A new meta-analysis of 49 randomized controlled trials (RCTs) revealed mixed results for adjunctive AI in improving detection for a variety of cancers on medical imaging.
For the meta-analysis, recently published in the Journal of the American College of Radiology, researchers reviewed data from 49 RCTs with colorectal cancer (CRC) imaging accounting for nearly 80 percent of the reviewed studies. The research also included evaluations of single RCTs assessing AI for prostate cancer, breast cancer and lung cancer detection.
For RCTs assessing AI for CRC detection, the meta-analysis authors found that AI use resulted in a 22 percent improvement in adenoma detection and a 20 percent increase in polyp detection based on pooled relative risks (RRs) from 39 RCTs. There was no significant impact with AI in detection of advanced adenomas or CRC, according to the researchers.
“This finding suggested that AI assistance provided less value for the detection of these advanced lesions, which were large and rarely overlooked …,” wrote lead meta-analysis author Jinlu Song, M.D., who is affiliated with the Xiangya School of Public Health at Central South University in Changsha, China, and colleagues
The researchers pointed out that adjunctive AI facilitated a 20 percent increase in breast cancer detection, a 40 percent improvement in prostate cancer detection and more than double the detection for actionable lung nodules and high-risk esophageal lesions. However, they also acknowledged that these pooled percentages were based on single RCTs.
Three RCTs for liver cancer and two RCTs for gastric cancer revealed no significant impact for adjunctive AI detection, according to the meta-analysis authors.
“AI-assisted examinations may improve certain detection rates but not all among seven cancer types,” added Song and colleagues.
Three Key Takeaways
- Adjunctive AI improves detection of early colorectal lesions but not advanced disease. Across 39 RCTs, AI increased adenoma detection by 22 percent and polyp detection by 20 percent, but showed no benefit for detecting advanced adenomas or colorectal cancer (CRC).
- Early evidence for other cancers is promising but limited. Single RCTs suggest AI may enhance detection of breast cancer (↑20 percent), prostate cancer (↑40 percent), and actionable lung nodules/high-risk esophageal lesions (more than doubled detection), but RCT data for AI-assisted detection of cancer remain sparse outside of CRC.
- No RCTs evaluated patient outcomes, highlighting a major evidence gap. While AI may boost some detection metrics, its impact on meaningful clinical outcomes is unknown, and future RCTs must prioritize patient-centered endpoints to determine true clinical value.
The researchers also noted that none of the 49 RCTs evaluated the impact of adjunctive AI on patient outcomes.
“Our results highlight the notable lack of patient-centered outcomes, which are essential for evaluating the actual benefits and risks for patients. We call for future research to prioritize the assessment of the effectiveness of AI on patient-centered outcomes beyond diagnostic accuracy, using well-designed RCTs,” emphasized Song and colleagues.
(Editor’s note: For related content, see “MRI-Based Deep Learning for Lymph Node Metastasis Detection in Colorectal Cancer: What a New Meta-Analysis Reveals,” “Study: AI-Generated ADC Maps from MRI More Than Double Specificity in Prostate Cancer Detection” and “New bpMRI Study Suggests AI Offers Comparable Results to Radiologists for PCa Detection.”)
In regard to limitations with the meta-analysis, the authors acknowledged that aside from colorectal cancer (CRC), a limited number of RCTs thwarted meta-analysis of AI detection for other cancers. They also conceded significant heterogeneity among the reviewed studies examining AI in CRC detection and that more than half of these studies were comprised of Asian cohorts, limiting extrapolation of the study findings to broader populations.
