Outlier detection and handling for robust 3-D active shape models search.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Volumetric segmentation using 3-D active shape models (ASMs) is crucial for medical image analysis.
• Outliers in feature points can significantly degrade the accuracy and convergence of ASM-based segmentation.
• Existing methods often struggle with a high density of erroneous feature points.

Purpose of the Study:

• To develop a robust outlier handling method for 3-D ASMs.
• To improve the accuracy and convergence of volumetric segmentation frameworks.
• To address the challenge of significant erroneous feature points in segmentation tasks.

Main Methods:

• A novel shape metric invariant to scaling, rotation, and translation using interlandmark distance ratios.
• Statistical tolerance intervals derived from training data to identify outlier feature points.
• A replacement strategy for outliers based on a statistical tolerance model and valid point positions.
• Introduction of a geometrically weighted fitness measure for feature point detection.

Main Results:

• The proposed method effectively handles outliers, regardless of their extremity or quantity.
• Improved convergence and robustness of the 3-D active shape model segmentation framework.
• Successful validation in 3-D magnetic resonance (MR) segmentation of carotid arteries and left ventricle myocardial borders.

Conclusions:

• The new outlier handling method significantly enhances the reliability of 3-D ASM volumetric segmentation.
• The technique offers a robust solution for dealing with erroneous feature points in complex medical imaging scenarios.
• Demonstrated practical utility in segmenting critical anatomical structures from 3-D MR images.