A novel robust kernel principal component analysis for nonlinear statistical shape modeling from erroneous data.

Area of Science:

• Medical image analysis
• Computational anatomy
• Machine learning for medical imaging

Background:

• Statistical Shape Models (SSMs) are crucial for medical image segmentation.
• Existing SSMs struggle with nonlinear shape distributions and noisy training data.
• Degradation in training data quality compromises SSM performance.

Purpose of the Study:

• To develop a generic SSM capable of modeling nonlinear shape distributions.
• To create an SSM robust to outliers and corrupted training data.
• To enhance the accuracy and reliability of medical image segmentation.

Main Methods:

• Proposed a novel Robust Kernel Principal Component Analysis (RKPCA) for statistical shape modeling.
• Constructed a low-rank nonlinear subspace to effectively discard outliers.
• Assumed sparsity in nonlinear distribution for improved modeling.

Main Results:

• The RKPCA-based SSM demonstrated superior performance in outlier recovery.
• Achieved a higher quality statistical shape model compared to existing methods.
• Resulted in significantly lower segmentation errors on CT kidney and MRI ankle bone datasets.

Conclusions:

• The proposed RKPCA method effectively addresses limitations of standard SSMs.
• This approach enhances robustness against noisy annotations in medical imaging.
• The developed SSM offers improved accuracy for anatomical structure segmentation.