Radiomics analysis of pulmonary nodules in low‐dose CT for early detection of lung cancer

• DOI: 10.1002/mp.12820
• Publication Date: 2018-02-21T16:37:20Z
• Authors: Wookjin Choi; Jung Hun Oh; Sadegh Riyahi; Chia‐Ju Liu; Feng Jiang; Wengen Chen; Charles White; Andreas Rimner; James G. Mechalakos; Joseph O. Deasy; Wei Lu
• Keywords: Lasso

To develop a radiomics prediction model to improve pulmonary nodule (PN) classification in low-dose CT. compare the with American College of Radiology (ACR) Lung CT Screening Reporting and Data System (Lung-RADS) for early detection lung cancer.We examined set 72 PNs (31 benign 41 malignant) from Image Database Consortium image collection (LIDC-IDRI). One hundred three radiomic features were extracted each PN. Before building process, distinctive identified using hierarchical clustering method. We then constructed by support vector machine (SVM) classifier coupled least absolute shrinkage selection operator (LASSO). A tenfold cross-validation (CV) was repeated ten times (10 × 10-fold CV) evaluate accuracy SVM-LASSO model. Finally, best 10 CV further evaluated 20 5- 50 2-fold CVs.The consisted only two features: bounding box anterior-posterior dimension (BB_AP) standard deviation inverse difference moment (SD_IDM). The BB_AP measured extension PN direction highly correlated (r = 0.94) size. SD_IDM texture feature that directional variation local homogeneity IDM. Univariate analysis showed both statistically significant discriminative (P 0.00013 0.000038, respectively). larger or smaller more likely malignant. SVM achieved an 84.6% 0.89 AUC. By comparison, Lung-RADS 72.2% 0.77 AUC four (size, type, calcification, spiculation). improvement comparing (McNemar's test P 0.026). misclassified 19 cases because it mainly based on size, whereas correctly classified these combining size (SD_IDM) feature. performance stable when leaving patients out five- twofold CVs (accuracy 84.1% 81.6%, respectively).We developed predict malignancy features. demonstrated 84.6%, which 12.4% higher than Lung-RADS.