet.com.
