EchoEFNet: Multi-task deep learning network for automatic calculation of left ventricular ejection fraction in 2D...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Left ventricular ejection fraction (LVEF) is a key metric for assessing left ventricular systolic function.
• Current clinical calculation methods involving manual segmentation are prone to errors and lack reproducibility.
• Accurate and reproducible LVEF assessment is crucial for patient diagnosis and management.

Purpose of the Study:

• To develop and validate a novel multi-task deep learning network, EchoEFNet, for automated LVEF calculation.
• To improve the accuracy and reproducibility of LVEF measurement compared to traditional methods.
• To simultaneously segment the left ventricle and detect critical landmarks for LVEF calculation.

Main Methods:

• A multi-task deep learning network, EchoEFNet, utilizing ResNet50 with dilated convolution as its backbone.
• A multi-scale feature fusion decoder for simultaneous segmentation and landmark detection.
• Automated LVEF calculation using the biplane Simpson's method based on network outputs.
• Model validation on the public CAMUS dataset and a private CMUEcho dataset.

Main Results:

• EchoEFNet demonstrated superior performance in geometrical metrics and keypoint detection compared to other deep learning methods.
• High correlation coefficients between predicted and true LVEF values were observed: 0.854 on CAMUS and 0.916 on CMUEcho.
• The proposed method significantly enhances the accuracy and reproducibility of LVEF assessment.

Conclusions:

• EchoEFNet offers a robust and automated solution for LVEF calculation from echocardiographic images.
• The deep learning approach addresses the limitations of manual segmentation, improving clinical workflow efficiency and diagnostic reliability.
• This technology has the potential to standardize LVEF assessment in clinical practice.