CT coronary perivascular fat attenuation combined with machine learning algorithms for diagnosis of myocardial ischemia in coronary heart disease
10.20039/j.cnki.1007-3949.2024.06.008
- VernacularTitle:CT冠状动脉周围脂肪衰减结合机器学习算法诊断冠心病心肌缺血
- Author:
Yige LU
1
,
2
,
3
;
Wei HE
;
Hongyan LIN
;
Furong HE
;
Hanbo ZHANG
;
Yao TAN
;
Hongming ZHU
Author Information
1. 复旦大学附属中山医院血管外科
2. 复旦大学血管外科研究所
3. 国家放射与治疗临床医学研究中心
- Keywords:
coronary heart disease;
myocardial ischemia;
fat attenuation index;
fractional flow reserve
- From:
Chinese Journal of Arteriosclerosis
2024;32(6):514-520
- CountryChina
- Language:Chinese
-
Abstract:
Aim To explore the feasibility of using machine learning algorithms combined with coronary computed tomography(CT)derived perivascular fat attenuation index(FAI)and plaque information to evaluate myocardial ischemia in stable coronary heart disease patients.Methods A retrospective analysis was conducted on the clinical and imaging data of patients who underwent preoperative coronary CT angiography(CCTA),invasive coronary angiography(ICA),and flow reserve fraction(FFR)measurements at Zhongshan Hospital Affiliated to Fudan University from April 2019 to October 2021.206 patients with stable coronary heart disease were selected.The semi-automatic plaque analysis software was used for quantification of plaque and lumen parameters and perivascular FAI measurement,with man-ual delineation of a 40 mm segment of the coronary artery starting 10 mm from the ostium for perivascular FAI measure-ment.Differences in plaque characteristics,perivascular FAI,and coronary perivascular FAI between stable coronary heart disease patients with FFR≤0.8 and FFR＞0.8 were compared.The diagnostic performance of combining perivascu-lar FAI,coronary perivascular FAI,and plaque features using machine learning algorithms for myocardial ischemia in stable coronary heart disease patients was evaluated through ROC curves.Results 206 stable coronary heart disease patients were divided into FFR≤0.8 group(50 cases)and FFR＞0.8 group(156 cases).The mean periplaque FAI of patients with FFR≤0.8 was-69.28±5.65 HU,significantly higher than that of patients with FFR＞0.8 at-80.10±7.75 HU(P＜0.001).Further analysis was conducted using machine learning models,including XGBoost,random forest,and Logistic regression models,all of which had an accuracy rate of over 0.8 in diagnosing myocardial ischemia.Among them,the XGBoost model performed the best with an accuracy of 0.903,an F1 value of 0.774,and an AUC of 0.931,in-dicating its high effectiveness in diagnosing myocardial ischemia.Conclusion The combination of FAI and machine learning algorithm XGBoost model is a new method for diagnosing myocardial ischemia,which has better diagnostic value in evaluating myocardial ischemia in stable coronary heart disease patients.
