Automated Local Adaptive Thresholding for Real-time Feature Detection and Rendering of 3D Endomicroscopic Images on GPU.
Muhammad Mobeen Movania, Feng Lin, Qian Kemao, Hock Soon Seah · CGVR · 2009
Laser scanning confocal fluorescence endomicroscopy (LSCEM) is an in vivo cross-sectional optical imaging modality for acquisition of live cell images. Effective use of this imaging modality in clinics largely depends on the capability of real-time 3D image reconstruction, cell and tissue feature detection, and rendering. To match the video rate of image streaming, the online detection and rendering of the volumetric features implied by the cellular images has been proven technically challenging. Upon a comprehensive survey and comparative analysis, in this paper, we present a novel GPU-based solution to the real-time 3D feature detection and rendering, for an interactive visualization of the endomicroscopic images. We preprocess the acquired datasets by employing mathematical morphology with multiple render targets in a multi-pass approach on the GPU. A transfer function is defined online for the volume of interest (VOI), based on an automated local adaptive thresholding technology. The mechanism and GPU implementation techniques for dynamically generation of the transfer functions are presented. Advantages of this novel approach are shown in the experimental results.