Automatic cardiac evaluations using a deep video object segmentation network.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate cardiac volume and function assessment are crucial for diagnosing ventricular dysfunction and cardiovascular disease.
• Current methods for cardiac measurements can be challenging, necessitating reliable AI-assisted tools.
• Deep neural networks offer potential for enhancing the accuracy and reliability of echocardiographic analyses.

Purpose of the Study:

• To develop and evaluate EchoRCNN, a semi-automated deep neural network for echocardiography sequence segmentation.
• To improve the accuracy of cardiac measurements, specifically left and right ventricle segmentation.
• To provide physicians with a reliable tool for more accurate cardiac function diagnosis.

Main Methods:

• Utilized a semi-automated deep neural network named EchoRCNN.
• Combined mask region-based convolutional neural network (R-CNN) for image segmentation.
• Integrated reference-guided mask propagation for video object segmentation.

Main Results:

• Achieved high accuracy in segmenting left and right ventricle regions with Dice similarity coefficients of 94.03% and 94.97%, respectively.
• Post-processing of segmented regions yielded low mean absolute errors for ejection fraction (3.13%) and fractional area change (2.03%).
• Demonstrated excellent performance in ventricle segmentation and cardiac parameter estimation.

Conclusions:

• EchoRCNN provides excellent performance for left and right ventricle segmentation in echocardiography.
• The method enables more accurate estimations of vital cardiac parameters like ejection fraction and fractional area change.
• The findings suggest EchoRCNN can support assured, accurate, and reliable cardiac function diagnosis in clinical settings.