A Multi-platform Biomarker Approach to Reduce False Positives in Lung Cancer Screening

Case ID:
C13949
Disclosure Date:
1/4/2016
Description:
The recently completed National Lung Screening Trial (NLST) demonstrated a 20% mortality reduction in low-dose CT (LDCT) screening as compared to chest radiography among smokers at high risk for lung cancer. This finding has led to Medicare coverage of LDCT for lung cancer screening. However, this mortality reduction is weighted against the harm from a high false positive (FP) rate with over 95% of LDCT positive nodules not resulting in a lung cancer diagnosis. CT images with identified pulmonary nodule(s) can be grouped into three categories: (1) images with all nodules having a high certainty of being benign, (2) images with at least one nodule having a high certainty of malignancy, and (3) images with one or more indeterminate nodules (IPNs) that require further evaluation. We investigate those subjects whose images belong to category 3 because the first two categories are easily classified by radiologists. Images with IPN are more challenging to evaluate and are the major source of FP screens. In the NLST study, about 94.7% of FP LDCT screens were from images with small IPNs that did not result in lung cancer diagnoses. We propose to reduce the FP and thus to improve the diagnostic accuracy rate by adding two additional noninvasive diagnostic tests for subjects with IPNs from LDCT scans, and help clinicians to optimize the nodule management strategy by providing two probabilities of lung cancer development within one year and within two years. The first test is based on a re-evaluation of existing LDCT scans using image texture markers; the second test is based on the combination of image texture markers, sputum and blood methylation biomarkers, and blood metabolic biomarkers. These diagnostic tests will be derived from four prediction models. Our goal will be achieved by the following three aims: Aim 1. To extract image features from the LDCTs for both the NLST training and test sets. All nodules will be identified by radiologists and will be segmented by a semi-automatic segmentation method. Image texture features will be extracted from within the nodule (intranodular), the surrounding lung parenchyma (perinodular), and outside the nodule (extranodular) including lymph nodes. Aim 2. To quantify plasma and sputum biomarkers for both the NLST training and test sets. A six-gene promoter methylation biomarker panel derived in earlier studies will be measured in sputum and plasma assay DNA methylation using a single tube DNA extraction and
processing technique dubbed “Methylation-On-Beads” (MOB) recently developed by our investigators. Plasma metabolites will be measured by high-resolution 1H MR spectroscopy. Aim 3. To identify biomarkers and develop four prediction models using the NLST training data, and to validate model prediction accuracies using the NLST test data. Biomarkers that are associated with lung cancer diagnosis and time to cancer development will be identified. Two diagnoses will be developed from four hierarchical models. The first diagnosis from model 1 uses image texture features. The second diagnosis is derived from models 2-4 with sequentially added biomarkers from sputum methylation, plasma methylation, and plasma metabolites. All models and biomarkers will be validated using independent test data. Our hypothesis is that one-year FP screens from one of the four models will be significantly lower than 50%. In an exploratory analysis, we will compare models 2-4 with model 1 in order to identify the most parsimonious, and yet, high accurate prediction model. Results from this study will help radiologists and clinicians to evaluate IPNs in their daily clinical decision making, to reduce the burden of follow-up imaging and lab tests, and to modify the time when an invasive procedure (such as biopsy or surgery) is needed
Patent Information:
Inventors:
Category(s):
For Information, Contact:
Jon Gottlieb
jgottl10@jhu.edu
410-614-0300
Save This Technology:
2017 © Johns Hopkins Technology Ventures. All Rights Reserved. Powered by Inteum