3-D Medical Image Interpolation Based on Context Classification

Abstract

<p>This thesis is concerned with interpolation along the Z-axis for application in medical imaging and increasing out-of plane resolution of 3-D medical image sets. Interpolation along the Z-axis is an essential task in clinical studies for better diagnosis and analysis of body organs and their functions. It is also necessary when sets of images with different out-of plane resolutions should be analyzed together.</p> <p>The first part of the thesis discusses a 3-D interpolation method based on a piece-wise autoregressive model that has been already proven to be efficient for 2-D image interpolation. The 3-D image set is modeled as a 3-D piece-wise autoregressive model and the model parameters are estimated within a cube that slides through the low resolution image set.</p> <p>The major part of this thesis is devoted to a new interpolation algorithm, called contextbased 3-D interpolation. The proposed method represents a new approach of aiding 3-D interpolation and improving its performance by efficient use of domain knowledge about the anatomy, orientation and imaging modalities. In the new approach a family of adaptive 3-D interpolation filters are designed and conditioned on different spatial contexts (classes of feature vectors). Training is used to incorporate the domain knowledge into the design of these interpolators. Experimental results show significant improvement of the new approach over some existing 3D interpolation techniques.</p>