Kang Jeonghyun, Lee Jae-Hoon, Lee Hye Sun, Jeon Tae Joo, Ryu Young Hoon
Department of Surgery, Gangnam Severance Hospital, Yonsei University College of Medicine, Seoul, Republic of Korea.
Department of Nuclear Medicine, Gangnam Severance Hospital, Yonsei University College of Medicine, Seoul, Republic of Korea.
Quant Imaging Med Surg. 2025 Aug 1;15(8):7524-7536. doi: 10.21037/qims-2025-117. Epub 2025 Jul 25.
Metabolic syndrome (MetS) is a complex health concern and the incidence of MetS is rising, even among the general population, necessitating effective identification and management strategies. This study aimed to determine if a predictive model using variables from fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT) and machine learning (ML) could enhance the prediction of MetS.
We retrospectively reviewed the medical records of 1,250 adults who underwent FDG PET/CT for cancer screening between 2014 and 2020. MetS was diagnosed according to the National Cholesterol Education Program Adult Treatment Panel III criteria. The study analyzed standardized uptake values (SUVs), area, and Hounsfield unit (HU) of various body organs from FDG PET/CT and developed a multivariable predictive model for MetS integrating FDG PET/CT variables using least absolute shrinkage and selection operator (LASSO) regression. The performance of a predictive model was assessed using the area under the receiver operating characteristic curve (AUC).
The study population comprised 720 men and 530 women with a median age of 54 years, and MetS was present in 26.3% of the subjects. The LASSO regression identified the area of visceral adipose tissue (VAT), mean HU of VAT, mean SUV of VAT, mean HU of skeletal muscle, mean SUV of blood pool, and body mass index as meaningful variables. Our multivariable LASSO model effectively predicted MetS with similar performance in both training and test sets (AUC, 0.792 and 0.828, respectively; P=0.173) and demonstrated superior predictive performance compared to univariable models in the test set (AUC, 0.828)-body mass index (0.794; P=0.017), the area of VAT (0.788; P<0.001), and the mean HU of VAT (0.777; P<0.001).
Our findings established the potential of FDG PET/CT, enhanced with ML, in predicting MetS.
代谢综合征（MetS）是一个复杂的健康问题，其发病率正在上升，即使在普通人群中也是如此，因此需要有效的识别和管理策略。本研究旨在确定使用来自氟-18氟脱氧葡萄糖正电子发射断层扫描/计算机断层扫描（FDG PET/CT）的变量和机器学习（ML）的预测模型是否能增强对MetS的预测。
我们回顾性分析了2014年至2020年间接受FDG PET/CT癌症筛查的1250名成年人的病历。根据美国国家胆固醇教育计划成人治疗小组第三次报告的标准诊断MetS。该研究分析了FDG PET/CT中各身体器官的标准化摄取值（SUV）、面积和亨氏单位（HU），并使用最小绝对收缩和选择算子（LASSO）回归开发了一个整合FDG PET/CT变量的MetS多变量预测模型。使用受试者工作特征曲线下面积（AUC）评估预测模型的性能。
研究人群包括720名男性和530名女性，中位年龄为54岁，26.3%的受试者患有MetS。LASSO回归确定内脏脂肪组织（VAT）面积、VAT平均HU、VAT平均SUV、骨骼肌平均HU、血池平均SUV和体重指数为有意义的变量。我们的多变量LASSO模型在训练集和测试集中均能有效预测MetS，性能相似（AUC分别为0.792和0.828；P=0.173），并且在测试集中与单变量模型相比显示出更好的预测性能（AUC为0.828）——体重指数（0.794；P=0.017）、VAT面积（0.788；P<0.001）和VAT平均HU（0.777；P<0.001）。
我们的研究结果证实了结合ML的FDG PET/CT在预测MetS方面的潜力。
