A fully automatic multi-atlas based segmentation method for prostate MR images.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Multi-atlas segmentation methods often struggle with local accuracy in prostate MRI due to field-of-view differences and anatomical variability.
• Global registration may not precisely capture the prostate region, limiting segmentation accuracy.

Purpose of the Study:

• To develop a fully automatic, two-stage multi-atlas framework for improved prostate segmentation in MRI.
• To enhance the accuracy of prostate segmentation by focusing registration on the local prostate region.

Main Methods:

• A two-stage approach: first, detecting a bounding cuboid around the prostate using atlas edge registration and edge detection.
• Second, refining segmentation by focusing registration within the detected prostate vicinity.

Main Results:

• The proposed method achieved a Dice Similarity Coefficient (DSC) of 83.4%±4.3% and a Hausdorff Distance (HD) of 9.3 mm±2.6 mm.
• Evaluation on 12 patient MRI volumes using a leave-one-out study demonstrated the method's effectiveness.

Conclusions:

• The developed two-stage automated method significantly improves prostate segmentation accuracy compared to traditional full-volume registration.
• This automated segmentation tool holds promise for various prostate imaging applications.