Automated Multi-class Fetal Cardiac Vessel Segmentation in Aortic Arch Anomalies Using T2-Weighted 3D Fetal MRI

Gilliland, Paula Ramirez; Uus, Alena; Poppel, Milou P. M. van; Grigorescu, Irina; Steinweg, Johannes K.; Lloyd, David; Pushparajah, Kuberan; King, Andrew P.; Deprez, Maria

doi:10.1007/978-3-031-17117-8_8

Cited by 2 publications

(5 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We expand on our prior research (Ramirez Gilliland et al, 2022) by incorporation of an anomaly classifier which improves segmentation discernment between anomalies. We also include detailed ablation studies to validate individual elements of our multi-task framework, which now consists of training using labels propagated from the anomaly-specific atlases and manual labels in individual images, while simultaneously classifying the anomaly from the predicted segmentation.…”

Section: Contributionsmentioning

confidence: 99%

“…Our benchmark framework is based on Ramirez Gilliland et al (2022), where Attention U-Net (Oktay et al, 2018) is trained using both manual binary labels and multi-class labels propagated from an atlas (Uus et al, 2022b).…”

Section: Multi-task Segmentation Frameworkmentioning

confidence: 99%

“…As presented in Ramirez Gilliland et al (2022) we conduct ablation studies on the inclusion of each type of labelling information, e.g. manual binary labels in individual subject images and multi-class labels propagated from the anomaly-specific atlases.…”

Section: Training Label Type Ablation Studymentioning

confidence: 99%

“…Expanding on our previous work (Ramirez Gilliland et al, 2022), we repeat our segmentation experiments three times, to account for network stochasticity.…”

Section: Training Label Type Ablation Studymentioning

confidence: 99%

“…3.1). Differences to Ramirez Gilliland et al (2022) are due to adaptations made to the atlas in pulmonary arteries and exclusion of pulmonary veins (following discussions with clinicians), and stochasticity due to network retraining. All metrics displayed are on unseen test set, comparing to our manually generated GT.…”

Section: Training Label Type Ablation Studymentioning

confidence: 99%

See 4 more Smart Citations

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

Ramirez,

Uus,

van Poppel

et al. 2023

Melba

View full text Add to dashboard Cite

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.<br>Our code is available at <a href='https://github.com/SVRTK/MASC-multi-task-segmentation-and-classification'>https://github.com/SVRTK/MASC-multi-task-segmentation-and-classification</a>

show abstract

Section: Contributionsmentioning

confidence: 99%

Section: Multi-task Segmentation Frameworkmentioning

confidence: 99%

Section: Training Label Type Ablation Studymentioning

confidence: 99%

“…Expanding on our previous work (Ramirez Gilliland et al, 2022), we repeat our segmentation experiments three times, to account for network stochasticity.…”

Section: Training Label Type Ablation Studymentioning

confidence: 99%

Section: Training Label Type Ablation Studymentioning

confidence: 99%

See 3 more Smart Citations