AN OBJECTIVE COMPARISON OF 3-D IMAGE INTERPOLATION METHODS

To aid in the display, manipulation, and analysis of biomedical image data, they usually need to be converted to data of isotropic discretiz ation through the process of interpolation, Traditional techniques con sist of direct interpolation of the grey values [1], When user interac tion is called fur in image segmentation, as a consequence of these in terpolation methods, the user needs to segment a much greater (typical ly 4-10x) amount of data, To mitigate this problem, a method called sh ape-based interpolation of binary data was developed [2], Besides sign ificantly reducing user time, this method has been shown to provide mo re accurate results than grey-level interpolation [2]-[5], We proposed [6] an approach for the interpolation of grey data of arbitrary dimen sionality that generalized the shape-based method from binary to grey data, This method has characteristics similar to those of the binary s hape-based method. In particular, we showed preliminary evidence [6], [7] that it produced more accurate results than conventional grey-leve l interpolation methods. In this paper, concentrating on the three-dim ensional (3-D) interpolation problem, we compare statistically the acc uracy of eight different methods: nearest-neighbor; linear grey-level, grey-level cubic spline [8], grey-level modified cubic spline [9], Go shtasby et nl, [10], and three methods from the grey-level shape-based class [6], A population of patient magnetic resonance and computed to mography images, corresponding to different parts of the human anatomy , coming from different three-dimensional (3-D) imaging applications, are utilized for comparison. Each slice in these data sets is estimate d by each interpolation method and compared to the original slice at t he same location using three measures: mean-squared difference, number of sites of disagreement, and largest difference. The methods are sta tistically compared pairwise based on these measures. The shape-based methods statistically significantly outperformed all other methods in all measures in all applications considered here with a statistical re levance ranging from 10% to 32% (mean = 15%) for mean-squat-ed differe nce.