A unified EM approach to bladder wall segmentation with coupled level-set constraints

Area of Science:

• Medical Imaging
• Oncology
• Computer Vision

Background:

• Magnetic resonance imaging-based virtual cystoscopy (VCys) is crucial for bladder cancer diagnosis and recurrence monitoring.
• Accurate segmentation of bladder wall thickness is essential for identifying abnormalities.

Purpose of the Study:

• To develop a unified expectation-maximization (EM) approach for precise bladder wall segmentation.
• To integrate an adaptive Markov random field (AMRF) model and coupled level-set (CLS) information for improved segmentation accuracy.

Main Methods:

• Proposed a unified EM-MAP approach incorporating AMRF and CLS models for bladder wall segmentation.
• Applied the method to T(1)-weighted MR images, enhancing wall visibility.
• Introduced scale-adaptive neighborhoods and adaptive weights in the AMRF model.
• Utilized CLS for geometrical constraints to mitigate artifacts and preserve wall continuity.

Main Results:

• The AMRF model accurately incorporates local image information.
• The combined approach demonstrated superior bladder wall extraction compared to previous CLS methods.
• Evaluated robustness using simulated phantoms and compared performance in volunteer and patient studies.

Conclusions:

• The novel segmentation scheme effectively extracts bladder walls, enabling accurate thickness measurement.
• Constructed 3D virtual bladder models with mapped wall thickness for improved bladder lesion detection.