Automatic prostate MR image segmentation with sparse label propagation and domain-specific manifold regularization.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Accurate prostate segmentation in MR images is crucial for prostate cancer diagnosis.
• Challenges include significant inter-subject shape variations and inhomogeneous appearance.
• Existing methods struggle with these inherent difficulties.

Purpose of the Study:

• To develop a hierarchical prostate MR segmentation method addressing shape variations and appearance inhomogeneity.
• To improve the accuracy and reliability of automatic prostate segmentation in medical imaging.

Main Methods:

• Learned salient features using subclass discriminant analysis (SDA) for anatomical signatures.
• Employed a multi-atlas sparse label fusion framework for coarse-level prostate likelihood estimation.
• Utilized domain-specific semi-supervised manifold regularization for fine-level segmentation refinement.

Main Results:

• The proposed hierarchical method achieved consistently higher segmentation accuracies compared to two state-of-the-art methods.
• Evaluated on a T2-weighted prostate MR image dataset of 66 patients.
• Demonstrated superior performance in handling prostate shape variations and appearance inconsistencies.

Conclusions:

• The novel hierarchical segmentation method effectively overcomes key challenges in prostate MR imaging.
• This approach offers improved accuracy for computer-aided diagnosis of prostate cancer.
• The method shows significant potential for clinical application in prostate cancer detection and management.