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 12bit integer or floating point representations, commonly used displays are only capable of handling 8bit gray or color levels. In a typical medical scenario, a 3D scanner will generate a 12bit dataset, which will be downsampled to an 8bit per-voxel accuracy. This downsampling is usually achieved by a linear windowing operation, which maps the active full accuracy data range of 0 up to 4095 into the interval between 0 and 255.

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 8bit volume datasets is significantly enhanced, in particular if the imaging modality allows for a high tissue differentiation (eg., MRI). Henceforth, 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.