Validation of a Whole Heart Segmentation from Computed Tomography Imaging Using a Deep-Learning Approach

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Research

Background:

• Accurate whole heart segmentation (WHS) is crucial for cardiovascular assessment.
• Manual segmentation of ECG-gated CT data is time-consuming and subjective.
• Deep learning offers potential for automated and efficient medical image analysis.

Purpose of the Study:

• To develop and validate an automated deep-learning-based WHS method for ECG-gated CT data.
• To assess the accuracy and efficiency of the proposed deep learning algorithm.
• To evaluate the clinical applicability of automated WHS.

Main Methods:

• A dataset of 71 ECG-gated CT scans acquired before transcatheter aortic valve implantation was used.
• A fully automatic deep learning method combining two convolutional neural networks was employed for segmentation.
• The algorithm segmented 10 cardiovascular structures, with results compared against manual segmentation using the Dice index.

Main Results:

• The deep learning algorithm achieved high segmentation accuracy with a Dice score of 0.920 (IQR: 0.906-0.925).
• The method demonstrated a low computing time of 13.4 seconds per scan.
• Satisfactory correlations and agreement were found for myocardial volumes and mass, though right-sided structures require further improvement.

Conclusions:

• Automated WHS from ECG-gated CT data using deep learning is feasible with high accuracy.
• The developed algorithm shows promise for reducing segmentation time and improving consistency.
• Further refinement is needed, particularly for right-sided cardiac structures, to enable routine clinical application.