Development of a Deep Learning-Based Predictive Model for Improvement after Holmium Laser Enucleation of the Prostate According to Detrusor Contractility

Abstract

Purpose Predicting improvements in voiding symptoms following deobstructive surgery for male lower urinary tract symptoms/benign prostatic hyperplasia (LUTS/BPH) is challenging when detrusor contractility is impaired. This study aimed to develop an artificial intelligence model that predicts symptom improvement after holmium laser enucleation of the prostate (HoLEP), focusing on changes in maximum flow rate (MFR) and voiding efficiency (VE) 1-month postsurgery. Methods We reviewed 1,933 patients who underwent HoLEP at Samsung Medical Center from July 2008 to January 2024. The study employed a deep neural network (DNN) for multiclass classification to predict changes in MFR and VE, each divided into 3 categories. For comparison, additional machine learning (ML) models such as extreme gradient boosting, random forest classification, and support vector machine were utilized. To address class imbalance, we applied the least squares method and multitask learning. Results A total of 1,142 patients with complete data were included in the study, with 992 allocated for model training and 150 for external validation. In predicting MFR, the DNN achieved a microaverage area under the receiver operating characteristic curve (AUC) of 0.884±0.006, sensitivity of 0.783±0.020, and specificity of 0.891±0.010. For VE prediction, the microaverage AUC was 0.817±0.007, with sensitivity and specificity values of 0.660±0.014 and 0.830±0.007, respectively. These results indicate that the DNN's predictive performance was superior to that of other ML models. Conclusions The DNN model provides detailed and accurate predictions for recovery after HoLEP, providing valuable insights for clinicians managing patients with LUTS/BPH.

Keywords

• artificial intelligence
• benign prostatic hyperplasia
• lower urinary tract symptoms
• urinary bladder
• underactive