摘要
目的:评估机器学习(ML)的性能,将临床参数与CT冠状动脉钙化(CAC)评分和自动心外膜脂肪组织(EAT)量化相结合,以预测无症状受试者的心肌梗死(MI)和心源性死亡的长期风险。方法:选取2013年1月至2015年10月于解放军总医院第五医学中心进行体检的1058名存在心血管危险因素且未发生冠心病症状的受试者,在进行CAC评分后进行长期随访。使用全自动深度学习方法量化EAT体积和密度。采用受试者工作特征(ROC)曲线分析临床资料、动脉粥样硬化性心血管疾病风险评分、CAC评分和自动EAT测量来训练ML极端梯度增强,并使用重复的10倍交叉验证进行模型验证。结果:在8年随访期间,1058名受试者中61例发生了MI和(或)心源性死亡事件。ML的ROC曲线下面积(AUC)明显高于动脉粥样硬化性心血管疾病(ASCVD)风险和CAC评分预测事件(ML:0.82;ASCVD:0.77;CAC:0.77)。根据ASCVD和CAC、EAT的ML比仅具有临床变量的ML更能预测MI和心脏死亡[AUC为0.82(95%CI:77~87)vs.0.78(95%CI:0.72~0.84),P=0.02]。ML高分数受试者生存率随时间增长下降程度较大,因此ML得分较高的受试者更有可能经历事件。结论:整合临床和定量影像学变量的ML可以为心血管危险因素患者提供长期风险预测。
Objective:To assess the performance of machine learning(ML),and integrate the clinical parameters with coronary artery calcium(CAC)score of computed tomography(CT)and quantification of automated epicardial adipose tissue(EAT),so as to predict the long-term risk of myocardial infarction(MI)and cardiogenic death in asymptomatic patients.Methods:A total of 1058 subjects with cardiovascular risk factors and without symptoms of coronary heart disease who underwent physical examination at the Fifth Medical Center of Chinese PLA General Hospital from January 2013 to October 2015 were selected as this study subjects.A long-term follow-up was conducted on them after CAC score.EAT volume and density were quantified using a fully automated deep learning method.ML extreme gradient boosting was trained by using clinical data,risk score of atherosclerotic cardiovascular disease,CAC score and automated EAT measure,and the repeated 10-fold cross validation was used to verify the model.Results:During the 8-year follow-up period,61 cases of 1058 subjects occurred events of MI and(or)cardiac death.The area under curve(AUC)value of ML was significantly higher than that of the atherosclerotic cardiovascular disease(ASCVD)risk and the predicting events of CAC score(ML:0.82,ASCVD:0.77,CAC:0.77).Compared with ML with only clinical variable,machine learning based on ASCVD,CAC and EAT had more predictive ability for MI and cardiac death[AUC 0.82(95%CI:77-87)vs.0.78(95%CI:0.72-0.84),P=0.02].The survival rate of subjects with high ML scores had a greater decline degree with the increasing of time,therefore,the subjects with higher ML scores were more likely to experience events.Conclusion:ML,which integrated clinical and quantitative imaging variables,can provide long-term risk prediction for patients with cardiovascular risk factors.
作者
苑文雯
高旭东
赵静
李筱涵
刘佳
皋月娟
庞君丽
赵利利
李伯安
Yuan Wenwen;Gao Xudong;Zhao Jing;Li Xiaohan;Liu Jia;Gao Yuejuan;Pang Junli;Zhao Lili;Li Boan(Department of Clinical Laboratory,The Fifth Medical Center of Chinese PLA General Hospital,Beijing 100039,China;Department of Hepatology,The Fifth Medical Center of Chinese PLA General Hospital,Beijing 100039,China;Department of Ultrasound Diagnosis,The Fifth Medical Center of Chinese PLA General Hospital,Beijing 100039,China)
出处
《中国医学装备》
2024年第6期56-61,共6页
China Medical Equipment
基金
北京市自然科学基金(7222172)。
关键词
CT冠状动脉钙化评分
心外膜脂肪组织量化
心肌梗死(MI)
机器学习(ML)
Computed tomography(CT)score of coronary artery calcification
Quantization of epicardial adipose tissue
Myocardial infarction(MI)
Machine learning(ML)
