Nonlinear performance analysis and prediction for robust low dose lung CT
Principal Investigator
Name
Grace Gang
Degrees
PhD
Institution
Johns Hopkins University
Position Title
Assistant Research Professor
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-728
Initial CDAS Request Approval
Nov 17, 2020
Title
Nonlinear performance analysis and prediction for robust low dose lung CT
Summary
The proliferation of nonlinear algorithms in CT has posed significant challenges to image quality assessment and optimization. Many of these algorithms defy traditional signal and noise tradeoff and are highly dependent on the data they operate on. This project aims to develop an understanding of important data features pertaining to CT imaging of the lung and apply these features in the systematic assessment and optimization of nonlinear algorithms.\\
To this end, we aim to acquire chest CT volume images from various sources to create a large lung lesion library. We will analyze the morphology and texture of lung nodules in order to quantify underlying diagnostic features. We will explore both traditional analytic decomposition methods and emerging data-driven feature discovery. These features will be used as inputs to characterize the performance of particular nonlinear algorithms including model-based and deep learning data processing. Furthermore, we will develop methods to increase the diversity of the lesion library with synthetic lesions which may be used in virtual clinical trials, data augmentation of machine learning algorithms, as well as system design and optimization.
To this end, we aim to acquire chest CT volume images from various sources to create a large lung lesion library. We will analyze the morphology and texture of lung nodules in order to quantify underlying diagnostic features. We will explore both traditional analytic decomposition methods and emerging data-driven feature discovery. These features will be used as inputs to characterize the performance of particular nonlinear algorithms including model-based and deep learning data processing. Furthermore, we will develop methods to increase the diversity of the lesion library with synthetic lesions which may be used in virtual clinical trials, data augmentation of machine learning algorithms, as well as system design and optimization.
Aims
Aim 1: Create large lung nodule library from various clinical lung datasets
Aim 2: Shape and texture analysis of various classes of lung nodules
Aim 3: Application of the aim 2 analysis in nodule synthesis and assessment of nonlinear, data-driven imaging algorithms
Collaborators
Jessica Flores. Johns Hopkins University
Shaoyan Pan, Johns Hopkins University
Grace Gang PhD, Johns Hopkins University
Webster Stayman PhD, Johns Hopkins University