Noninvasive Extraction of Input Function from Carotid Artery in H2 O Dynamic Brain Positron Emission Tomography Using Independent Component Analysis

For the absolute quantification of regional cerebral blood flow (rCBF) by means of H2 15 O positron emission tomography (PET) and kinetic modeling, arterial input function should be determined accurately. Even if arterial blood sampling, as an input function, provides an accurate time-activity curve (TAC), it is invasive and delay from carotid artery to radial artery should be corrected. Since the TAC from the region of interest (ROI) drawn on the small area containing carotid artery could not be determined objectively, more reproducible and objective method for noninvasive derivation of the input function is required. We applied independent component analysis (ICA) to dynamic brain H2 O PET image (canine study) to obtain input function from carotid artery noninvasively. The independent components were estimated by recursively minimizing the mutual information (measure of dependency) among the estimated independent components. We calculated area under the curve (AUC) of input functions from ICA method and from blood samples and compared each other to validate the accuracy of ICA method. ICA successfully extracted carotid artery TACs from the H2 O PET images and their corresponding images. The extracted TACs by ICA method were well correlated with the arterial blood samples.