High dimensional statistical shape model for medical image analysis

Statistical shape models have been widely used in biomedical image analysis, e.g. segmentation, registration, and shape classification. The traditional statistical shape models forced all shape parameters of each shape into one vector and put all vectors together to generate the point distribution model (PDM). The standard principal component analysis (PCA) was employed to project all shapes onto subspaces for dimensionality reduction. Since the shape vectors have a large dimension, the previous methods is computational expensive. In this paper, we propose a novel statistical shape models using natural PDM representations by multiple matrices and two dimensional PCA (2DPCA) is used to reduce the dimensionality of shape parameters. Because 2DPCA considers the correlations of row by row and column by column, our technique can fast extract the principle shape parameters. Combining with spherical harmonics shape representation, we create a framework for biomedical anatomic structures´ shape analysis and classification. The experimental results using real cardiac left ventricle shapes have demonstrated our method outperforms the previous statistical shape modeling.