264 - Automated Measurement of Left Atrium to Aortic Root Ratio in Targeted Neonatal Echocardiograms

Publication Number: 4260.264

Daniel Dagle, University of Iowa Stead Family Children's Hospital, Iowa City, IA, United States; Danielle R. Rios, University of Iowa Stead Family Children's Hospital, Iowa City, IA, United States; Craig Rusin, Baylor College of Medicine, Houston, TX, United States; Numra A. Aleem, University of Iowa Stead Family Children's Hospital, Iowa City, IA, United States; Amy HC. Stanford, University of Iowa Stead Family Children's Hospital, Iowa City, IA, United States

Background: Patent ductus arteriosus (PDA) carries an increased risk of complications in preterm neonates. Targeted neonatal echocardiography (TnECHO) is the gold standard method of diagnosing PDA. The University of Iowa uses a seven-element scoring system to determine the hemodynamic significance of a PDA, giving more a granular assessment of risk to peripheral organs. Many elements of the Iowa PDA score are measured from the TnECHO images. This requires additional time from a physician trained in neonatal hemodynamics. Even with hemodynamics training, there is still variation in measurements between physicians.

Objective: The goal of this work is to create an automated image processing pipeline to measure the left atrium (LA) to aortic root (Ao) diameter ratio (LA/Ao), one of the elements of the Iowa PDA score.

Design/Methods: An image processing pipeline was constructed to locate the positions of M-mode TnECHO images corresponding to the walls of the left atrium and aortic root. These measurements were compared to the clinical physician measurements using Bland-Altman analysis.

Results: 200 M-mode TnECHO images were retrospectively retrieved from 74 patients at the University of Iowa from May through September 2025. 25% of the images were held out for validation. The left atrium measurements achieved a non-parametric reproducibility coefficient (RPC) of 0.32 for the training set and 0.33 for the validation set with RMSE of 0.18 and 0.13 respectively. The aortic root measurements achieved RPCs of 0.22 for the training set and 0.17 for the validation set with RMSE of 0.26 and 0.28 respectively. The LA/Ao ratio achieved RPCs of 0.53 for the training set and 0.74 for the validation set with RMSE of 0.18 and 0.13 respectively.

Conclusion(s): This image processing pipeline enables automated LA/Ao measurement with reasonable accuracy compared to human measurements. This algorithm is the first step in automating the calculation of the Iowa PDA score.

Computer Annotated TnECHO
M-mode TnECHO with computer annotations for left atrium (red) and aortic root diameter (yellow).

LA/Ao Ratio Validation Performance
Parity and Bland-Altman plots of LA/Ao ratio for the validation set.