人工智能CT定量分析对肺腺癌Ki-67LI的预测研究  

Study on Prediction of Ki-67LI in Lung Adenocarcinoma by Artificial Intelligence CT Quantitative Analysis

作　　者：左孟哲[1] 骆磊 张臻[1] 张春雷[1] 王建良[1] ZUO Meng-zhe;LUO lei;ZHANG Zhen;ZHANG Chun-lei;WANG Jian-liang(Radiology Department of Affiliated Kunshan Hospital of Jiangsu University,91 Qianjin West Road,Kunshan 215300,Jiangsu Province,China)

机构地区：[1]江苏大学附属昆山医院放射科,江苏昆山215300

出　　处：《中国CT和MRI杂志》2023年第2期64-67,共4页Chinese Journal of CT and MRI

基　　金：昆山市科技发展社会专项(KS18075);苏州市卫生科技创新项目(SKY2022077);2021年度昆山市一院院内科技专项(KRY-YN034)。

摘　　要：目的 探索肺磨玻璃密度结节(GGN)人工智能(AI)CT定量分析对肺腺癌Ki-67表达指数(Ki-67LI)的预测方法,并评价其对肺腺癌病理类型的预测价值。方法 回顾性分析肺腺癌者230例,共308个GGN,分为不典型腺瘤样增生(AAH)/原位腺癌(AIS)组168个、微浸润性腺癌(MIA)组86个和浸润性腺癌(ICA)组54个。分析CT定量参数和Ki-67LI的组间差异和相关性,采用多元线性回归分析建立Ki-67LI预测模型,并评价该模型对不同病理类型肺腺癌的鉴别价值。结果 3D长径和CT平均值是Ki-67LI的独立预测参数(P<0.05),并建立预测模型:Ki-67LI预测值=1.476+0.311*3D长径+0.003*CT平均值;Ki-67LI预测值在各组间均有统计学差异(P<0.05),且对AAH/AIS和MIA以及MIA和ICA的鉴别价值均较高(AUC值0.837和0.862)。结论 AI CT定量分析可于术前对肺腺癌Ki-67LI进行预测,对GGN的侵袭性有一定预判价值。Objective To explore the prediction method of Ki-67 Labeling index(Ki-67LI)in lung adenocarcinoma by artificial intelligence CT quantitative analysis of lung ground glass nodules(GGN),and to evaluate its value for predicting the pathologic types of lung adenocarcinoma. Methods Retrospective analysis was performed on 230 patients with pathologically confirmed lung adenocarcinoma,including 308 GGN,which were divided into 168 atypical adenomatous hyperplasia/adenocarcinoma in situ(AAH/AIS)group,86 minimally invasive adenocarcinoma(MIA) group and 54 invasive adenocarcinoma(ICA)group. The differences between groups and correlations of CT quantitative parameters and Ki-67LI were analyzed. The prediction model of Ki-67LI was established by multiple linear regression analysis and its value in the differentiation of different pathological types of lung adenocarcinoma was evaluated. Result 3D long diameter and CT mean value were independent predictive parameters of Ki-67LI(P<0.05),and the predictive model was established: Ki-67LI-prediction=1.476+0.311*3D long diameter +0.003* CT mean value. The differences of Ki-67LI-prediction among the multiple groups and between any two groups were statistically significant(P<0.05),and the identification value of it between AAH/AIS and MIA,MIA and ICA was high(AUC value was 0.837 and 0.862). Conclusion Ki-67LI of lung adenocarcinoma can be predicted by artificial intelligence CT quantitative analysis of lung GGN preoperatively,and can be helpful to predict the aggressiveness of GGN.

关 键 词：肺腺癌 磨玻璃结节 Ki-67表达指数 CT定量参数 人工智能 

分 类 号：R445.3[医药卫生—影像医学与核医学] R734.2[医药卫生—诊断学]