View Abstract

A4417 - Quantitative Imaging of the Indeterminate Pulmonary Nodule and Surrounding Parenchyma for Lung Cancer Diagnosis Within the DECAMP Consortium
Author Block: E. Billatos1, E. Moses2, F. Duan3, H. Marques4, C. Stevenson5, M. Lorenzi6, M. E. Lenburg7, A. Spira8, G. R. Washko9, DECAMP Consortium; 1Pulmonary and Critical Care, Boston University, Boston, MA, United States, 2Boston University School of Medicine, Boston, MA, United States, 3Biostatistics, Providence, RI, United States, 4Brown University, Providence, RI, United States, 5Janssen Pharmaceuticals, London, United Kingdom, 6Janssen Pharmaceuticals, Spring House, PA, United States, 7Boston University, Brookline, MA, United States, 8Boston University, Newton, MA, United States, 9Brigham and Women's, West Roxbury, MA, United States.
Rational: Quantitative Computed Tomography (QCT) imaging has increasingly been used to evaluate smoking-related lung diseases including emphysema, interstitial lung disease, and lung nodules. In this study, we leverage DECAMP, a unique cohort of military and veteran personnel to evaluate the ability of QCT derived metrics of the nodule and its immediate surrounding lung parenchyma to identify patients with lung cancer among smokers with indeterminate pulmonary nodules.
Methods: CT imaging was obtained on 144 current and former smokers undergoing bronchoscopy in the diagnostic work-up for indeterminate pulmonary nodules (7-30 mm). Subjects were followed clinically for up to two years until a final diagnosis of lung cancer or benign disease was made (84 lung cancer, 60 benign). Differences between benign and malignant nodules were assessed using the Wilcoxon signed rank test for continuous variables, and using Chi-square or Fisher’s Exact test for categorical variables. LASSO regressions were performed to select imaging features that differentiate malignant from benign nodules and 100 bootstraps were used to increase robustness of the fitted models. The area under the curve (AUC) was computed to assess feature performance, along with the net reclassification improvement (NRI).
Results: Between subjects with malignant and benign nodules, there were no differences in age, gender, BMI, smoking status, or COPD status. Patients with lung cancer had significantly higher pack years, larger nodule sizes, more family history of lung cancer, and a higher pretest probability of malignancy. QCT features of the nodule and its immediate surrounding parenchyma (10 mm radius from nodular surface) distinguished malignancy with an AUC of 0.77. No further benefit was observed when extending the boundary to 15 or 20mm. The nodule and 10 mm of surrounding lung parenchyma resulted in a 5% improvement in sensitivity and 7% improvement in specificity over nodular features alone. Further inclusion of pack-years did not significantly improve sensitivity or specificity.
Conclusion: These findings suggest that QCT imaging of a lung nodule in addition the nodule itself may improve diagnostic yield. Of note, incorporation of the parenchyma beyond 10mm decreased feature performance, suggesting that features in the parenchyma immediately surrounding the nodule play a significant role in distinguishing malignancy. Mild aberrations detected only by QCT in the surrounding lung parenchyma that are specific to lung cancer may be leveraged to develop a radiomic biomarker for the early detection of lung cancer in high-risk smokers.