To main content

Left-Ventricular Volume Estimation in Contrast-Enhanced Echocardiography Using Deep Learning

Abstract

Accurate estimation of left ventricular (LV) volume is important for the diagnosis and management of cardiac disease. Contrast-enhanced ultrasound (CEUS) helps improve the visualization of the endocardial borders and is essential for patients with poor image quality. This study aims to develop automated CEUS segmentation and volume estimation, evaluate the inter-observer variability, and compare its accuracy to automated B-mode segmentation for patients with suboptimal image quality.We included N=105 patients (492 contrast images) with diverse cardiac conditions to develop a U-Net-based CNN for LV segmentation. We evaluated the inter-observer variability between two annotators, and with an existing B-mode pipeline for reference, we evaluated LV volume for both automated B-mode and CEUS towards the manual CEUS reference for the same patients.Results showed good accuracy in LV segmentation and volume estimation on CEUS images, with an average Dice score of 0.92±0.04, similar to the inter-observer variability at 0.92±0.03. Compared to manual volume measurements, our automated approach had an average bias of –10.0 mL (-7.0%) and a standard deviation of 17.6 mL (17.3%). A significantly higher standard deviation (26.6 mL, 27.1%) was found for automated B-mode measurements, mainly due to indistinct borders and subpar segmentation.Our study demonstrates the potential of LV volume estimation using contrast echocardiography images.
Read the publication

Category

Academic chapter

Language

English

Author(s)

  • Jieyu Hu
  • Erik Smistad
  • Bjørnar Leangen Grenne
  • Espen Holte
  • Håvard Dalen
  • Lasse Løvstakken

Affiliation

  • SINTEF Digital / Health Research
  • St. Olavs Hospital, Trondheim University Hospital
  • Norwegian University of Science and Technology

Year

2023

Publisher

IEEE (Institute of Electrical and Electronics Engineers)

Book

Proceedings of the 2023 IEEE International Ultrasonics Symposium (IUS)

ISBN

9798350346459

View this publication at Norwegian Research Information Repository