Cardiac atlas development and validation for automatic segmentation of cardiac substructures

Area of Science:

• Medical imaging
• Radiotherapy
• Computational anatomy

Background:

• Accurate delineation of cardiac substructures is crucial for radiotherapy planning.
• Manual contouring is time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate a set of cardiac atlases for automated contouring of cardiac substructures.
• To improve the efficiency and consistency of cardiac substructure delineation in CT imaging.

Main Methods:

• Eight radiation oncologists manually delineated 15 cardiac substructures from noncontrast CT images.
• Atlases were created by fusing and editing contours from 6 patients.
• Multi-atlas segmentation was applied to delineate contours in other patients.

Main Results:

• Auto-segmented contours showed agreement within one standard deviation of expert variability for Dice Similarity Coefficient and Mean Surface Distance.
• Good agreement was observed for the heart, chambers, and great vessels.
• Independent validation on 19 patients demonstrated reasonable agreement for heart chambers.

Conclusions:

• A validated set of cardiac atlases for auto-contouring from noncontrast CT images was successfully created.
• The accuracy of auto-contouring for key cardiac structures supports its potential clinical application.