AI对肺磨玻璃结节筛查及定性的临床应用研究

A clinical study of AI on the detection and classification of pulmonary ground glass nodules

目的:探讨人工智能(AI)对肺磨玻璃结节(GGN)筛查及定性的临床应用价值。方法:搜集行胸部CT平扫的200例患者(共1230个GGN),比较A组(住院医师)、B组(AI)、C组(住院医师结合AI)对诊断GGN的敏感度、误判率、漏诊率、阳性预测值和平均诊断时间。将其中经手术病理证实的137例GGN患者按其病理结果分为良性组(54例)、恶性组(83例),比较两组间AI量化参数的差异,对有统计学差异的参数行ROC曲线分析,再以病理结果为因变量,各指标为自变量行Logistic回归分析。结果:B组误诊率高于A、C两组,阳性预测值小于A、C两组;A组漏诊率高于B、C两组,敏感度小于B、C两组,差异均有统计学意义(P值均<0.05)。GGN良、恶性结节的长径、最大面积、体积、平均CT值、最大CT值和恶性概率差异均有统计学意义(P值均<0.05),对各参数行ROC曲线分析,曲线下面积(AUC)均大于0.7,Logistic回归分析显示长径和最大面积是GGN恶变的独立危险因素。结论:AI协助医生阅片可明显提高肺GGN检出敏感度,并可降低误诊率和漏诊率,同时对GGN的良恶性预判具有一定参考价值。

Objective: To explore the clinical application value of artificial intelligence (AI) in the detection and classification of pulmonary ground glass nodules (GGNs). Methods: A total of 1230 GGNs in 200 patients with chest CT plain scan were collected.The detection sensitivity,false positive,rage of missed diagnosis,positive predictive value and average diagnosis time were compared in the following groups:group A (resident),group B (AI) and group C (resident combined with AI).137 GGNs confirmed by surgery and pathology were divided into benign group (54 cases) and malignant group (83 cases) according to their pathological reports.The differences of AI quantification parameters between the two groups were compared,ROC curve analysis was performed for the parameters with statistical differences.Then,logistic regression analysis was performed considering the pathological results as the dependent variables and the parameters as independent variables. Results: The false positive of group B was higher than that of group A and C,while the positive predictive value was lower than that of group A and C.The rate of missed diagnosis of group A was higher than that of group B and C,while the detection sensitivity was lower than that of group B and C ( P <0.05).There were statistically significant differences in longest diameter,maximum area,volume,average CT value,maximum CT value and malignant rate between benign and malignant groups ( P <0.05).All parameters obtained AUCs greater than 0.7 after ROC curve analysis and the longest diameter and maximum area were identified as independent risk factors for malignant transformation of GGN after logistic regression analysis. Conclusion: AI-aided CT reviewing can significantly improve the detection sensitivity and reduce the false positive,rage of missed diagnosis in terms of pulmonary GGNs.Meanwhile,it has potential reference value for the prediction of benign and malignant GGN.