Toward lower contrast computer vision in vivo flow cytometry.

There are many applications in biomedical research where detection and enumeration of circulating cells (CCs) is important. Existing techniques involve drawing and enriching blood samples and analyzing them ex vivo. More recently, small animal "in vivo flow cytometry" (IVFC) techniques have been developed, where fluorescently-labeled cells flowing through small arterioles (ear, retina) are detected and counted. We recently developed a new high-sensitivity IVFC technique termed "Computer Vision(CV)-IVFC". Here, large circulating blood volumes were monitored in the ears of mice with a wide-field video-rate near-infrared (NIR) fluorescent camera. Cells were labeled with a membrane dye and were detected and tracked in noisy image sequences. This technique allowed enumeration of CCs in vivo with overall sensitivity better than 10 cells/mL. However, an ongoing area of interest in our lab is optimization of the system for lower-contrast imaging conditions, e.g. when CCs are weakly labeled, or in the case higher background autofluorescence with visible dyes. To this end, we developed a new optical flow phantom model to control autofluorescence intensity and physical structure to better mimic conditions observed in mice. We acquired image sequences from a series of phantoms with varying levels of contrast and analyzed the distribution of pixel intensities, and showed that we could generate similar conditions to those in vivo. We characterized the performance of our CV-IVFC algorithm in these phantoms with respect to sensitivity and false-alarm rates. Use of this phantom model in optimization of the instrument and algorithm under lower-contrast conditions is the subject of ongoing work in our lab.