TwinTree Insert

16-02 Inherent Problems


n order to process the time series (= the series of images starting with one or se­ve­ral precontrast images followed by the contrast-enhanced images) on a pixel-by-pixel basis, the spatial location of picture elements within the image ma­trix must be exactly the same. In many instances, this is not the case due to mo­ve­ment (Figures 16-02 and 16-03).


Figure 16-02:
Top: Organs in the human body may move in all three dimensions and rotate at the same time.
Center: Thus, multichannel images can easily be out of align­ment due to motion and rotation.
Bottom: Pixels have to be realigned for image-pro­cessing and analy­sis. Anatomically, the yellow pixels cor­res­pond to the same voxel, but have a different lo­ca­tion on the slices.


Figure 16-03:
Four images taken out of a dynamic study of the kidneys. During the examination, both kidneys move in all three dimensions and rotate. In this example, the right kidney moves several pixels up and down during the time series (yellow line). Mathematical processing of the images can help eli­mi­na­te some of the movement and facilitate processing the time series.


This hampers the calculation and evaluation of time intensity curves and pa­ra­met­ric images [⇒ Gehrig 1991]. Pixels or regions-of-interest have to be reallocated ma­­nu­­al­­ly or, preferably, automatically, e.g., by contour recognition. The re­align­ment of the images is often referred to as image registration.

This problem is even more complicated in contracting organs such as the heart [⇒ Hig­gins 1996]. Here artifacts can be corrected by highlighting the edges of the or­gan by drawing the boundaries either manually, semiautomatically or au­to­ma­ti­cal­ly if there is some a priori knowledge, e.g., previous knowledge of the outline of the or­gan and the ratio between its length and width (Figure 16-04) [⇒ von Schult­hess 1991]. A more sophisticated approach is the automatic creation of regions of in­­te­r­est (auto-ROIs) [⇒ Torheim 1997].


Figure 16-04:
Left: A reference image is selected.
Center: The boundaries of the kidney can be manu­ally drawn and ad­justed in each image of the exa­mi­­na­tion series.
Right: An edge-detection program can define the boundaries through contour-enhancement and au­to­­ma­ti­cal­ly or semiautomatically adjust them by com­paring the ratios between fixed pa­ra­me­ters.


spaceholder redImage registration of contrast-enhanced images of non-rigid organs is a dif­fi­cult prob­lem due to the changes not only in contrast but also in the structures vi­si­ble.

Whereas rigid organs like the brain can be aligned by performing trans­la­tions and ro­ta­tions, non-rigid require non-rigid reformations, which po­ten­ti­al­ly can change the structures to be observed.

An additional problem is the propagation of mistakes, for instance through ar­ti­facts which are not recognizable any more on pro­ces­sed images. They can be caused by a change of relative or absolute signal intensity on the image influenced by outside factors such as surface coils.


spaceholder redFurthermore, due to the frequent presence of time constraints, the signal-to-noise ratio is often quite low in dynamic contrast-enhanced MR imaging.

This necessitates the need for noise-reduction techniques which are com­mon­ly per­form­ed by applying so-called filters either in the spatial (image) or in the temporal domain [⇒ Sebastiani 1996].

In the meantime there are some filters which can remove noise in a way which was previously only possible by increasing the magnetic field strength or im­pro­ve­ments in the amplifiers or other hardware.