Registration Based Medical Image Interpolation

A new technique is presented for interpolating between grey-scale images in a medical data set. Registration between neighboring slices is achieved with a modified control grid interpolation algorithm that selectively accepts displacement field updates in a manner optimized for performance. A cubic interpolator is then applied to pixel intensities correlated by the displacement fields. Special considerations are made for efficiency, interpolation quality, and compression in the implementation of the algorithm. Experimental results show that the new method achieves good quality, while offering dramatic improvement in efficiency relative to the best competing method.

THREE-DIMENSIONAL (3-D) medical imaging modalities often present acquired data as a set of slices. The thickness of each slice is usually significantly greater than the distance between voxel centers within an imaging plane, resulting in a data set composed of voxels that are anisotropic. In many applications that deal with 3-D data, it is desirable to have voxel dimensions that are isotropic or nearly so. These include multi-planar reconstruction (MPR), maximum intensity projection, and shaded surface rendering to name a few. Here, a novel interpolation technique for increasing the out-of-plane resolution of medical image data sets, analogous to decreasing slice thickness, is presented. The new methodology performs well in comparison to other state-of-the-art techniques based on quality and offers significant advantages in terms of both computational requirements and ease of implementation in commercial applications.
Interpolation methods for this type of problem can be generally classified as being scene-based or object-based [1]. Scenebased approaches use uniform registration, customarily interpolating between intensity values that are correlated based on their matrix locations within respective images. This category includes many well-known methods such as linear, spline, and truncated sinc function interpolation [2]. Several enhancements along these lines have also been proposed that make use of more complex convolution kernels [3], [4]. Scene-based techniques are efficient and easily implemented, but can produce signifi- cant artifacts when pixels that occupy the same matrix location in contiguous images belong to different anatomical structures. Object-based interpolation techniques exploit information contained in the image slices to facilitate more accurate interpolation. This category can be further subdivided into interpolators that operate on extracted features or contours [5]–[9], and ones that operate directly on image intensity values [10]. In this paper, we are primarily concerned with the latter category, but also with special cases of the former category that allow whole-image interpolation, as that is the problem being addressed. Specific examples include the shape-based method proposed by Grevera and Udupa [11], the registration-based method proposed by Penney et al. [12], and a number of optical flow-based method

Free download research paper