AI-aided diagnostic performance for prostate MRI: systematic review and meta-analysis

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• AI is increasingly integrated into the prostate cancer diagnostic pathway.
• This study evaluates the diagnostic accuracy of AI assistance for prostate cancer MRI.
• The research focuses on clinically significant prostate cancer (csPCa).

Purpose of the Study:

• To estimate the diagnostic accuracy of AI assistance for csPCa detection using MRI.
• To compare the performance of AI-assisted radiologists versus standalone radiologists.
• To assess AI's utility in improving diagnostic performance across different experience levels.

Main Methods:

• Systematic literature search (Jan 2017-Oct 2024) across major databases.
• Meta-analysis using hierarchical summary receiver operating characteristic modeling.
• Pairwise comparison of AI vs. radiologists using sensitivity, specificity, PPV, NPV, CDR, and accuracy.

Main Results:

• Included 29 studies with 7398 patients; AI as assistant showed superior sensitivity, specificity, PPV, and NPV.
• AI assistance improved diagnostic performance for radiologists of all experience levels.
• Standalone AI showed higher specificity but lower sensitivity compared to human readers.

Conclusions:

• Integrating AI as an assistant enhances diagnostic accuracy in csPCa detection via MRI.
• AI assistance is particularly beneficial for less experienced radiologists.
• AI shows potential to optimize prostate cancer diagnostic workflows.