A model was introduced to select radiomic features from T2-weighted MRI images to classify patients based on continence a year post-procedure. With a cohort of 97 patients, our chosen features displayed impressive sensitivity, specificity, accuracy, and area under the ROC curve metrics.
The study rigorously assessed several classification algorithms such as Decision Trees, Naive Bayes, Support Vector Machines, Nearest Neighbor, XGBoost, and Neural Networks. The XGBoost algorithm, a dynamic machine-learning technique, emerged superior. This gradient-boosting framework focuses on sequentially refining models to rectify preceding errors, ensuring accuracy and resilience.
While our study aligns with prior research linking age with postoperative continence rates, it differs in its findings concerning the LA muscle. Unlike earlier studies that emphasized LA muscle thickness as a determinant of continence, our study spotlighted LA muscle textural features as pivotal. Specifically, the GLCM features, which define the statistical relationship of gray levels between neighboring pixels in images, have been deemed significant in deciphering underlying pathomorphological patterns. The conspicuous association between certain LA muscle textural features and continence emphasizes radiomics' potential to unearth deeper insights about post-RARP urinary continence. Our findings challenge the traditional reliance on morphological features, suggesting that more intricate textural metrics might reveal essential pathophysiological changes.
Significant correlations between features suggest overlapping or complementary information. Such correlations highlight potential multicollinearity, a phenomenon where multiple features provide similar insights. In our model, such correlations played a pivotal role, steering the selection of significant features while sidelining redundant ones. Recent research identified a distinct radiomic signature linking muscle heterogeneity detected by ultrasound radiomic features to various age-related ailments. This theory suggests that age might bring about textural or heterogeneity alterations in the LA muscle, mirroring underlying tissue changes at the functional level.
Our findings could revolutionize patient care by aiding physicians in pre-surgical identification of potential incontinence risks. Such insights could refine patient counseling, optimize surgical candidate selection, and promote alternative treatments when required. However, our study has limitations. With a small, single-institution sample size and a retrospective design, plus other unaccounted factors like pre-existing urinary issues and nerve-sparing surgery predominance, our findings require cautious interpretation. Although our approach awaits external validation, the initial outcomes are promising.
Written by: Mohammed Shahait,1 Ruben Usamentiaga,2 Yubing Tong,3 Alex Sandberg,4 David I Lee,5 Jayaram K Udupa,3 and Drew A Torigian3
- Department of Surgery, Clemenceau Medical Center, Dubai, United Arab Emirates.
- Department of Computer Science and Engineering, University of Oviedo, Gijon, Spain.
- Medical Image Processing Group, Department of Radiology, University of Pennsylvania, Philadelphia, PA
- Temple Medical School, Temple University, Philadelphia, PA
- Department of Urology, University of California Irvine, Irvine, CA