Quantifying Biological Variation
Special Symposium - Innovations in Food Technology (LMC Congress)
Modern Analysis: Chemical & Multivariate Analysis (Food-6a)
Keywords: Image analysis, computed tomography, 3D statistical shape model, biological variation
Statistical shape models are often used to encapsulate complex biological variation in medical applications. They can be used for obtaining a better understanding of a shape, as prior knowledge in automated segmentation schemes or as discriminating features in classification tasks. In order to do statistics on shapes corresponding points are required, which are not feasible to acquire manually in 3D.
In this work dense 3D surfaces of bones are estimated from anisotropic voxels using Radial Basis Functions. The data consists of 2D CT scans of 40 porcine carcasses separated along the medial plane. Each scan has a slice thickness of 10mm with a spacing of 10mm between each scan. Voxel dimensions are [x, y, z] = [0.88, 0.88, 10]mm. The full length of the carcasses is scanned resulting in approximately 130 scans per carcass, but only 30 scans per carcass are used in this application, covering the parts around the region of the pelvic bone. Using an extended iterative closest point algorithm a rigid registration is obtained and a compact statistical shape model is built. The model quantifies the biological variation of the bone structures in such a way that the modes are localized and their interpretation is simpler. Furthermore it is investigated how well the shape model can predict the weight of specific parts of the carcass obtained by manually dissection.
Presented Thursday 20, 09:20 to 09:35, in session Modern Analysis: Chemical & Multivariate Analysis (Food-6a).
