A temporally constrained ICA (TCICA) technique for artery-vein separation of cerebral microvasculature

A fully automatic ICA based data driven technique which incorporates additional a priori information from physiological modeling of the cerebral microcirculation (gamma variate model) is developed for the separation of arteries and veins in contrast-enhanced studies of the cerebral microvasculature. A dynamic data set of 50 images taken by a two-photon laser scanning microscopy technique that monitors the passage of a bolus of dye through artery and vein is used here. A temporally constrained ICA (TCICA) technique is developed to extract the vessel specific dynamics of artery and vein by adding two constraints to classical ICA algorithm. One of the constraints guarantees that the extracted curves follow the gamma variate model of blood passage through vessels. Positivity as the second constraint indicates that none of the extracted component images that correspond to the artery, vein or capillaries in the imaging field of view, has negative impact on the acquired images. Experimental results show improved performance of the proposed temporally constrained ICA (TCICA) over the most commonly used classical ICA technique (fast-ICA) in generating physiologically meaningful curves; they are also closer to that of pixel by pixel model fitting algorithms and perform better in handling noise. This technique is also fully automatic and does not require specifying regions of interest which is critical in model based techniques.