A 19-study meta-analysis suggests robust detection of clinically significant prostate cancer (csPCa) with the combination of biparametric magnetic resonance imaging (bpMRI) and artificial intelligence (AI).
For the meta-analysis, recently published in Academic Radiology, researchers reviewed data from 6,286 patients, including 4,594 patients in internal validation testing cohorts, 795 patients from external validation testing groups and 897 patients who had unassisted radiologist interpretations.
Internal validation cohorts revealed average sensitivity of 88 percent and average specificity of 79 percent. Studies with external validation had average sensitivity and specificity of 85 percent and 83 percent respectively, according to the meta-analysis authors.
The researchers noted the combination of bpMRI and AI had a 91 percent average area under the receiver operating characteristic curve (AUC) for csPCa detection in both internal and external validation cohorts in contrast to 78 percent for radiologist assessments of bpMRI without AI.
“AI technologies enable unprecedented quantitative feature extraction from medical images, transcending traditional visual limitations and providing noninvasive insights into tumor microenvironments. Machine learning and deep learning algorithms can systematically analyze complex imaging data, identifying subtle patterns that human radiologists might overlook,” wrote lead study author Guangzhao Yan, M.D., who is affiliated with the Department of Emergency Medicine in the Emergency and Critical Care Center at Zhejiang Provincial People’s Hospital in Hangzhou, China, and colleagues.
While bpMRI provides advantages over multiparametric MRI (mpMRI) with respect to shorter exam times, cost-effectiveness and patient safety, the meta-analysis authors noted that morphological constraints and subjective evaluations can hinder the effectiveness of bpMRI.
Three Key Takeaways
1. High diagnostic performance. The combination of biparametric MRI (bpMRI) and AI demonstrated high sensitivity (85-88 percent) and specificity (79-83 percent) for detecting clinically significant prostate cancer (csPCa), outperforming unassisted radiologist assessments.
2. AI enhances image interpretation for csPCa. The meta-analysis authors noted that adjunctive AI can improve tumor classification and reduce interobserver variability. The use of adjunctive AI improved the diagnostic capability of bpMRI, achieving an average AUC of 91 percent compared to 78 percent for radiologists interpreting bpMRI alone.
3. Potential for clinical implementation. While bpMRI offers advantages such as reduced scan time and cost-effectiveness over multiparametric MRI (mpMRI), AI can help mitigate its limitations by refining morphological feature assessments and providing more objective, reproducible results.
However, the researchers emphasized that deep learning and machine learning technologies can enhance bpMRI capabilities in characterizing csPCa.
“AI improves the accuracy and reliability of tumor classification by effectively extracting morphological features pertinent to PCa. Moreover, AI reduces the variability associated with the subjective interpretations of radiologists in conventional diagnostic practices, thus providing more objective and consistent analytical results,” added Yan and colleagues.
(Editor’s note: For related content, see “Meta-Analysis Shows No Difference Between bpMRI and mpMRI in Ruling Out csPCa,” “Can Deep Learning Ultra-Fast bpMRI Have an Impact in Prostate Cancer Imaging?” and “Emerging AI Software for Prostate MRI Offers 95 Percent Sensitivity for csPCa.”)
In regard to study limitations, the authors conceded that only four of the reviewed studies had external validation cohorts and acknowledged a lack of consistent reporting on the expertise level of reviewing radiologists in the studies included in the meta-analysis.
