Improving robustness of automatic cardiac function quantification from cine magnetic resonance imaging using synthetic image data

Bogdan A Gheorghiță,Teodora Chițiboi,Puneet Sharma,Mohamed Ali Asik Ali,Jeanette Schulz-Menger,Jens Wetzl,Aaron M Lee,Stefan Neubauer,Constantin Suciu,Steffen E Petersen,Lucian M Itu,Christian Geppert,Stefan K Piechnik

doi:10.1038/s41598-022-06315-3

Abstract

Although having been the subject of intense research over the years, cardiac function quantification from MRI is still not a fully automatic process in the clinical practice. This is partly due to the shortage of training data covering all relevant cardiovascular disease phenotypes. We propose to synthetically generate short axis CINE MRI using a generative adversarial model to expand the available data sets that consist of predominantly healthy subjects to include more cases with reduced ejection fraction. We introduce a deep learning convolutional neural network (CNN) to predict the end-diastolic volume, end-systolic volume, and implicitly the ejection fraction from cardiac MRI without explicit segmentation. The left ventricle volume predictions were compared to the ground truth values, showing superior accuracy compared to state-of-the-art segmentation methods. We show that using synthetic data generated for pre-training a CNN significantly improves the prediction compared to only using the limited amount of available data, when the training set is imbalanced.

Highlights

Having been the subject of intense research over the years, cardiac function quantification from magnetic resonance imaging (MRI) is still not a fully automatic process in the clinical practice
We hypothesize that synthetically generated cardiac MRI can substantially improve the performance of our regression model
For a small to moderate training data size, this data imbalance can lead to suboptimal results for the pathological cases, i.e. an AI algorithm trained on such data distributions may perform poorly on the less represented low or high ejection fraction (EF) cases

Summary

Introduction

Having been the subject of intense research over the years, cardiac function quantification from MRI is still not a fully automatic process in the clinical practice. This is partly due to the shortage of training data covering all relevant cardiovascular disease phenotypes. We propose to synthetically generate short axis CINE MRI using a generative adversarial model to expand the available data sets that consist of predominantly healthy subjects to include more cases with reduced ejection fraction. We introduce a deep learning convolutional neural network (CNN) to predict the end-diastolic volume, end-systolic volume, and implicitly the ejection fraction from cardiac MRI without explicit segmentation. The usage of synthetic data has been previously shown to improve deep-learning based segmentation models, when little training data is available[18]

Methods

Results

Conclusion