Extended Abstract: Classical Machine Learning vs. Large Language Models for Prostate Cancer Detection
Conference
Omar, I, Elshinawy, M, Arafa, MA et al. (2027). Extended Abstract: Classical Machine Learning vs. Large Language Models for Prostate Cancer Detection
. Communications in Computer and Information Science, 2971 CCIS 473-476. 10.1007/978-3-032-28463-1_38
Omar, I, Elshinawy, M, Arafa, MA et al. (2027). Extended Abstract: Classical Machine Learning vs. Large Language Models for Prostate Cancer Detection
. Communications in Computer and Information Science, 2971 CCIS 473-476. 10.1007/978-3-032-28463-1_38
This study builds on earlier work using classic machine learning (ML) for prostate cancer (PCa) detection by employing Large Language Models (LLMs) to construct and evaluate prognostic prediction models. We compare five LLMs against classic ML algorithms using data from 751 patients from two leading hospitals in Riyadh, Saudi Arabia, who underwent prostate biopsy. Features including age, prostate volume, PSA, BMI, mpMRI score, and lesion characteristics were used. ClaudeAI and Gemini demonstrated the highest performance among the LLMs. We evaluated different preprocessing approaches, finding that performance varied significantly. Using F1-score and AUC, we found that providing a training sample via In-Context Learning (ICL) yielded the best LLM results, with ClaudeAI and Gemini achieving the highest accuracy. ClaudeAI showed excellent prediction capability compared to other LLMs.
