Study
PLCO
(Learn more about this study)
Project ID
2017-9001
Initial CDAS Request Approval
Sep 20, 2017
Title
Identifying lethal lung cancers from non-lethal disease using CT and pathology image markers
Summary
This project will create high resolution H&E images from the National Lung Screening Trial (NLST) H&E slides using hyperspectral slide scanning system, and to create second harmonic generation microscopy scans from a subset of these slides using multiphoton Olympus microscope. The objective is to extract features from CT images, H&E images, and microspcopy collagen 1 fibers by recognizing various lung abnormality patterns. Machine learning will be used to develop diagnostic algorithms to predict lung cancer progression. Several risk scores will be developed to help clinicians to determine who needs aggressive treatment and who can be followed conservatively. Images generated from this project will be made available to lung cancer research communities.
Aims
Aim 1. To extract features from CT images, H&E pathology images, and microspcopy images, and to identify markers that are associated with overall survival.
Aim 2. To develop risk scores to identify lethal lung cancers from non-lethal disease by combining markers from aim 1 and epidemiology risk factors.
Aim 3. To determine differences in collagen 1 fiber patterns in lung cancer patients with poor survival and prolonged survival, and to test if the addition of collagen 1 fiber features could increase the prediction accuracy.
Collaborators
Peng Huang (Johns Hopkins University)
Calum MacAulay (British Columbia Cancer Agency)
Pei-Hsun Wu (Johns Hopkins University)
Zaver M. Bhujwalla (Johns Hopkins University)
Kristine Glunde (Johns Hopkins University)
Peter B. Illei (Johns Hopkins University)
