Cardiac-DeepIED: Automatic Pixel-Level Deep Segmentation for Cardiac Bi-Ventricle Using Improved End-to-End...

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate cardiac bi-ventricle segmentation is crucial for cardiovascular function assessment.
• The complexity of cardiac structures presents a significant challenge for automated segmentation in MR images.

Purpose of the Study:

• To develop an improved end-to-end encoder-decoder network for pixel-level cardiac bi-ventricle segmentation.
• To enhance the accuracy and robustness of automated cardiac segmentation from MR images.

Main Methods:

• Proposed Cardiac-DeepIED framework utilizing an improved encoder-decoder architecture with Fire and D-Fire dilated modules.
• Incorporated a convolutional long-term and short-term memory structure to capture spatiotemporal correlations in cardiac MR sequences.
• Evaluated on 145 clinical subjects using leave-one-out cross-validation.

Main Results:

• Achieved high average Dice metrics: 0.96 for the left ventricle, 0.89 for the myocardium, and 0.903 for the right ventricle.
• Demonstrated superior performance compared to state-of-the-art methods in cardiac bi-ventricle segmentation.
• Validated the effectiveness and advantages of the proposed method for pixel-level segmentation.

Conclusions:

• The Cardiac-DeepIED system provides accurate and robust segmentation of cardiac bi-ventricles, outperforming existing methods.
• The automated system can be integrated into clinical settings to expedite cardiovascular quantification and analysis.
• Potential for expansion to various analyses including volume, wall thickness, and three-dimensional measurements.