AI-based modeling of treatment decisions in benign prostatic hyperplasia: a transformer-based comparative study

Area of Science:

• Urology
• Artificial Intelligence
• Medical Informatics

Background:

• Benign Prostatic Hyperplasia (BPH) significantly impacts aging men's quality of life due to urinary symptoms.
• Predicting the optimal treatment path (surgery vs. medical therapy) for BPH is crucial for patient outcomes.

Purpose of the Study:

• To evaluate the efficacy of advanced AI models, including Large Language Models (LLMs) and deep learning, in predicting BPH management strategies.
• To compare the performance of GEMMA, GPT, Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) models.

Main Methods:

• Utilized a dataset of 883 BPH patient cases from Jordan University Hospital (JUH).
• Included 15 clinical attributes such as Prostate-Specific Antigen (PSA) levels, prostate size, and treatment history.
• Trained and tested five predictive models: GEMMA, GPT, RNN, CNN, and LSTM.

Main Results:

• The GEMMA model achieved the highest prediction accuracy (92%) and ROC AUC score (0.94).
• The GPT model demonstrated strong performance with 91% accuracy.
• Long Short-Term Memory (LSTM) outperformed other deep learning models (CNN, RNN) in capturing sequential data dependencies.

Conclusions:

• AI models, particularly GEMMA and GPT, show high potential for predicting BPH treatment decisions.
• Key clinical features like PSA levels and prostate size are critical for accurate BPH management prediction.
• These predictive capabilities can aid clinicians in optimizing treatment strategies for BPH patients.