基于AI评估磨玻璃样肺腺癌浸润性及病理亚型的研究  

Evaluation of Invasiveness and Pathological Subtypes of Ground glass Lung Adenocarcinoma Based on AI

作　　者：汪琼 朱建国[1] 李海歌[1] 曹波[1] 唐继来[1] WANG Qiong;ZHU Jian-guo;LI Hai-ge;CAO Bo;TANG Ji-lai(Department of Radiology,The Second Affiliated Hospital of Nanjing Medical University,Nanjing 210003,Jiangsu Province,China)

机构地区：[1]南京医科大学第二附属医院放射科,江苏南京210003

出　　处：《罕少疾病杂志》2024年第9期36-39,共4页Journal of Rare and Uncommon Diseases

摘　　要：目的 基于AI(人工智能)评估高分辨计算机断层扫描(HRCT)上表现为磨玻璃样肺腺癌的浸润性及病理亚型。方法 收集经病理证实为肺腺癌的GGN共205例,其中男79例,女126例,平均年龄59.87±11.49岁。根据病理浸润性可分为微浸润性腺癌(MIA)和浸润性腺癌(IA);IA又根据病理亚型分为低危组(含有乳头、贴壁或腺泡型为主且没有微乳头亚型和实体亚型的成分)和高危组(含有微乳头或实体亚型)。AI自动提取结节CT形态学特征并采用密度直方图自动计算结节CT定量指标以综合评估危险度。采用二元logistic回归分析并绘制风险指数的受试者工作特征(ROC)曲线进行评估。结果 经统计分析显示,GGN形态、垂直径、CT平均值是浸润性病变组的独立风险因素,其中GGN垂直径预测浸润性的诊断效能最大;GGN的CT最小值、实性占比、毛刺征是IA中高危组的独立风险因素,其中实性占比预测预后不良的诊断效能最大。结论 基于AI提取GGN的CT形态学特征及定量指标有助于磨玻璃样肺腺癌浸润程度及病理亚型的预测。Objective An AI-based(artificial intelligence)study to evaluate the invasiveness of malignant nodules that manifested as ground glass on high-resolution computed tomography(HRCT).Methods 205 cases of GGN confirmed by surgery and pathology were retrospectively analyzed,including 79 males and 126 females,with an average age of 59.87±11.49 years.According to pathological infiltration,it can be divided into microinvasive adenocarcinoma(MIA)and invasive adenocarcinoma(IA).Based on pathological subtypes,IA was divided into low-risk groups(containing papillary,adherent or acinous types without micropapillary or solid subtypes)and high-risk groups(containing micropapillary or solid subtypes).AI automatically extracted CT morphological features and calculated quantitative indicators of nodules by density histogram to evaluate the risk comprehensively.Using binary logistic regression analysis and receiver operating characteristic(ROC)curve of risk index to evaluation.Results According to statistical analysis,GGN shape,vertical diameter and CT mean value were independent risk factors for invasive group,and the vertical diameter of GGN had the greatest diagnostic efficiency in predicting invasiveness.The CT minimum value,solid proportion and spiculation sign of GGN were independent risk factors in IA high-risk group,the solid proportion had the greatest diagnostic efficiency in predicting poor prognosis.Conclusion The CT morphological features and quantitative parameters of GGN extracted based on AI are helpful to predict the infiltration degree and pathological subtypes.

关 键 词：人工智能 磨玻璃样肺腺癌 病理亚型 高分辨率计算机断层扫描 

分 类 号：R734.2[医药卫生—肿瘤]