Predicting Mortality and Causes from Chest CT Images
Principal Investigator
Name
david wilson
Degrees
MD, MPH
Institution
UPMC
Position Title
Assoc Professor
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-762
Initial CDAS Request Approval
Mar 9, 2021
Title
Predicting Mortality and Causes from Chest CT Images
Summary
We have been pioneers in the development of low-dose computed tomography (LDCT) chest imaging for lung cancer screening. The Pittsburgh Lung Screening Study (PLuSS), ongoing since 2002, has resulted in many advances in the field, particularly imaging biomarker discovery. We reported that the presence of emphysema on LDCT is a strong predictor of lung cancer risk (an early imaging biomarker). We developed vessel number as a strong predictor of lung cancer risk in indeterminant pulmonary nodules (IPN). We were the first group to utilize complex neural networks and artificial intelligence to differentiate IPNs into cancer or no cancer. We, along with others, have promoted the potential utility of LDCT for the evaluation of other diseases, and there have been studies of CT phenotypes and CT derived biomarkers and their association with comorbid diseases and outcomes, for example in chronic obstructive pulmonary disease (COPD).
The purpose of this project is to use deep learning or artificial intelligence to form risk profiles that predict disease specific mortality from chest CT images. Tens of millions of chest CT examinations are performed annually in the U.S. Most chest CT examinations are performed to study a specific disorder based on a patient’s complaints, like rule out pulmonary embolism. Many factors, which are implied in the volumetric CT images beyond visual perception and may affect patients’ health, are largely ignored, particularly when not associated with the clinical question under investigation. As a result, many image features are not fully explored in terms of early detection and disease prevention. In this project, we propose to leverage a longitudinal cohort, namely the PLuSS, and use deep learning to analyze and develop imaging biomarkers in the chest CT images. Our goal is to identify image signals of early disease which may have an impact on prognosis. The ongoing PLuSS cohort has enrolled and actively followed more than 4,000 participants. Patient demographics, lung function, death dates, and causes of death are well documented. As of January 1st, 2020, more than 1,000 persons died due to various health-related causes (e.g., heart attack, diabetes, or lung cancer). We will integrate this information and develop a novel computer model to predict an individual’s life expectancy based on a baseline CT scan. We expect that individualized life expectancy prediction can be used to predict future health risk and inform timely screening and potential life-style changes. We propose a novel approach to detect early thoracic and associated disorders. The prediction model can be applied to any chest CT examination regardless of the reason for performing the exam. If successful, the proposed concept will be applicable to many other health problems and may result in a new paradigm of early disease screening, prevention, and optimal patient management.
The purpose of this project is to use deep learning or artificial intelligence to form risk profiles that predict disease specific mortality from chest CT images. Tens of millions of chest CT examinations are performed annually in the U.S. Most chest CT examinations are performed to study a specific disorder based on a patient’s complaints, like rule out pulmonary embolism. Many factors, which are implied in the volumetric CT images beyond visual perception and may affect patients’ health, are largely ignored, particularly when not associated with the clinical question under investigation. As a result, many image features are not fully explored in terms of early detection and disease prevention. In this project, we propose to leverage a longitudinal cohort, namely the PLuSS, and use deep learning to analyze and develop imaging biomarkers in the chest CT images. Our goal is to identify image signals of early disease which may have an impact on prognosis. The ongoing PLuSS cohort has enrolled and actively followed more than 4,000 participants. Patient demographics, lung function, death dates, and causes of death are well documented. As of January 1st, 2020, more than 1,000 persons died due to various health-related causes (e.g., heart attack, diabetes, or lung cancer). We will integrate this information and develop a novel computer model to predict an individual’s life expectancy based on a baseline CT scan. We expect that individualized life expectancy prediction can be used to predict future health risk and inform timely screening and potential life-style changes. We propose a novel approach to detect early thoracic and associated disorders. The prediction model can be applied to any chest CT examination regardless of the reason for performing the exam. If successful, the proposed concept will be applicable to many other health problems and may result in a new paradigm of early disease screening, prevention, and optimal patient management.
Aims
Aim 1: An individual’s cause of death can be predicted from a chest CT exam.
Aim 2: An individual’s life expectancy can be predicted from a chest CT exam.
Aim 3: The inclusion of patient demographic information can improve the accuracy of the predictions.
Collaborators
Jiantao Pu PhD, UPMC Dept of Radiology
Jacob Sechrist MD, UPMC Dept of Radiology
Andriy Bandos PhD, UPitt Dept of Biostatistics