Prediction of extracapsular extension and seminal vesicle invasion in prostate cancer using preoperative MRI images combined with deep learning algorithms.

Abstract

e17139 Background: Extracapsular extension (ECE) and seminal vesicle invasion (SVI) are critical factors in determining nerve-sparing decisions during radical prostatectomy for prostate cancer. This study aims to predict ECE and SVI in prostate cancer patients using preoperative MRI images combined with deep learning algorithms. Methods: We retrospectively collected data from patients who underwent radical prostatectomy at our center between 2018 and 2022. After screening, 354 patients with complete clinicopathological information, postoperative whole-mount pathological slides, and preoperative 3.0T contrast-enhanced MRI images were included in the study. These patients were divided into training and validation sets in a 2:1 ratio. Using whole-mount pathological slides as the gold standard, two senior radiologists delineated the lesion areas on preoperative 3.0T contrast-enhanced MRI images. A predictive model was then constructed based on the delineated MRI images using the novel deep learning algorithm NAFNet. Results: A deep learning model was developed using the training cohort to predict the probability of ECE and SVI. The probability values generated by the model, along with other clinicopathological indicators, were incorporated into a multivariate regression model. Finally, a deep learning-integrated model (DL-nomogram) was constructed by combining clinical T-stage and preoperative biopsy pathology grading. The results in the validation cohort demonstrated that the integrated model had good predictive performance for ECE and SVI (AUC: 0.852, 95% CI: 0.798–0.910). Conclusions: The integrated model, developed using the novel deep learning algorithm NAFNet combined with traditional clinicopathological indicators, can predict ECE and SVI in prostate cancer patients based on preoperative MRI images. This provides a potential new tool for individualized treatment planning in prostate cancer patients.