Automatic bladder segmentation on CBCT for multiple plan ART of bladder cancer using a patient-specific bladder model

Area of Science:

• Radiation Oncology
• Medical Imaging
• Computational Anatomy

Background:

• Adaptive radiotherapy (ART) for bladder cancer relies on accurate daily imaging to adjust treatment plans.
• Current methods for bladder segmentation on cone-beam CT (CBCT) can be time-consuming and operator-dependent.
• Efficient and accurate bladder segmentation is crucial for selecting the optimal treatment plan in ART.

Purpose of the Study:

• To develop and validate an automatic bladder segmentation technique for CBCT images.
• To assess the performance of this automatic method in selecting the correct treatment plan for bladder cancer ART.
• To leverage patient-specific deformation models for robust bladder contouring.

Main Methods:

• Developed a two-step automatic segmentation approach using patient-specific principal component analysis (PCA) models.
• Trained PCA models on planning CT and early CBCT scans to capture bladder deformation patterns.
• Validated segmentation accuracy using volume and distance metrics, and evaluated plan selection agreement with manual delineations.

Main Results:

• The automatic segmentation achieved a 78.5% mean conformity index compared to manual delineations.
• The mean standard deviation of local residual error was 0.24 cm.
• Plan selection agreement between automatic and manual methods was 77.5%.

Conclusions:

• Principal component analysis (PCA) effectively models patient-specific bladder deformations for ART.
• The statistical-shape-based segmentation is robust for low-quality CBCT images.
• This automated approach enables fast and reliable bladder segmentation on CBCT, facilitating accurate plan selection in bladder cancer ART.