Automatic detection of cardiac abnormalities in low dose chest CT scans using machine learning techniques
Principal Investigator
Name
Serkan Cimen
Degrees
Ph.D.
Institution
Siemens Medical Solutions USA Inc.
Position Title
Senior Research Scientist
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-457
Initial CDAS Request Approval
Dec 11, 2018
Title
Automatic detection of cardiac abnormalities in low dose chest CT scans using machine learning techniques
Summary
Low dose chest CT, acquired for the main purpose of lung cancer screening, can be utilized to extract other clinically useful information. In this project, we will explore several machine learning algorithms i) to detect and quantify coronary artery calcifications in order to automate calcium scoring in low-dose chest CT images, ii) to segment and measure aortic diameters and ii) to segment and measure pulmonary artery diameter. These automated measurements could potentially provide the first evidence of serious diseases, such as coronary atherosclerosis, thoracic aortic disease, and pulmonary hypertension.
A small randomly selected subset of the data will be used for training the machine learning algorithms. Remaining large dataset will be used to perform quantitative evaluation and to understand generalization abilities of the proposed algorithms.
A small randomly selected subset of the data will be used for training the machine learning algorithms. Remaining large dataset will be used to perform quantitative evaluation and to understand generalization abilities of the proposed algorithms.
Aims
1) Develop an automated heart segmentation algorithm to measure heart volume and to define a region of interest for the calcium detection algorithm.
2) Develop an automated calcium scoring algorithm in low-dose chest CT.
3) Develop an automated aorta segmentation algorithm and measure aortic diameters.
4) Develop an automated pulmonary artery segmentation algorithm and measure pulmonary diameter.
5) Perform a quantitative evaluation of the algorithms in a large dataset
Collaborators
None