A concurrent, deep learning-based computer-aided detection system for prostate multiparametric MRI: a performance...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Prostate cancer diagnosis relies heavily on magnetic resonance imaging (MRI).
• Radiologist experience significantly impacts diagnostic accuracy in prostate MRI.
• Computer-aided diagnosis (CAD) systems show potential to augment radiologist performance.

Purpose of the Study:

• To assess the impact of a deep learning-based CAD (DL-CAD) system on the diagnostic performance of radiologists with varying experience levels in prostate MRI.
• To compare diagnostic accuracy, lesion detection, and grading correlation with and without DL-CAD assistance.
• To evaluate the effect of DL-CAD on reading times.

Main Methods:

• Retrospective multi-reader, multi-case study involving 172 patients with prostate MRI.
• Four radiologists (two experienced, two less-experienced) interpreted MRIs twice: once without and once with DL-CAD assistance.
• Diagnostic accuracy (AUC, sensitivity, specificity) and correlation between PI-RADS category and Gleason score were analyzed. Reading times were also compared.

Main Results:

• DL-CAD significantly improved the AUC for less-experienced radiologists (from 0.66 to 0.80).
• Experienced radiologists showed a smaller, non-significant AUC increase with DL-CAD.
• DL-CAD enhanced the correlation between PI-RADS category and Gleason score (0.45 to 0.57) and reduced median reading time.

Conclusions:

• DL-CAD significantly boosts diagnostic performance, particularly for less-experienced radiologists, enabling them to achieve accuracy comparable to experienced peers.
• The system aids in differentiating benign from cancerous prostate lesions and improves the correlation between imaging findings and cancer grade.
• DL-CAD integration into prostate MRI workflow offers a valuable tool for enhancing diagnostic consistency and accuracy across different experience levels.