Multi-task learning for joint weakly-supervised segmentation and aortic arch anomaly classification in fetal cardiac MRI

Paula Ramirez10000-0002-0705-5296, Alena Uus10000-0001-5796-2145, Milou P.M. van Poppel10000-0002-1739-4726, Irina Grigorescu10000-0002-9756-3787, Johannes K. Steinweg10000-0002-3366-0932, David F.A. Lloyd10000-0003-1759-6106, Kuberan Pushparajah10000-0003-1541-1155, Andrew P. King10000-0002-9965-7015, Maria Deprez10000-0002-2799-6077
1: Biomedical Engineering and Imaging Sciences, King's College London
Publication date: 2023/11/11
PDF · Code · arXiv


Congenital Heart Disease (CHD) is a group of cardiac malformations present already during fetal life, representing the prevailing category of birth defects globally. Our aim in this study is to aid 3D fetal vessel topology visualisation in aortic arch anomalies, a group which encompasses a range of conditions with significant anatomical heterogeneity. We present a multi-task framework for automated multi-class fetal vessel segmentation from 3D black blood T2w MRI and anomaly classification. Our training data consists of binary manual segmentation masks of the cardiac vessels’ region in individual subjects and fully-labelled anomaly-specific population atlases. Our framework combines deep learning label propagation using VoxelMorph with 3D Attention U-Net segmentation and DenseNet121 anomaly classification. We target 11 cardiac vessels and three distinct aortic arch anomalies, including double aortic arch, right aortic arch, and suspected coarctation of the aorta. We incorporate an anomaly classifier into our segmentation pipeline, delivering a multi-task framework with the primary motivation of correcting topological inaccuracies of the segmentation. The hypothesis is that the multi-task approach will encourage the segmenter network to learn anomaly-specific features. As a secondary motivation, an automated diagnosis tool may have the potential to enhance diagnostic confidence in a decision support setting. Our results showcase that our proposed training strategy significantly outperforms label propagation and a network trained exclusively on propagated labels. Our classifier outperforms a classifier trained exclusively on T2w volume images, with an average balanced accuracy of 0.99 (0.01) after joint training. Adding a classifier improves the anatomical and topological accuracy of all correctly classified double aortic arch subjects.
Our code is available at


Deep Learning · Fetal Cardiac Imaging · Congenital Heart Disease · Automated Diagnosis · Fetal Cardiac MRI · Aortic Arch Segmentation · Multi-Task Learning · Multi-Class Vessel Segmentation · Anomaly Segmentation

Bibtex @article{melba:2023:015:ramirez, title = "Multi-task learning for joint weakly-supervised segmentation and aortic arch anomaly classification in fetal cardiac MRI", author = "Ramirez, Paula and Uus, Alena and van Poppel, Milou P.M. and Grigorescu, Irina and Steinweg, Johannes K. and Lloyd, David F.A. and Pushparajah, Kuberan and King, Andrew P. and Deprez, Maria", journal = "Machine Learning for Biomedical Imaging", volume = "2", issue = "PIPPI 2022 special issue", year = "2023", pages = "406--446", issn = "2766-905X", doi = "", url = "" }
RISTY - JOUR AU - Ramirez, Paula AU - Uus, Alena AU - van Poppel, Milou P.M. AU - Grigorescu, Irina AU - Steinweg, Johannes K. AU - Lloyd, David F.A. AU - Pushparajah, Kuberan AU - King, Andrew P. AU - Deprez, Maria PY - 2023 TI - Multi-task learning for joint weakly-supervised segmentation and aortic arch anomaly classification in fetal cardiac MRI T2 - Machine Learning for Biomedical Imaging VL - 2 IS - PIPPI 2022 special issue SP - 406 EP - 446 SN - 2766-905X DO - UR - ER -

2023:015 cover