基于深度学习的FFRCT在可疑冠心病病人中的应用

Application of deep learning computed tomographic angiography-based fractional flow reserve in patients with suspected coronary artery disease

目的探讨基于深度学习的CT血流储备分数(FFR_(CT))在可疑冠心病病人中应用的可行性,分析缺血性病变(FFR_(CT)≤0.80)的预测因素及对治疗决策的影响。方法回顾性纳入因疑似冠心病行冠状动脉CT血管成像(CCTA)的病人292例,其中男187例,女105例,平均年龄(65.8±10.3)岁。利用CCTA影像将狭窄程度分为轻度(≥25%且<50%)、中度(≥50%且<70%)和重度(≥70%且<99%)。采用基于深度学习的FFR_(CT)软件对病人的CCTA数据进行测量。根据FFR_(CT)数值范围将病人分为阳性组(FFR_(CT)≤0.80,102例)和阴性组(FFR_(CT)>0.80,190例)。2组病人的一般资料、CCTA上的血管特征及血运重建,以及基于FFR_(CT)与CCTA制定的治疗策略的比较采用Mann-Whitney U检验、t检验及卡方检验。采用Logistic回归分析FFR_(CT)≤0.80的独立预测因素。结果阳性组病人的年龄更大,男性更多,高血压、糖尿病和吸烟的比例均高于阴性组(均P<0.05)。阳性组较阴性组病人更多的表现为中重度狭窄(分别为80.4%和28.4%),更多的病人行血运重建术(分别为56.8%和11.1%),均P<0.05。74例病人(25.3%)基于FFR_(CT)的结果治疗决策发生改变。多因素Logsitic回归分析显示,高血压(OR=2.245)、糖尿病(OR=2.238)及中重度狭窄(OR=8.837)是FFR_(CT)≤0.80的独立预测因素(均P<0.05)。结论基于深度学习的FFR_(CT)技术在可疑冠心病病人中的应用是可行的,高血压、糖尿病及中重度狭窄是FFR_(CT)≤0.80的独立预测因素,FFR_(CT)可能影响病人的治疗决策。

Objective To explore the feasibility of deep learning-based CT flow reserve fraction(FFR_(CT))in patients with suspected coronary artery disease,and to analyze the predictors of ischemic disease(FFR_(CT)≤0.80)and its influence on therapeutic decision-making Methods A total of 292 patients with suspected coronary artery disease who underwent coronary computed tomography angiography(CCTA)were retrospectively enrolled,including 187 males and 105 females,with an average age of 65.8±10.3 years.The stenosis on CCTA images were analyzed,which was classified into 3 categories:mild(≥25%-50%),moderate(≥50%-70%),and severe(≥70%-99%).CCTA data were measured using FFR_(CT) software based on deep learning.The patients were divided into positive group(n=102,FFR_(CT)≤0.80)and negative group(n=190,FFR_(CT)≤0.80).Mann-Whitney U test,t test and Chi-square test were used to compare the baseline information,artery features on CCTA,revascularization,and treatment strategies based on FFR_(CT) and CCTA between the two groups.Logistic regression model was used to analyze the independent predictors of FFR_(CT)≤0.80.Results Compared with the negative FFR_(CT) group,the positive FFR_(CT) group was older,more likely to be men,the prevalence of hypertension,diabetes,and smoking were higher(all P<0.05).More patients in the positive group had moderate to severe stenosis(80.4%vs.28.4%,respectively),and more patients underwent revascularization(56.8%vs.11.1%)(all P<0.05).Seventy-four patients(25.3%)changed treatment decisions based on FFR_(CT) results.Logistic regression analysis showed that hypertension(OR=2.245),diabetes(OR=2.238),and stenosis degree(OR=8.837)were independent predictors of FFR_(CT)≤0.80(all P<0.05).Conclusion FFR_(CT) based on deep learning is feasible in patients with suspected coronary heart disease.Hypertension,diabetes,and moderate to severe stenosis are independent predictors of FFR_(CT)≤0.80,and FFR_(CT) may influence treatment strategies.