Abstract
Benign Prostatic Hyperplasia (BPH) is a common condition among aging men that often causes significant urinary symptoms, impacting their quality of life. This study employs advanced LLMs and deep learning models to predict whether BPH patients were managed with TURP or continued medical therapy using historical clinical data. We utilized a dataset of 883 patient cases from Jordan University Hospital (JUH), comprising 15 clinical attributes, including PSA levels, prostate size, and treatment history. Five models were tested: GEMMA, GPT, and three deep learning models-Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM). The GEMMA model achieved the highest performance, with an accuracy of 92% and an ROC AUC score of 0.94. The GPT model was a close second with 91% accuracy, demonstrating its robustness in handling multimodal data. Deep learning models showed promising results, with LSTM outperforming CNN and RNN because of its ability to capture sequential dependencies. The findings emphasize the importance of choosing relevant features, like PSA levels and prostate size, for better prediction accuracy, which significantly impacts the decision to either keep medication or move to surgery.