Global strategy to extract automatically relevant subdominant perfusion information: application to skeletal muscle NMR imaging with arterial spin labeling

This paper describes a global strategy of image processing to automatically extract perfusion information, when it is not the predominant information in the sequence of images. It is applied to MR perfusion studies of skeletal muscles, acquired with arterial spin tagging sequences. First the dynamic images are registered, then methods based on factor analysis are applied to discriminate between highly perfused regions and low perfused regions. Finally a functional segmentation is achieved on the parametric images, which are computed by factor analysis. An example of the whole processing is given, which shows the possible functional discrimination between different regions inside one muscle. The effect of misregistration on the final segmentation is studied. The method is proved to be efficient provided that image data could be either well registered or discarded.