AI深度学习对早期肺腺癌不同病理类型的精准识别  

AI deep learning for accurate recognition of different pathological types of early lung adenocarcinoma

作　　者：陈唐欣汐 肖运平[2] 潘永军 CHEN Tangxinxi;XIAO Yunping;PAN Yongjun(Guangxi University of Traditional Chinese Medicine,Nanning 530200,China;Department of Radiology,Liuzhou People's Hospital of Guangxi,Liuzhou 545026,China)

机构地区：[1]广西中医药大学,广西南宁530200 [2]广西柳州市人民医院放射科,广西柳州545026

出　　处：《医学影像学杂志》2024年第5期56-60,共5页Journal of Medical Imaging

基　　金：广西壮族自治区卫生健康委员会自筹经费科研课题(编号:Z-B20221376)。

摘　　要：目的 探讨人工智能(AI)深度学习对早期肺腺癌的不同病理类型诊断效能,分析AI深度学习对不同病理类型应用价值。方法 选取经手术病理证实为早期肺腺癌患者90例的CT图像,按照病理类型分为原位腺癌组、微小浸润性腺癌组、浸润性腺癌组,每组30例,并按照2:1的比例随机分为训练组(n=60例)与验证组(n=30例),利用联影AI科研平台半自动分割靶病灶并提取影像组学特征,利用验证组的数据测试训练深度学习模型,分析病灶CT形态特征,计算病灶CT定量参数,并对肺腺癌的病理类型进行预测评分,分析诊断效能。结果 三组在性别、年龄、结节位置、结节大小、结节体积、结节质量、最小CT值、空泡征等比较,差异无统计学意义(P>0.05);在结节密度、长径、短径、体积、AI预测得分、最大CT值、平均CT值、分叶征、血管集束征、毛刺征、胸膜凹陷征结果比较,差异有统计学意义(P<0.05)。浸润性腺癌的长径、短径、直径高于原位腺癌组和微小浸润性腺癌组;浸润性腺癌组的AI预测得分、最大CT值、平均CT值高于原位腺癌组和微小浸润性腺癌组;浸润性腺癌组在分叶征、血管集束征、毛刺征、胸膜凹陷征的比例高于原位腺癌组和微小浸润性腺癌组。结论 基于AI学习在诊断肺腺癌不同病理类型中具有较高的诊断价值,能够精准识别不同的病理类型,具有较高的可重复性和操作性。Objective To investigate the diagnostic efficacy of AI deep learning in different pathological types of early lung adenocarcinoma,and to analyze the application value of AI deep learning in different pathological types.Methods CT im⁃ages of 90 patients with early lung adenocarcinoma confirmed by surgery and pathology were selected and divided into in situ adenocarcinoma group,microinvasive adenocarcinoma group and invasive adenocarcinoma group,with 30 cases in each group,and were randomly divided into training group(n=60 cases)and verification group(n=30 cases)according to the ratio of 2:1.The joint film artificial intelligence research platform was used to semi-automatically segment the target lesions and ex⁃tract the image omics features,and the data of the validation group wene used to test the training deep learning model made.We analyzed the CT morphological features of the lesions,calculated the quantitative parameters of the lesions,and made a prediction score for the pathological types of lung adenocarcinoma so as to analyze the diagnostic effectiveness.Results There were no significant differences in gender,age,nodule location,nodule size,nodule volume,nodule mass,minimum CT value and cavitation sign among the three groups(P>0.05).There were statistically significant differences in nodule den⁃sity,long diameter,short diameter,volume,AI prediction score,maximum CT value,average CT value,lobular sign,vas⁃cular bunching sign,burr sign and pleural depression sign(P<0.05).The long diameter,short diameter and diameter of in⁃vasive adenocarcinoma were higher than those of in situ adenocarcinoma group and microinvasive adenocarcinoma group.The predicted AI score,maximum CT value and average CT value of invasive adenocarcinoma group were higher than those of in situ adenocarcinoma group and microinvasive adenocarcinoma group.The proportion of lobular sign,vascular bunching sign,burr sign and pleural sag sign in invasive adenocarcinoma group was higher than that in situ adenocarcinoma group and micro⁃i

关 键 词：人工智能 AI深度学习 肺腺癌 精准识别 体层摄影术 X线计算机 

分 类 号：R734.2[医药卫生—肿瘤] R814.42[医药卫生—临床医学]