A Generative Framework for Predicting Myocardial Strain from Cine-Cardiac Magnetic Resonance Imaging

Nina Cheng, Rodrigo Bonazzola, Nishant Ravikumar, Alejandro F. Frangi

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Myocardial strain is an important measure of cardiac performance, which can be altered when ejection fraction (EF) and other ventricular volumetric indices remain normal, providing an additional indicator for early detection of cardiac dysfunction. Cardiac tagging MRI is the gold standard for myocardial strain quantification but requires additional sequence acquisition and relatively complex post-processing procedures, which limit its clinical application. In this paper, we propose a framework for learning a joint latent representation of cine MRI and tagging MRI, such that tagging MRI can be synthesised and used to derive myocardial strain, given just cine MRI as inputs. Specifically, we use a multi-channel variational autoencoder to simultaneously learn features from tagging MRI and cine MRI, and project the information from these distinct channels into a common latent space to jointly analyse the multi-sequence data information. The inference process generates tagging MRI using only cine MRI as input, by conditionally sampling from the learned latent representation. Finally, automated tag tracking was performed using a cardiac motion tag tracking network on the generated tagging MRI, and myocardial strain was estimated. Experiments on the UK Biobank dataset show that our proposed framework can generate tagging images from cine images alone, and in turn, can be used to estimate myocardial strain effectively.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
EditorsGuang Yang, Angelica Aviles-Rivero, Michael Roberts, Carola-Bibiane Schönlieb
PublisherSpringer Nature
Pages482-493
Number of pages12
ISBN (Print)9783031120527
DOIs
Publication statusPublished - 2022
Event26th Annual Conference on Medical Image Understanding and Analysis, MIUA 2022 - Cambridge, United Kingdom
Duration: 27 Jul 202229 Jul 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13413 LNCS

Conference

Conference26th Annual Conference on Medical Image Understanding and Analysis, MIUA 2022
Country/TerritoryUnited Kingdom
CityCambridge
Period27/07/2229/07/22

Keywords

  • Cardiac cine MRI
  • Cardiac tagging MRI
  • Convolutional neural network
  • Machine learning
  • Myocardial strain estimation

Fingerprint

Dive into the research topics of 'A Generative Framework for Predicting Myocardial Strain from Cine-Cardiac Magnetic Resonance Imaging'. Together they form a unique fingerprint.

Cite this