Volumetric high dynamic range windowing for better data representation

Abstract

Volume data is usually generated by measuring devices (eg. CT scanners, MRI scanners), mathematical functions (eg., Marschner/Lobb function), or by simulations. While all these sources typically generate 12 bit integer or floating point representations, commonly used displays are only capable of handling 8 bit gray or color levels. In a typical medical scenario, a 3D scanner will generate a 12 bit dataset, from which a subrange of the active full accuracy data range of 0 up to 4096 voxel values will be downsampled to an 8 bit per-voxel accuracy. This downsampling is usually achieved by a linear mapping operation and by clipping of value ranges left and right of the chosen subrange.In this paper, we propose a novel windowing operation that is based on methods from high dynamic range image mapping. With this method, the contrast of mapped 8 bit volume datasets is significantly enhanced, in particular if the imaging modality allows for a high tissue differentiation (eg., MRI). Thus, it also allows better and easier segmentation and classification. We demonstrate the improved contrast with different error metrics and a perception-driven image difference to indicate differences between three different high dynamic range operators.