Motion-enhanced, differential interference contrast video microscopy using a GPU and CUDA

Abstract

Optical video microscopy is widely used to observe living cells and their moving parts. The smaller moving parts of the cells, such as vesicles, have low contrast and are often obscured by membranes and cell walls. Large images (1k x 1k) showing many cells are most helpful to the microscopist; limited memory prohibits storing such images for the entire life of a cell. As a result, it is imperative that image enhancement calculations be performed in real time, so that the researcher can observe moving vesicles immediately, rather than by post-processing. The MEDIC algorithm uses background subtraction to remove or at least minimize the effects of the immobile parts of the cell, including the cell wall. With MEDIC, moving objects are visible to the naked eye. In this paper, we extend the MEDIC algorithm to take advantage of fast computing on GPUs. Current mainstream CPUs are not fast enough to execute the MEDIC algorithm in real time with fast cameras. Dedicated image processing boards, made by companies like Matrox Imaging, are faster, but they are also expensive. GPUs, which are designed for rendering video game graphics, are made to perform calculations in parallel, and they can be obtained for a few hundred dollars. While not as fast, they are still well suited to executing the MEDIC algorithm in real time. The GPU can provide a significant speedup over CPU computations, making real time imaging possible with fast cameras for a fraction of the price of dedicated image processing boards.