Virtual Library
Start Your Search
J. Sicks
Author of
-
+
P2.13 - Radiology/Staging/Screening (ID 714)
- Event: WCLC 2017
- Type: Poster Session with Presenters Present
- Track: Radiology/Staging/Screening
- Presentations: 1
- Moderators:
- Coordinates: 10/17/2017, 09:30 - 16:00, Exhibit Hall (Hall B + C)
-
+
P2.13-014 - Computed Tomography-Based Radiomic Classifier Distinguishes Malignant from Benign Pulmonary Nodules in the National Lung Screening Trial (ID 10244)
09:30 - 09:30 | Author(s): J. Sicks
- Abstract
Background:
In the National Lung Screening Trial (NLST), indeterminate pulmonary nodules were detected in 40% of high-risk individuals screened by low dose high-resolution computed tomography (HRCT). However 96% of these nodules were benign indicating that overdiagnosis represents a major challenge for the clinical implantation of CT based lung cancer screening. While current clinical-radiological risk prediction models are very valuable, optimization of the clinical management of larger (≥ 7 mm) screen-detected nodules to avoid unnecessary diagnostic interventions including futile thoracotomies better strategies are needed. Herein we demonstrate the potential value of a novel radiomics based approach for the classification of screen-detected indeterminate nodules.
Method:
Independent quantitative variables assessing various radiologic nodule features such as sphericity, flatness, elongation, spiculation, lobulation and curvature, using 726 nodules (all ≥ 7 mm) were developed from the NLST dataset (benign, n=318 and malignant, n=408). Multivariate analysis was performed using least absolute shrinkage and selection operator (LASSO) method for variable selection and regularization in order to enhance the prediction accuracy and interpretability of the multivariate model. To increase the stability of the modeling, LASSO was run 1,000 times and the variables that were selected in at least 50% of the runs were included into the final multivariate model. The bootstrapping method was then applied for the internal validation and the optimism-corrected AUC was reported for the final model.
Result:
Eight radiologic features were selected by LASSO multivariate modeling out of 57 quantitative radiological variables considered for inclusion. These 8 features include variables capturing vertical location (centroid_Z), volume estimate (Min Enclosing Brick), flatness, texture analysis (SILA_Tex), surface complexity (Max_SI and Avg_SI), and estimates of surface curvature (Avg_PosMeanCurv and Min_MeanCurv), all with P<0.01. The optimism-corrected AUC is 0.939.
Conclusion:
Our novel radiomic HRCT-based approach to non-invasive screen-detected nodule characterization appears extremely promising. Independent external validation is needed.
