JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Automatic Segmentation of the Left Ventricle in Apical Four-Chamber View on Transesophageal Echocardiography Based on UNeXt Deep Neural Network

  • 0Department of Anesthesiology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100730, China.
Diagnostics (basel, Switzerland) +

|

December 17, 2024

|

December 17, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces UNeXt for automatic left ventricle segmentation in transesophageal echocardiography (TEE) apical four-chamber views (A4CV), achieving superior accuracy and efficiency. The developed large-scale TEE A4CV dataset supports this advancement in cardiac imaging analysis.

Area Of Science

  • Medical Imaging
  • Artificial Intelligence in Medicine
  • Cardiovascular Imaging

Background

  • Automatic left ventricle segmentation in transesophageal echocardiography (TEE) is crucial for cardiac assessment.
  • Existing methods require improvement in accuracy and computational efficiency.
  • A large-scale, annotated dataset for TEE apical four-chamber view (A4CV) segmentation is needed.

Purpose Of The Study

  • To construct a large-scale TEE A4CV image dataset.
  • To propose and evaluate an automatic left ventricle segmentation method using the UNeXt deep neural network for TEE A4CV.
  • To compare UNeXt's performance against U-Net, TransUNet, and Attention U-Net.

Main Methods

  • Development of a TEE A4CV dataset from 60 cardiac surgery patients, comprising 3000 images.
  • Implementation of the UNeXt deep neural network, a U-Net variant with a multilayer perceptron, for segmentation.
  • Comparative analysis using Dice Similarity Coefficient (DSC) and Intersection over Union (IoU) metrics, with Kruskal-Wallis test for significance.

Main Results

  • UNeXt achieved the highest average DSC (88.60%) and IoU (77.60%), outperforming U-Net, TransUNet, and Attention U-Net.
  • UNeXt demonstrated significantly better performance than TransUNet and Attention U-Net (p < 0.05).
  • UNeXt exhibited reduced parameters, computational complexity, and faster inference times compared to other models.

Conclusions

  • This study presents the first large-scale TEE A4CV dataset and validates UNeXt for left ventricle segmentation.
  • The UNeXt-based method offers a promising solution for accurate and efficient automatic segmentation in TEE A4CV.
  • The developed dataset and model contribute to advancing automated cardiac image analysis.

Keywords: