Evaluating Artificial Intelligence-Assisted Prostate Biparametric MRI Interpretation: An International Multireader...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Prostate Cancer Diagnostics

Background:

• Variability in prostate biparametric MRI (bpMRI) interpretation impacts diagnostic reliability for prostate cancer (PCa).
• Artificial intelligence (AI) offers potential to standardize interpretations and enhance diagnostic accuracy in bpMRI.

Purpose of the Study:

• To evaluate the impact of a deep learning AI model on lesion- and patient-level detection rates for clinically significant PCa (csPCa) and PCa.
• To assess the effect of AI on interreader agreement in bpMRI interpretations.

Main Methods:

• Retrospective, multireader, multicenter study involving 180 patients undergoing bpMRI.
• Six radiologists interpreted bpMRI scans with and without AI assistance.
• Reference standard included whole-mount pathology or systematic biopsies; lesion-level sensitivity, PPV, patient-level AUC, and interreader agreement were assessed.

Main Results:

• AI significantly improved lesion-level positive predictive value (PPV) for both csPCa and PCa detection (p < .001).
• AI assistance led to a slight reduction in lesion-level sensitivity (p = .01) but showed no significant difference in patient-level AUC.
• Interreader agreement for PI-RADS scores (lesion and patient level) and lesion size measurements significantly improved with AI assistance (p < .001).

Conclusions:

• AI integration in bpMRI interpretation enhances lesion-level PPV and interreader agreement.
• While sensitivity slightly decreased, AI shows potential to improve consistency and reduce false positives in PCa diagnosis.
• Further AI model optimization is needed to balance sensitivity and specificity for improved clinical impact.