Related Experiment Video
Updated: Jul 18, 2026
09:38
Following Endocardial Tissue Movements via Cell Photoconversion in the Zebrafish Embryo
Published on: February 20, 2018
6.5K
A Semi-supervised Four-Chamber Echocardiographic Video Segmentation Algorithm Based on Multilevel Edge Perception and
Yuexin Wan1, Dandan Li1, Zhi Li1
1State Key Laboratory of Public Big Data, College of Computer Science and Technology, Guizhou University, Guiyang, China.
Ultrasound in Medicine & Biology
|June 4, 2024
Summary
This study introduces a deep learning model for improved echocardiographic endocardium segmentation, enhancing cardiac function analysis and disease diagnosis accuracy. The model effectively addresses edge blurring and feature fusion challenges using multilevel edge perception and calibration fusion.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiology
Background:
- Automatic semantic segmentation of endocardium in echocardiographic videos is vital for cardiac function evaluation and heart disease diagnosis.
- Key challenges include edge blurring due to noise and ineffective multilevel feature fusion.
Purpose of the Study:
- To propose a deep learning model for enhanced endocardium segmentation in echocardiography.
- To address challenges of edge blurring and feature fusion for improved accuracy.
Main Methods:
- A deep learning model featuring a multilevel edge perception module for robust edge feature extraction.
- A calibration fusion module to integrate semantic and detailed features.
- Semi-supervised learning utilizing a memory architecture for labeled and unlabeled data.
Main Results:
- Achieved average Dice coefficients of 93.05% and 93.93% on public echocardiography datasets.
- Demonstrated a Pearson correlation of 0.765 for predicting left ventricular ejection fraction on clinical data.
Conclusions:
- The proposed semi-supervised model effectively overcomes segmentation challenges in echocardiography, improving ventricular segmentation accuracy.
- This model can aid cardiologists in achieving precise research and diagnostic outcomes.
