Segmenting the prostate and rectum in CT imagery using anatomical constraints

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Accurate segmentation of prostate and rectum in 3D CT images is crucial for image-guided cancer therapy.
• Existing automatic segmentation methods face challenges in precision and reliability.

Purpose of the Study:

• To develop a novel automatic segmentation algorithm for prostate and rectum using deformable organ models.
• To enhance segmentation accuracy by incorporating anatomical constraints and appearance models.

Main Methods:

• A new segmentation cost function utilizing a Bayesian framework with anatomical constraints from surrounding bones.
• A novel appearance model learning nonparametric intensity histogram distributions inside and outside organ contours.
• Deformable organ models trained on previously segmented data.

Main Results:

• The algorithm demonstrated improved segmentation performance on 185 prostate site datasets.
• The proposed cost function and appearance model contributed to enhanced accuracy.
• The method showed superior results compared to previous segmentation models.

Conclusions:

• The developed automatic segmentation algorithm offers improved accuracy for prostate and rectum.
• This advancement is critical for enhancing the effectiveness of image-guided therapy.
• The integration of anatomical and appearance models represents a significant step forward in medical image segmentation.