high performance computing on graphics processing units: hgpu.org

Laser scanning confocal endomicroscopy (LSCEM) is emerging as an in vivo 3D cellular imaging technology. However, the image datasets acquired from LSCEM are generally with low z-depth resolution, which causes artifacts in volume rendering. Intermediate data between consecutive images needs to be generated. One common solution uses trilinear interpolation which more or less reduces the artifacts, but blocky artifacts are clearly visible especially in the areas of low spatial resolution. The current shape-based interpolation techniques cannot effectively remove such artifacts since they rely on the hard classification for identification of shape features. At the same time, the higher order interpolation schemes like cubic b-spline interpolation take much time to process the slices. In this paper, we propose a novel GPU-based surface oriented inter-slice directional interpolation algorithm to recover the missing information and thus minimize the interpolation artifacts. Our algorithm enjoys full floating point precision offered by the modern GPU to enhance the visualization result. Experimental results on several datasets clearly demonstrate the effectiveness of the new method.