Functional statistics based method for the evaluation of the registration of sequences of 3D perfusion MR images

Accurate registration of medical images is a necessary task for several important diagnosis techniques. Nevertheless, it is a difficult challenge due to movement of the patient, deformations, noise in the signal, etc. Besides, evaluation of the quality of the performed registration is also troublesome, specially when no golden pattern (true result) is available and/or when the signal values may have changed between successive images/volumes to be registered. This work introduces a method to evaluate the accuracy of the registration of a series of volumes, magnetic resonance images (MKI) of the liver and adjacent areas, that have been taken while a contrast is diffused through the liver tissue. This means that intensity of each observed point changes because of two reasons: contrast diffusion/perfusion and deformation of the liver (due to movement of the body and breathing). The main idea is to compare the accuracy of several different registration methods (or variants of a single method) by examining the perfusion curves at selected points. By modeling the function that gives the intensity at a given point as a functional datum and using statistical techniques, we try to assess its change in comparison with other functions. The underlying idea is that the more accurate the registration is, the smoother this function will become. This is because one of the reasons for variation of the perfusion function is the movement (points of the series of volumes are not accurately matched) whereas the other reason (contrast diffusion) is the same for all methods. The method is evaluated on a set of perfusion MKI liver images and results for the comparison of two variants of a state-of-the-art registration method are given.