Automatic thoracic anatomy segmentation on CT images using hierarchical fuzzy models and registration

Area of Science:

• Medical Image Analysis
• Computational Anatomy
• Radiology

Background:

• Automatic organ segmentation is crucial for medical image analysis.
• Previous automatic anatomy recognition (AAR) methods achieved speed and accuracy without registration.
• Existing AAR approaches faced challenges in segmentation accuracy.

Purpose of the Study:

• To investigate the influence of image/object registration on the AAR methodology.
• To improve the accuracy of organ recognition and delineation by integrating registration.
• To refine the AAR approach by tightly coupling recognition and delineation steps.

Main Methods:

• Developed a 3D fuzzy set model for each organ by registering labeled binary images.
• Incorporated hierarchical and spatial relationships between organs into the model.
• Employed hierarchical affine registration of fuzzy shape models to target images for recognition.
• Utilized fuzzy connectedness delineation with seed points from recognition for final segmentation.

Main Results:

• Achieved high delineation accuracy for eight thoracic organs on 30 real images.
• Mean false positive and false negative volume fractions were 0.34% and 4.02% for nonsparse organs.
• Mean false positive and false negative volume fractions were 0.16% and 12.6% for sparse organs.
• Mean boundary distances to ground truth were 1.31 mm for nonsparse and 2.28 mm for sparse objects.

Conclusions:

• Hierarchical structure and location relations enhance registration efficiency and robustness.
• 3D fuzzy models combined with hierarchical affine registration ensure accurate recognition of diverse organ types.
• Organ-specific registration criteria and refined intensity properties improve overall segmentation performance.