Optimization of image quality of simulated multiple detectors computed tomography acquisition parameters using machine learning and Catphan phantom

The study successfully employed Monte Carlo (MC) simulation and a Machine Learning (ML) approach using a Random Forest Regression (RFR) model to develop optimized Multi-Detector CT (MDCT) protocols that significantly reduce radiation dose while maintaining diagnostic image quality. The MC engine accurately modeled X-ray spectra, and the RFR model demonstrated high predictive power for key metrics, achieving R2 scores of 0.97 for CTDIvol and over 0.92 for image quality metrics (Noise, CNR). Through multi-objective optimization guided by the RFR, the final protocol (Optimization-3) was found on the Pareto front, achieving a notable 35% dose reduction (from 15.5 mGy to 9.9 mGy) compared to the standard protocol. Furthermore, using Iterative Reconstruction (IR) contributed a 22–35% noise reduction. Quantitative analysis on the Catphan phantom confirmed that the optimized settings preserved or enhanced image quality metrics like CNR (improving from 1.8 to 1.9) and Line Spread Function (LSF), validating the ML-driven approach's effectiveness in balancing patient safety with clinical requirements.