The Diagnosis Of Lung Cancer In CT And MRI:Comparison Between Radiomics And Conventional Imaging Metrics

Part 1 Comparison of prediction models with radiological semantic features and radiomics in lung cancer diagnosis of the pulmonary nodules:A case-control studyPurpose:To compare the ability of radiological semantic and quantitative texture features in lung cancer diagnosis of pulmonary nodules.Materials and Methods:We performed a retrospective study with 121 NSCLCs and 117control cases.The nodules were contoured by a radiologist through chest CT images and texture features calculated.Radiologists also evaluated semantic features including size,shape,density,emphysema,invasiveness and lymphadenopathy from CT images with or without contrast enhancement.Three different models were compared using LASSO logistic regression:'CS'using clinical and semantic variables,'T'using texture features,and'CST'using clinical,semantic,and texture variables.For each model we performed 100trials of 5-fold cross-validation and the average receiver operating curves was accessed,The AUC of the cross-validation study(AUC_CV)was calculated together with its 95%confidence interval.Results:The AUC_CV?and 95%confidence interval?for models T,CS and CST was 0.85?0.71-0.96?,0.88?0.77-0.96?and 0.88?0.77-0.97?.After separating the data into two groups with or without contrast enhancement,the AUC?without cross-validation?of the model T was 0.86 both for images with and without contrast-enhancement suggesting that contrast enhancement did not impact the utility of texture analysis.Conclusions:The models with semantic and texture features provided cross-validated AUCs of 0.85-0.88 for classification of benign versus cancerous nodules,showing potential to aid in the management of patients.Part 2 An exploratory whole-lesion DCE-MRI intensity histogram analysis for indeterminate pulmonary lesion classificationPurpose: To explore the value of DCE-MRI intensity histogram radiomics,relative to conventional metrics,for classifying indeterminate pulmonary lesions.Materials and Methods: This retrospective study enrolled 49 patients with indeterminate pulmonary lesions who underwent DCE-MRI scans and subsequently had histopathologic confirmation of diagnosis.Three conventional metrics?maximum enhancement ratio,peak time [Tmax] and slope?and eight intensity histogram metrics?volume,integral,maximum,minimum,median,coefficient of variation [Co V],skewness,and kurtosis?were extracted from the DCE-MRI images.All conventional metrics and intensity histogram metrics between benign and malignant group were compared using Wilcoxon rank-sum test.For significant metrics,the univariate receiver operating characteristic?ROC?analysis and an exploratory multivariate logistic regression analysis were conducted to evaluate the diagnostic performance for discriminating benign and malignant lesions and the incremental value of the metrics.We also assessed the diagnostic performance of DCE-MRI radiomic signatures using logistic regression,decision tree,random forest,adaptive boosting,and support vector machine classification algorithms.Diagnostic accuracy,sensitivity and specificity of each classifier was estimated using leave-one-out cross-validation?LOOCV?.Results: Based on the final histopathology,the 49 patients were classified into 2 groups:malignant group?n=33?and benign group?n=16?.Lower Co V?OR=0.2 per 1-SD increase,p=0.0006?,lower Tmax?OR=0.4 per 1-SD increase,p=0.005?and steeper slope?OR=2.4 per1-SD increase,p=0.010?were significantly associated with malignancy.The estimated AUC?area under ROC curve?for Co V,Tmax and slope were 0.81,0.75 and 0.73,respectively.Under multivariate analysis,Co V was significantly independently associated with malignancy after accounting for either Tmax?OR = 0.3 per 1-SD increase,p=0.007?,slope?OR = 0.3 per 1-SD increase,p=0.011?,or both combined?OR = 0.2 per 1-SD increase,p=0.005?.The estimated AUC of the two bivariate and the trivariate combination of metrics were 0.84,0.81,and 0.85.Under multivariate analysis,Co V achieved the highest importance scores for pulmonary lesion classification using the random forest algorithm.Across different machine learning algorithms,classifiers trained with intensity histogram features achieved LOOCV average diagnostic accuracy of 77%,while classifiers trained with quantitative conventional DCE-MRI metrics achieved LOOCV average diagnostic accuracy of 65%.Conclusions: Our exploratory study found that Co V,a new imaging biomarker based in DCE-MRI intensity histogram analysis,was independently associated with malignancy,which can be used to classify indeterminate pulmonary lesions and may complement conventional quantitative DCE MRI metrics.Larger prospective studies are warranted to validate the value of DCE-MRI intensity histogram analysis.