Semiautomatic bladder segmentation on CBCT using a population-based model for multiple-plan ART of bladder cancer

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Anatomy

Background:

• Adaptive radiotherapy (ART) requires accurate patient anatomy delineation for treatment planning.
• Current bladder segmentation methods can be time-consuming and may lack consistency.
• Efficient bladder segmentation is crucial for optimizing multiple-plan ART strategies.

Purpose of the Study:

• To develop a novel semiautomatic bladder segmentation approach for multiple-plan ART.
• To utilize a statistical shape model derived from principal component analysis (PCA) for bladder segmentation.
• To evaluate the performance and efficiency of the proposed segmentation method.

Main Methods:

• A population-based statistical bladder model was created using spherical harmonic expansion and PCA on training data.
• The model was applied to segment bladders in validation CBCT scans, deforming a prior contour to fit the boundary.
• A cost function and simplex optimizer were used for automatic segmentation, followed by optional manual correction.

Main Results:

• The statistical shape model effectively captured bladder shape variations using seven PCA modes.
• Automatic segmentation achieved a mean Dice similarity coefficient of 70.5% and residual error of 0.39 cm.
• Semiautomatic segmentation, with manual correction, significantly improved results to 77.7% Dice and 0.30 cm error, enhancing plan selection agreement to 80.7%.

Conclusions:

• Statistical shape-based segmentation offers a viable method for automatic bladder delineation on CBCT with moderate accuracy.
• Limited manual intervention can substantially enhance segmentation accuracy and reliability.
• This approach is suitable for selecting appropriate treatment plans in multiple-plan ART for bladder cancer.