Automatic segmentation of multiple cardiovascular structures from cardiac computed tomography angiography images...

Area of Science:

• Cardiovascular Imaging
• Deep Learning in Medical Imaging
• Medical Image Segmentation

Background:

• Manual segmentation of cardiovascular images is time-consuming and labor-intensive.
• Automated methods are needed to improve efficiency and consistency in cardiovascular image analysis.
• Coronary Computed Tomography Angiography (CCTA) provides detailed anatomical information crucial for diagnosis.

Purpose of the Study:

• To develop and evaluate an automated deep learning method for segmenting multiple cardiovascular structures.
• To assess the accuracy of the deep learning model in segmenting various cardiac components from CCTA images.
• To provide a more efficient and reliable tool for cardiovascular image analysis.

Main Methods:

• A U-net-derived deep learning model was designed for automated segmentation.
• The model was trained and validated on CCTA images from a multicenter registry (70:20:10 split).
• Segmentation accuracy was evaluated using the Dice score for multiple cardiovascular structures including aorta, vena cavae, pulmonary artery, coronary sinus, and atrial/ventricular walls.

Main Results:

• The deep learning model achieved an overall median Dice score of 0.820 across all segmented structures.
• High median Dice scores were observed for the proximal ascending aorta (0.969) and descending aorta (0.953).
• Performance was consistent across different sexes and age groups, with the exception of the coronary sinus.

Conclusions:

• An automated deep learning model successfully segmented multiple cardiovascular structures from CCTA images.
• The model demonstrated reasonable overall accuracy at the pixel level, indicating its potential clinical utility.
• This automated approach offers a promising solution to the resource-intensive nature of manual cardiovascular image segmentation.