Paul E Kinahan

PhD

University of Washington

Professor and Vice-Chair for Research

NLST (Learn more about this study)

NLST-544

Jul 30, 2019

Risk stratification for pulmonary nodules detected by CT imaging using plasma and imaging biomarkers

Lung cancer is the leading cause of cancer deaths worldwide with >159,000 deaths annually in the US alone. The National Lung Screening Trial (NLST) employed low-dose Computed Tomography (CT) imaging of the chest to screen for lung cancer in a high-risk population (smokers aged 55-74). This study demonstrated a 20% reduction in mortality in the group receiving CTs when compared to standard care and has led to generalized acceptance of lung cancer screening in heavy smokers. Unfortunately, pulmonary nodules are a relatively common finding with 25-56% of smokers >50 years of age having CT identifiable pulmonary nodules but less than 2.5% of these actually were cancerous. For diagnosis of incidentally detected pulmonary nodules, current guidelines call for additional imaging and/or invasive biopsy procedures. For both of these scenarios we propose to combine two novel approaches to improve risk stratification for subjects with pulmonary modules. The first involves an antibody array platform for proteomic, glycomic, and autoantibody-antigen complex interrogation that has yielded a four-marker panel with an area under the ROC curve (AUC) of 0.82 in prediagnostic samples and 0.83 in a validation diagnostic set of malignant and benign nodules. The second novel component is the analysis of quantitative nodule features extracted from CT images using the methods of 'radiomics'. We have developed a validated radiomics pipeline that used machine learning algorithms for image texture features that when combined with radiologist-described shape, or semantic features yielded an AUC of 0.82 using the same diagnostic sample set described above. We have created a rule that combines clinical factors (age, smoking etc.), plasma biomarkers, radiomic CT image semantic and texture features for classification of CT-detected nodules as malignant or benign. The addition of both radiomic and biomarkers to the rule significantly increase the AUC (p<0.005) over clinical and semantic CT measures alone. This rule will be tested first in a Vanderbilt CVC incidental/diagnostic cohort, then fixed and tested in the Detection of Early lung Cancer Among Military Personnel Study 1 (DECAMP-1) cohort (Aim 1) with the goal of improving nodule evaluation. We will also test the rule in the NLST screening cohort (Aim 2) to create a final rule that models lung cancer early detection. In Aim 3 we will test the fixed rules from aims 1 and 2 in University of Colorado diagnostic and DECAMP-2 (prediagnostic) cohorts, respectively.

1. Use plasma samples and CT image data from the Vanderbilt diagnostic cohort to finalize a rule combining clinical, biomarker and imaging radiomic features to distinguish malignant from benign nodules and test this fixed rule in samples/images from the Detection of Early lung Cancer Among Military Personnel Study 1 for nodule evaluation in a diagnostic setting.

2. Use the NLST pre-diagnostic plasma samples and CT images to retrospectively validate a combination rule of biomarker and CT clinical and radiomic biomarkers/features for lung cancer early detection screening.

3. After creating final, fixed combination rules of clinical, biomarker, CT semantic and CT radiomic features for screening and diagnostic settings, we will attempt to validate them using samples and CT images from Detection of Early lung Cancer Among Military Personnel Study 2 and LCEDPC screening cohorts and diagnostic samples and CT images from Colorado.