分层分析影像组学模型在肺腺癌诊断中的价值

The value of stratified analysis of radiomics model in the diagnosis of lung adenocarcinoma

目的:建立并验证可高效鉴别肺腺癌及其浸润程度的预测模型,并根据结节/肿块性质分层分析模型的预测效能。方法:回顾性分析本院2011年10月-2018年12月经病理证实的肺结节/肿块患者2105例。根据肿瘤性质,分为磨玻璃组(A组,1711例)和实性组(B组,394例),组内以2017年10月为界,分为训练集和测试集。收集患者的传统影像特征、人口统计学资料;采用3D slicer对兴趣区(ROI)进行手动勾画,Pyraodiomics提取影像组学特征,最大相关-最小冗余(mRMR)和最小绝对收缩选择算子(LASSO)算法对特征进行筛选、降维,构建预测模型,并采用受试者操作特征(ROC)曲线及曲线下面积(AUC),校准曲线和拟合优度检验,临床决策曲线分别评价模型的预测准确性、校准度、临床实用性。结果:在A组、B组分别建立独立组学模型、传统模型和融合组学模型。A组:融合组学模型在训练集、测试集、外部验证集的AUC分别为0.92(0.90~0.93)、0.94(0.93~0.96)、0.87(0.82~0.91);B组:融合组学模型在训练集、测试集以及外部验证集的AUC分别为0.85(0.80~0.90)、0.80(0.72~0.89)、0.70(0.54~0.85)。De-long检验A组测试集组学模型与传统模型间效能差异有统计学意义(P<0.05);B组组学模型与传统模型间的效能结果差异无统计学意义。无论是A组还是B组,Hosmer-Lemeshow拟合优度检验融合组学模型、传统模型P值均大于0.05,校准曲线中融合组学模型预测值与标准基线重合度较高。决策曲线中融合组学模型的净利润率最高。该结果在外部验证集中得到部分验证。结论:影像组学进一步提升了传统诊断模式对肺腺癌的诊断效能。相比于实性肺结节/肿块,融合组学模型在磨玻璃肺结节/肿块中的预测准确性、校准度、临床实用性更为优异。

Objective:To establish and verify an efficiently model to distinguish lung adenocarcinoma and its invasiveness,and further identify the predictive performance of the model according to the types of nodule/mass.Methods:In the study,2105 patients with pathologically confirmed lung nodules/masses were retrospectively enrolled from October 2011 to December 2018.According to the characteristics of the tumor,they were divided into ground glass group(group A-1711 cases)and solid group(group B-394 cases).With the boundary of October 2017,patients in each group were divided into the training set and test set.Image features and demographic data were collected.3D slicer was used to manually delineate region of interest(ROI).Pyraodiomics was used to extract radiomic features,and max-relevance and min-redundancy(mRMR)and least absolute shrinkage and selection operator(LASSO)algorithms were used to filter features,reduce dimensionality.The prediction models were then constructed,and the area under the curve(AUC)value of the receiver operator characteristic curve,the calibration curve and the goodness of fit test,and the clinical decision curve were used to respectively evaluate the prediction accuracy,calibration,and clinical applicability of the model.Results:Independent radiomics model,traditional model and fusion radiomics model were established in group A and B respectively.In group A:the AUC values of fusion radiomics model in training set,test set,and external validation set were 0.92(0.90~0.93),0.94(0.93~0.96),0.87(0.82~0.91).In group B:The AUC values of the fusion radiomics model in the training set,test set and external validation set were 0.85(0.80~0.90),0.80(0.72~0.89),0.70(0.54~0.85).The De-long test showed a statistical difference in the comparison of the efficacy between the test set radiomics models and the traditional model in group A(P<0.05).The efficacy results between the radiomics models and the traditional model in group B were not statistically different.In both group A and group B,the Hosmer-Lemeshow goodness-of-fit test fusion radiomics model,traditional model P-value was greater than 0.05,and the predicted values of fusion radiomics model in the calibration curve overlapped with the standard baseline.The decision curve had the highest net profitability for the fusion radiomics model.The results were partially validated in the external validation set.Conclusion:Radiomics can improve the diagnostic efficiency of lung adenocarcinoma compared with traditional diagnostic models.Furthermore,fusion radiomics model demonstrated a better performance in ground-glass lung nodules/mases with a better prediction accuracy,calibration,and clinical applicability,compared with solid lung nodules/masses.