AI ups detection of small prostate lesions by nearly 20%
New research indicates that artificial intelligence could be especially promising within the realm of prostate cancer, helping radiologists identify smaller, more inconspicuous lesions on imaging.
In fact, rads’ sensitivity increases by nearly 20% when they utilize the support of AI detection tools, according to the new data. Researchers involved in this latest study suggest that AI support could help address issues with reader variability on prostate imaging.
“Prostate MRI has proven highly effective for PCa detection, leading to its adoption in international guidelines as a primary diagnostic tool for suspected cases,” Jiule Ding, MD, with the department of radiology at Third Affiliated Hospital of Soochow University in China, and colleagues note. “Despite its accuracy, prostate MRI interpretation varies significantly across medical institutions and depends heavily on reader expertise.”
Numerous studies have shown that AI improves PCa detection on MRI, but most have tested the tool in limited settings, leaving a gap in understanding its capabilities. This study addresses this shortcoming with its fully-crossed, multi-reader, multi-case (MRMC) design.
“The fully-crossed design of MRMC minimizes bias in case-to-reader assignment, thereby strengthening the validity of the evaluation,” the authors explain. “Despite its advantages, to date, only three studies have employed the MRMC design to investigate the assistive value of AI systems in PCa identification via MRI.”
Across three institutions, 10 non-expert radiologists were tasked with interpreting biparametric prostate MRI cases comprising T2WI, diffusion-weighted images both with and without the help of AI. These performances were compared against each other to determine how AI affected their decision-making and accuracy, and also to AI’s performance as a solo reader.
The group reported on a total of 407 cases. When utilizing AI support, readers recorded significantly higher lesion-level sensitivity, which climbed from 67.3% without AI to 85.5% with it. Case-level sensitivity rose by around 4% with AI’s help, while the AFROC-AUC was also higher for aided versus unaided reading, at 86.9% compared to 76.1%. Although AI performed well a solo reader, it did not outshine readers with AI assistance.
AI use was especially beneficial in spotting smaller lesions, the authors reported.
“Notably, for small PCa lesions (maximal diameter ≤1 cm), the lesion-level sensitivity was 38.3% for radiologists alone, but it increased by 94.8% with the addition of the AI system, whereas for larger lesions (maximal diameter >3 cm), the sensitivity increased by only 7.2%,” the group explains. “These results suggest that the AI system primarily enhanced the diagnosis and localization of smaller PCa lesions.”
While the study had limitations, researchers suggest their study design provides “robust evidence” supporting the use of similar tools in prostate cancer clinical settings.
Learn more about the findings here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.