Can dynamic contrast-enhanced MR imaging based on radiomics improve the diagnostic efficiency of clinically significant prostate cancer?

Abstract

Aims/Background Prostate cancer stands out as one of the most prevalent malignant tumours among males. The non-invasive identification of clinically significant prostate cancer via magnetic resonance imaging plays a critical role in circumventing unnecessary biopsies and determining suitable treatment strategies for patients. Our study aimed to evaluate the potential improvement in predictive accuracy for clinically significant prostate cancer by incorporating perfusion data obtained from dynamic contrast-enhanced magnetic resonance imaging acquisition protocols into multiparametric magnetic resonance imaging parameters. Methods Radiomics extracted from perfusion parameters (Ktrans, Kep, Ve) of dynamic contrast-enhanced magnetic resonance imaging were analysed in patients suspected of prostate cancer who underwent 3T multiparametric magnetic resonance imaging between January 2017 and June 2023 in this retrospective study. The pathological findings obtained from biopsy or therapy were categorised into groups based on the Gleason sum score as either clinically significant prostate cancer (Gleason sum score > 7) or non-clinically significant prostate cancer (Gleason sum score ≤ 6). Diagnostic models were constructed using logistic regression analysis, incorporating prostate imaging reporting and data system V2.1 scores and clinical data, with or without radiomics extracted from dynamic contrast-enhanced. The area under curve (AUC) values were compared using the DeLong test. Results Overall, 214 men (clinically significant prostate cancer [n=78] and non-clinically significant prostate cancer [n=136]) were included. The clinical-prostate imaging reporting and data system model demonstrated an AUC of 0.89 (95% confidence interval: 0.84–0.95) in the training cohort and 0.91 (95% confidence interval: 0.84–0.98) in the test cohort. For the clinical-prostate imaging reporting and data system-radscore model, the AUC values were 0.97 (95% confidence interval: 0.95–0.99) for Ktrans, 0.98 (95% confidence interval: 0.96–1.00) for Ve, and 0.96 (95% confidence interval: 0.93–0.98) for Kep in the training cohort, and 0.97 (95% confidence interval: 0.94–1.00) for Ktrans, 0.95 (95% confidence interval: 0.91–1.00) for Ve, and 0.97 (95% confidence interval: 0.94–1.00) for Kep in the test cohort. Radiomics based on perfusion parameters exhibited good diagnostic performance in predicting clinically significant prostate cancer. The clinical-prostate imaging reporting and data system-radscore model demonstrated superior diagnostic capability compared to perfusion-based radiomics or clinical-prostate imaging reporting and data system models alone. Conclusion The application of radiomics, which involves extracting perfusion parameters from dynamic contrast-enhanced imaging, has the potential to enhance diagnostic accuracy for clinically significant prostate cancer.