Automated MRI system for clinically significant prostate cancer detection development validation and real-world...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Prostate MRI is crucial for detecting clinically significant prostate cancer (csPCa).
• PI-RADS scoring variability hinders reproducibility and efficiency in prostate MRI interpretation.
• There is a need for standardized, automated tools to improve csPCa detection and reporting.

Purpose of the Study:

• To develop and validate an automated MRI-based decision aid (ProAI) for estimating patient-level csPCa risk.
• To assess ProAI's performance against PI-RADS scoring in terms of accuracy and consistency.
• To evaluate the impact of ProAI on clinical workflow, including radiologist accuracy, reading time, and workload.

Main Methods:

• Development and validation of ProAI using biparametric MRI data from 7849 examinations across multiple centers and public datasets.
• Comparative analysis of ProAI's diagnostic performance (AUC) against PI-RADS scoring.
• Multi-reader, multi-case study to evaluate ProAI's effect on clinician accuracy and reading time.
• Prospective implementation study to assess real-world performance and workload reduction.

Main Results:

• ProAI achieved a pooled external test AUC of 0.93, comparable to PI-RADS, with improved inter-case consistency.
• Clinician accuracy increased from 0.80 to 0.86 with ProAI assistance, alongside reduced reading times.
• Prospective implementation maintained performance (AUC 0.92) and resulted in a 32% reduction in radiology workload.
• ProAI demonstrated generalizable performance on the TCIA cohort (AUC 0.83).

Conclusions:

• An automated MRI-based decision aid (ProAI) can standardize prostate cancer reporting and enhance diagnostic accuracy.
• ProAI improves inter-reader consistency and reduces radiologist workload, optimizing prostate cancer care pathways.
• This AI tool shows potential for widespread clinical adoption in prostate cancer diagnostics.