Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Master's Award: Canada Graduate Scholarships

University of Toronto

$17,500.00

Domain-Adversarial Neural Networks for Robust Segmentation of the Left and Right Ventricle in Cardiac MR Images

FY 2017-18

Summary
Cardiac magnetic resonance imaging is frequently used for the assessment of cardiac function. Segmentation of the left and right ventricles of the heart is required to calculate functional parameters and diagnose cardiovascular disease. Over the past years, several groups have developed automated segmentation tools in order to have more accurate and efficient diagnosis. Recently, convolutional neural networks have been shown to reach human-level performance in the segmentation of the left and right ventricles on the UK Biobank dataset. However, these neural networks usually fail when tested on images from a different scanner or images acquired with slightly different parameters. This is a problem if we want to apply them to images from different medical centres. A promising approach to this problem is the use of a domain-adversarial neural network which attempts to learn features that are domain-invariant from the source and target datasets. The domain-adversarial neural network will be trained and tested on different pairs of source and target datasets to evaluate the generalizability of the network. The Dice similarity coefficient, mean contour distance, and Hausdorff distance will be used as metrics for the performance of the network. This work will enable robust segmentation of the ventricles in the real-world setting and more efficient diagnosis of cardiovascular disease.
Program
Master's Award: Canada Graduate Scholarships
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
CIHR Funding Decisions Database

As published

Id
386633
Name
Ng Matthew
Region
Ontario
Country
Canada
Orgtype
Institution Paid; Research Institution
Prcname2
Special Cases - Awards Programs
Projecttitle
Domain-Adversarial Neural Networks for Robust Segmentation of the Left and Right Ventricle in Cardiac MR Images
Programtype2
Studentship Programs
Orgnameinp2
University of Toronto
Cihramount2
$17,500
Cihrequipment2
$0
Programname2
Master's Award: Canada Graduate Scholarships
Keyworddelim
Cardiac Magnetic Resonance Imaging; Cardiac Segmentation; Machine Learning
Abstract
Cardiac magnetic resonance imaging is frequently used for the assessment of cardiac function. Segmentation of the left and right ventricles of the heart is required to calculate functional parameters and diagnose cardiovascular disease. Over the past years, several groups have developed automated segmentation tools in order to have more accurate and efficient diagnosis. Recently, convolutional neural networks have been shown to reach human-level performance in the segmentation of the left and right ventricles on the UK Biobank dataset. However, these neural networks usually fail when tested on images from a different scanner or images acquired with slightly different parameters. This is a problem if we want to apply them to images from different medical centres. A promising approach to this problem is the use of a domain-adversarial neural network which attempts to learn features that are domain-invariant from the source and target datasets. The domain-adversarial neural network will be trained and tested on different pairs of source and target datasets to evaluate the generalizability of the network. The Dice similarity coefficient, mean contour distance, and Hausdorff distance will be used as metrics for the performance of the network. This work will enable robust segmentation of the ventricles in the real-world setting and more efficient diagnosis of cardiovascular disease.
Competitiondate
201712
Approvedterm2
1 yr 0 mth
Pinamesdelim
Ng, Matthew
Approvedterm
1 an 0 mois
Cihrcontribution2
~ $17,500~ $0~