Encrypted login | home

Program Information

Radiomics Analysis of Pulmonary Nodules in Low Dose CT for Early Detection of Lung Cancer

W Choi

W Choi1*, J Oh1 , S Riyahi Alam1 , F Jiang2 , W Chen2 , J Deasy1 , W Lu1 , (1) Memorial Sloan Kettering Cancer Center, New York, NY, (2) University of Maryland School of Medicine, Baltimore, MD


TH-AB-201-10 (Thursday, August 3, 2017) 7:30 AM - 9:30 AM Room: 201

Purpose: To develop a radiomics prediction model to improve pulmonary nodule classification in low dose CT. To compare the model against the ACR Lung-RADS for early detection of lung cancer.

Methods: The Lung Image Database Consortium image collection (LIDC-IDRI) in the Cancer Imaging Archive (TCIA) was examined. We evaluated a subset of 79 pulmonary nodules (36 benign and 43 malignant). CT radiomic features (n=103) were extracted from each nodule. Redundant features were removed with a hierarchical clustering method. We then constructed a prediction model by using a support vector machine (SVM) classifier with important features chosen by a least absolute shrinkage and selection operator (LASSO). A 10-fold cross-validation was repeated ten times to obtain the model accuracy.

Results: The ACR Lung-RADS achieved 73.6% accuracy and 0.74 AUC with four features (axial largest diameter (LD), type, calcification, and spiculation). The SVM-LASSO model achieved 80.9% accuracy and 0.81 AUC with two features: the bounding box anterior-posterior dimension (BB_AP) and the standard deviation of inverse difference moment (SD_IDM). Both features were always chosen (100 times) by the LASSO. The BB_AP measured the extension of a nodule in anterior-posterior direction and it was highly correlated (r=0.96) with the LD. The SD_IDM is a texture feature that measured the directional variation of the local homogeneity feature IDM. In univariate analysis, both features were significant predictors of nodule malignancy (P<0.0001, Wilcoxon rank sum test). The ACR Lung-RADS misclassified some cases since it is mainly based on size (LD), while the SVM-LASSO model correctly classified them by combining size (BB_AP) and texture (SD_IDM) features.

Conclusion: We developed a SVM-LASSO model to predict malignancy of pulmonary nodules with CT radiomic features. We demonstrated that the model achieved higher accuracy than the ACR Lung-RADS.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the National Cancer Institute Grants R01CA172638.

Contact Email: