Development and validation of a deep learning-based model for predicting prostate cancer in patients with gray-zone PSA levels: a comparative study with clinician observations.
This study aimed to develop and validate a Deep Learning(DL)-based model for detecting Prostate cancer (PCa) in patients with gray-zone PSA levels using multiparametric Magnetic Resonance Imaging (mp-MRI).
This retrospective multi-center study initially reviewed data from 1778 patients with suspected prostate lesions who presented at multiple hospitals between December 2015 and October 2023. The DL algorithm was trained on T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC) values. The performance of the algorithm was evaluated using the area under the receiver operating characteristic (ROC) curve and compared to the diagnostic performance of radiologists with varying levels of expertise.
The final study cohort comprised 274 patients, who were divided into training, validation, and test sets in a 6:2:2 ratio, corresponding to 162, 56, and 56 patients, respectively. The mean age of the patients was 71.8 years, and the cohort included 90 patients with prostate cancer and 184 without. The multi-attribute Convolutional Neural Networks (CNNs) model achieved an area under the curve (AUC) of 0.84, surpassing individual ADC and T2WI models. It also outperformed a transformer-based model (AUC 0.79). When compared to radiologists, the DL model performed similarly to experienced radiologists (AUC 0.86) and outperformed less experienced radiologists (AUC 0.71).
The multimodal CNNs, integrating ADC and T2WI data, demonstrated high accuracy for PCa detection in patients with gray-zone PSA levels, comparable to senior radiologists. This model can aid clinical decision-making, assist in challenging cases, and help reduce unnecessary biopsies.
This retrospective multi-center study initially reviewed data from 1778 patients with suspected prostate lesions who presented at multiple hospitals between December 2015 and October 2023. The DL algorithm was trained on T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC) values. The performance of the algorithm was evaluated using the area under the receiver operating characteristic (ROC) curve and compared to the diagnostic performance of radiologists with varying levels of expertise.
The final study cohort comprised 274 patients, who were divided into training, validation, and test sets in a 6:2:2 ratio, corresponding to 162, 56, and 56 patients, respectively. The mean age of the patients was 71.8 years, and the cohort included 90 patients with prostate cancer and 184 without. The multi-attribute Convolutional Neural Networks (CNNs) model achieved an area under the curve (AUC) of 0.84, surpassing individual ADC and T2WI models. It also outperformed a transformer-based model (AUC 0.79). When compared to radiologists, the DL model performed similarly to experienced radiologists (AUC 0.86) and outperformed less experienced radiologists (AUC 0.71).
The multimodal CNNs, integrating ADC and T2WI data, demonstrated high accuracy for PCa detection in patients with gray-zone PSA levels, comparable to senior radiologists. This model can aid clinical decision-making, assist in challenging cases, and help reduce unnecessary biopsies.
Authors
Wang Wang, Zhao Zhao, Wang Wang, Qi Qi, Zhang Zhang, Xi Xi, Wang Wang, Yan Yan, Lei Lei
View on Pubmed