Automatic 3D segmentation of prostate in MRI combining a priori knowledge, Markov fields and Bayesian framework.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Radiology

Background:

• Accurate prostate localization and contouring are critical for prostate cancer diagnosis and therapy.
• Existing semi-automatic and automatic segmentation methods often require manual expert correction.
• There is a need for fully automated, accurate prostate segmentation techniques.

Purpose of the Study:

• To introduce an original automatic 3D segmentation method for the prostate gland using Magnetic Resonance Imaging (MRI) data.
• To evaluate the accuracy of the proposed method compared to expert manual segmentation.
• To assess the method's performance in cases with cancer and benign prostatic hypertrophy (BPH).

Main Methods:

• Utilized a statistical shape model for prior knowledge.
• Modeled gray level distribution using Gaussian mixture fitting of histogram modes.
• Employed Markov fields for contextual information and Iterative Conditional Mode (ICM) algorithm for Bayesian a posteriori classification.
• Developed an automatic method free from manual correction.

Main Results:

• Achieved a mean Hausdorff distance (HD) of 9.94 mm and an Overlap Ratio (OR) of 0.83 across 6 random cases.
• Demonstrated satisfying results, including at the prostate's base and apex.
• Offered fast computing times.

Conclusions:

• The developed automatic 3D prostate segmentation method shows promising results and accuracy.
• This technique can potentially reduce the reliance on manual expert correction in clinical workflows.
• The method is effective even for challenging cases at the prostate's base and apex.