Contrast-Enhanced Ultrasound-Magnetic Resonance Imaging Radiomics Based Model for Predicting the Biochemical Recurrence of Prostate Cancer: A Feasibility Study

Abstract

Objective. This study was aimed at developing a model for predicting postoperative biochemical recurrence of prostate cancer (PCa) using clinical data-CEUS-MRI radiomics and at verifying its clinical effectiveness. Methods. The clinical imaging data of 159 patients pathologically confirmed with PCa and who underwent radical prostatectomy in Xiangyang No. 1 People’s Hospital and Jiangsu Hospital of Chinese Medicine from March 2016 to December 2020 were retrospectively analyzed. According to the 2-5-year follow-up results, the patients were divided into the biochemical recurrence (BCR) group ( $n=59$ ) and the control group ( $n=100$ ). The training set and test set were established in the proportion of 7/3; 4 prediction models were established based on the clinical imaging data. In training set, the area under the curve (AUC) and decision curve analysis (DCA) by R was conducted to compare the efficiency of 4 prediction models, and then, external validation was performed using the test set. Finally, a nomogram tool for predicting BCR was developed. Results. Univariate regression analysis confirmed that the SmallAreaHighGrayLevelEmphasis, RunVariance, Contrast, tumor diameter, clinical T stage, lymph node metastasis, distant metastasis, Gleason score, preoperative PSA, treatment method, CEUS-peak intensity (PI), time to peak (TTP), arrival time (AT), and elastography grade were the influencing factors for predicting BCR. In the training set, the AUC of combinatorial model demonstrated the highest efficiency in predicting BCR [AUC: 0.914 (OR 0.0305, 95% CI: 0.854-0.974)] vs. the general clinical data model, the CEUS model, and the MRI radiomics model. The DCA confirmed the largest net benefits of the combinatorial model. The test set validation gave consistent results. The nomogram tool has been well applied clinically. Conclusion. The previous clinical and imaging data alone did not perform well for predicting BCR. Our combinatorial model firstly using clinical data-CEUS-MRI radiomics provided an opportunity for clinical screening of BCR and help improve its prognosis.