Deep Learning Model Noninferior to Radiologists in Detecting Clinically Significant Prostate Cancer at MRI – Radiology

Study Design and Population: This retrospective study evaluated the performance of a deep learning (DL) model for detecting clinically significant prostate cancer (csPCa) using multiparametric MRI (mpMRI) images from 5215 patients (5735 examinations) with a mean age of 66 years. The study included patients who underwent prostate MRI between January 2017 and December 2019 at a single academic institution. The DL model was trained on T2-weighted, diffusion-weighted, and contrast-enhanced MRI sequences, with pathologic diagnosis as the reference standard.

Main Findings: The DL model achieved an area under the receiver operating characteristic curve (AUC) of 0.89 on the internal test set and 0.86 on an external test set, demonstrating noninferiority to radiologists, who had AUCs of 0.89 and 0.84, respectively. Additionally, the combination of the DL model and radiologists improved diagnostic performance (AUC of 0.89). Gradient-weighted class activation maps (Grad-CAMs) effectively localized csPCa lesions, overlapping with true-positive cases in 92% of internal test set and 97% of external test set cases.

Implications for Practice: The DL model showed comparable performance to experienced radiologists in detecting csPCa at MRI, suggesting its potential to assist radiologists in improving diagnostic accuracy and reducing interobserver variability. Future research should focus on integrating the model into clinical workflows and assessing its impact on biopsy targeting.

Reference: Cai JC, Nakai H, Kuanar S, et al. (2024). Fully Automated Deep Learning Model to Detect Clinically Significant Prostate Cancer at MRI. Radiology, 312(2): e232635. DOI: https://doi.org/10.1148/radiol.232635.