Enhancing Physiological Consistency in Dynamic PET/CT for Hepatocellular Carcinoma through Prior Information and NSGA-III

• Published in: International Conference on Information and Computer Technologies (Online) pp. 377 - 381
• Main Authors: Xiong, Yiwei, Li, Siming, He, Jianfeng, Wang, Shaobo
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.03.2024
• Subjects: Heuristic algorithms; kinetic modeling; NSGA-III; Parameter estimation; PET/CT; Pharmacokinetics; Physiology; Positron emission tomography; prior information; Visualization
• ISSN: 2769-4542
• DOI: 10.1109/ICICT62343.2024.00067

Dynamic positron emission tomography/computed tomography (PET/CT) is a promising medical imaging technique that allows the visualization of biological information through the activity of tracers. Analyzing the time-activity curves (TACs) of PET using pharmacokinetic models enables the extraction of quantifiable physiological process information relevant to hepatocellular carcinoma (HCC). The difficulty in the research lies in ensuring that estimated modeling parameters translate into effective diagnostic outcomes. In this study, the nondominated sorting genetic algorithm III (NSGA-III) is employed for multi-objective optimization, enhancing physiological consistency in parameter estimation by introducing supplementary physiological objectives alongside the fitting error objective. Utilizing prior information on kinetic parameter in each subpopulation, the prior-based multi-population NSGA-III (p-MPNSGA-III) algorithm was proposed for parameter estimation in pharmacokinetic models. Parameters estimated from PET TACs of 24 real HCC patients undergo statistical difference and receiver operating characteristic (ROC) analyses, using P-values and areas under the curve (AUC) values to evaluate diagnostic performance between normal and tumor groups. Compared to the standard Genetic Algorithm (GA), the p-MPNSGA-III exhibits statistically significant differences in K1 (P=0.006), k2 (P=0.002), k3 (P=0.024), fa (P<0.001), and vb (P=0.039) and correct physiological characteristics in all parameters. In ROC analyses, five parameters from the proposed method achieved higher AUC values compared to the standard GA. The results demonstrate that p-MPNSGA-III effectively enhances the physiological consistency of parameter estimation, leading to excellent diagnostic performance.