Leveraging Machine Learning to Optimize Donor Recipient Size Matching in Heart Transplantation
Principal Investigator
Name
Veli Topkara
Degrees
M.D.
Institution
Columbia University Medical Center
Position Title
Assistant Professor of Medicine
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-959
Initial CDAS Request Approval
Sep 20, 2022
Title
Leveraging Machine Learning to Optimize Donor Recipient Size Matching in Heart Transplantation
Summary
The purpose of this study is to utilize donor and recipient chest CT images to perform virtual size matching by automated cardiac segmentation and size quantification using a machine learning algorithm. One in every five donor heart offers are declined due to donor-recipient size mismatch since recipient undersizing or oversizing can lead to adverse patient outcomes. The current practice for safe heart size matching is to use a donor to recipient body weight ratio above 0.7. Even a ratio of predicted heart mass based on a MESA MRI study formula using gender, age, height, and weight, has a correlation coefficient of 0.69 against total cardiac volume based on actual donor CT images. Since 2021, 90% of donor heart offers have CT chest images available at the time of decision making, we propose that actual organ size matching using non-gated non-contrast chest CT images will lead to better outcomes than predicted organ size matching. We will use chest CT images in individuals without heart disease, to calculate the total heart volume in normal individuals across age/gender groups and train an AI model on this dataset so that we can automate the segmentation process. The calculation of an actual heart mass ratio will assist in the determination of whether the donor is a good match for the potential recipient.
Aims
1. Develop an AI algorithm which will automate whole heart segmentation and size estimation using chest CT DICOM image files.
2. Formulate a regression equation for the CT measured total cardiac volume using demographic data such as age, gender, height, and weight.
3. Test whether AI guided donor recipient size matching results in improved donor selection.
Collaborators
Veli Topkara, MD, Columbia University Medical Center
Kevin Clerkin, MD, Columbia University Medical Center
Mert Sabuncu, PhD, Weill Cornell Medicine
Benjamin Lee, PhD, Weill Cornell Medicine