Differences between algorithms for spatial interpolation of 3D echocardiography images

In 3D, ultrasound systems sample a conical volume of space with a given angular resolution, resulting in sparsely sampled regions far from the transducer. Image analysis operators require that each voxel of a cartesian volume has a value, and some form of interpolation is necessary. 3D images of the left ventricle (LV) of 10 patients were constructed from sequences of conventional B-mode frames which were successively rotated about the beam axis. Three different methods of interpolation were used to assign values to unsampled voxels. Reconstructed LV short axis views were extracted from the 3D images and analysed using standard edge detection and threshold operators. Results from the image analysis were used to compare the three different methods of interpolation. Results of this preliminary study show that the choice of interpolation method influences the results of image analysis operations.