%0 Journal Article
%T 基于卷积神经网络的心脏射血分数评估研究
Research on the Evaluation of Cardiac Ejection Fraction Based on Convolutional Neural Network
%A 高媛
%A 宋金超
%J Modeling and Simulation
%P 649-655
%@ 2324-870X
%D 2025
%I Hans Publishing
%R 10.12677/mos.2025.144317
%X 心血管疾病是全球范围内的重要死亡原因，涵盖了一系列涉及心脏和血管的疾病。射血分数作为临床检查中的关键指标，其变化与心力衰竭、心肌梗死和心脏纤维化等常见心血管疾病密切相关。本研究旨在通过开发一种高效准确的计算方法，协助临床医生及时诊断这些疾病。为此，本文基于EchoNet-Dynamic数据集，提出了一种新的半监督深度学习网络EF-Net。该网络结合了U-Net架构、深度监督和注意力模块，通过半监督学习增加训练样本，从而增强了模型的图像分割能力。与现有方法相比，EF-Net在数据集的大多数评估标准上均表现出显著提升。研究结果表明，该方法显著提高了心脏超声成像中射血分数的计算精度，展示了其在临床诊断中的潜在应用价值。
Cardiovascular diseases are a leading cause of death globally, encompassing a range of conditions that affect the heart and blood vessels. Ejection fraction, a key metric in clinical examinations, is closely associated with common cardiovascular diseases such as heart failure, myocardial infarction, and cardiac fibrosis. This study aims to assist clinicians in timely diagnosis by developing an efficient and accurate computational method. To this end, we propose a novel semi-supervised deep learning network, EF-Net, based on the EchoNet-Dynamic dataset. This network integrates the U-Net architecture, deep supervision, and attention mechanisms, enhancing the model's image segmentation capabilities through semi-supervised learning with additional training samples. Compared to existing methods, EF-Net demonstrates significant improvements across most evaluation metrics on the dataset. The results indicate that this approach significantly enhances the accuracy of ejection fraction calculation in cardiac ultrasound imaging, underscoring its potential clinical diagnostic value.to the Hans standard, which illustrates all the formats.
%K 半监督学习，
%K 射血分数，
%K 超声图像，
%K 图像分割
Semi-Supervised Learning
%K Ejection Fraction
%K Ultrasound Images
%K Image Segmentation
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=112194