Feature Based Image Registration for Functional MR images using prior shape information

摘要

Motion can have a significant impact on signal changes in functional MR images, and affects the detection of task activation. To minimize the effects of motion on the fMRI signal, we propose a feature based image registration model. Due to T_2~* weighted signal loss, decreased resolution and low contrast in fMR images, the feature to be used for realignment can't be detected reliably by any edge detector that uses only the image gradient, like the popular model of geometric active contours. Our approach uses a prior given shape to find the feature in a low-resolution image. In addition to this, we can find an affine alignment of the segmented feature with the given shape.Given a time series of images, we use this method to find transformations to realign the images. In this model, we minimize an energy functional that depends on the image gradient and the given shape, so that the boundary of the object captured occurs at high gradients and is as close as possible to the given shape. This model has been tested both on synthetic data and fMR brain image data. The. experimental results show the effectiveness of this model in feature determination and time series image registration.