Design and evaluation of clinical decision support system to determine the clinical significance of prostate cancer based on machine learning techniques

Abstract

Introduction: Prostate cancer diagnosis and management rely heavily on accurate assessment of clinical significance, which is often determined through a combination of medical imaging and laboratory tests. Traditional methods can be complex and inconsistent. This study aimed to develop and evaluate a Clinical Decision Support System (CDSS) that integrates artificial intelligence to enhance the accuracy of prostate cancer diagnosis and treatment recommendations. Objectives: The primary objective was to design and evaluate a CDSS that leverages medical images and clinical data to assess the clinical significance of prostate cancer. Specifically, the study sought to integrate multi-parametric MRI (mpMRI) data with clinical markers to improve diagnostic accuracy and treatment planning. Method: The study was conducted in three stages. First, data collection and preparation involved obtaining ethical approval and gathering patient information, including PSA levels, mpMRI results, and biopsy outcomes. The dataset comprised 356 patients with detailed clinical and imaging data. The second stage focused on system design and development, where an inference engine was created using AI modules. The ResNet50 model classified MRI images into PI-RADS scores, and a Gradient Boosting Classifier (XGBoost) predicted clinical significance based on these scores combined with clinical data. The knowledge base was constructed from European Association of Urology (EAU) guidelines, and a user-friendly interface was developed using Python’s Tkinter library. The third stage evaluated the CDSS by comparing its recommendations with actual physician decisions for 192 patients. Results: CDSS recommendations, which consisted of clinical data of 356 patients, two artificial intelligence modules in the inference engine part and 45 rules in the knowledge base part, were aligned with the doctor's decisions in 85.4% of cases. Also, the creation of the user interface of the system enabled simple data entry and production of treatment recommendations. Key words: prostate cancer, CDSS, Clinical Decision Support System, Gleason Score, multiparametric MRI