Machine Learning with Scarce Data: Ejection Fraction Prediction Using PLAX View

Zhiyuan Gao, Dominic Yurk, Yaser S. Abu-Mostafa

Keywords: Scarce data, Ejection fraction, Echocardiography, Parasternal Long-Axis (PLAX), Video View Classification, Proxy Labeling, PhysioNet MIMIC Dataset

TL;DR: We leveraged scarce public echo data to develop a large PLAX dataset with proxy EF labels, achieving a 6.86% MAE for EF prediction from PLAX views, setting the first reproducible benchmark for EF estimation when apical views aren’t available.

Abstract: We developed a machine learning model to predict left ventricular ejection fraction (LVEF/EF) from parasternal long-axis (PLAX) echocardiographic videos. Because public datasets with labeled PLAX videos are virtually non-existent, our work focuses on an innovative data generation strategy to overcome this scarcity. By leveraging a time-based correlation between clinical notes and echocardiographic videos, combined with fine-tuning view classifiers and proxy labeling, we effectively created a large labeled PLAX dataset and achieved a mean absolute error (MAE) of 6.86%. Given that Apical four-chamber methods, the clinical standard, report MAE values of 6%-7%, our results demonstrate that EF estimation from PLAX views is both feasible and clinically relevant. This surpasses the performance of existing methods and provides a clinically useful solution for situations where apical views may not be feasible. The EF labels for PLAX videos, derived from publicly available datasets, are accessible at https://github.com/Jeffrey4899/PLAX_EF_Labels_202501.

Primary Subject Area: Learning with Noisy Labels and Limited Data

Secondary Subject Area: Application: Cardiology

Paper Type: Both

Registration Requirement: Yes

Reproducibility: https://github.com/Jeffrey4899/PLAX_EF_Labels_202501

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., All authors and co-authors are correctly listed with proper spelling and avoid Unicode characters., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.

Submission Number: 205