A constrained independent component analysis technique for artery-vein separation of two-photon laser scanning microscopy images of the cerebral microvasculature

Understanding brain hemodynamics as well as the coupling between microvascular hemodynamics and neural activity is important in pathophysiology of cerebral microvasculature. When local increases in neuronal activity occur, the blood volume changes in the surrounding brain vasculature. Dynamic contrast enhanced imaging (DCE) is a powerful technique that quantifies these changes in the blood flow by repeatedly imaging the vasculature over time. Separating artery, vein and capillaries in the images and extracting their intensity-time curves from the DCE image sequence is an important first step in understanding vascular function. A constrained independent component analysis (ICA) technique is developed to analyze the two photon laser scanning microscopy (2PLSM) images of rat brain microvasculature, where a bolus of fluorescent dye is administered to the vascular system as the contrast agent. A priori information inferred from the gamma variate model of cerebral microvasculature is incorporated with the data driven technique in temporal and spatial domains using two constraints. The constraints are: no independent component (IC) is allowed to have negative contribution in forming the images (positivity constraint) and the component curves follow a gamma variate function (model fitting constraint). Experimental and simulation studies are conducted to demonstrate the improved performance of the proposed constrained ICA (CICA) technique over the most commonly used classical ICA algorithm (fast-ICA) in providing physiologically meaningful ICs and its ability to separate the model following factors from other factors are shown. The efficiency of CICA in handling noise is compared to model based techniques. Its capability in providing improved separation between artery, vein and capillaries compared to the other two techniques is also demonstrated.