15-03 Subtraction or Overlay Images
Multispectral images of the same body region can simply be overlaid to give an impression of the exact location of certain contrast-enhanced structures (data fusion). Usually, a high spatial resolution T1-weighted MR image is used as a background image to show the anatomical structures and the contrast-enhanced image is projected onto this picture.
In MR imaging this has been first shown with perfluorinated ventilation images in 1982 (Figure 15-07 [⇒ Rinck 1983 + 1984]).
Diagram explaning the theory of the first fusion images in MRI (1982). Subtraction of hydrogen and fluorine images of the lungs.
Today the method is used as a multiparametric or multimodal fusion of data as well as longitudinal (time domain) integration of single modality data, commonly applied in fMRI studies or in intermodality comparison between, e.g., CT or MR and PET images. Here, the information obtained from PET is overlaid or imprinted onto the more detailed anatomic images acquired with MR imaging or CT.
Practical Applications. The method is useful since it allows a better visualization to locate certain processes. The implementation is relatively simple. It is mainly used as an auxiliary tool to facilitate visualization of enhancement visible on postcontrast images (Figure 15-08), but also in MR angiography to highlight veins after subtraction of the CE-MRA images of the arterial phase, and in MRSI (Figure 05- 08).
15-04 Quantification of MR Parameters
With numerous tissue parameters, MR imaging has substantial – theoretical – potential for tissue discrimination in different organs. The most important intrinsic contrast factors are proton density, T1 and T2 relaxation, and bulk flow.
The use of relaxation times for medical applications was first proposed in 1955 [⇒ Odeblad]. Voxel-by-voxel in vivo relaxation-time measurements, partly turned into T1- and T2-maps, have been tried out over the years by a large number of researchers [⇒ Skalej]. However, parametric T1 and T2 images did not enter into clinical routine. They were restricted to a single parameter only and revealed less information than images representing several parameters combined with different parameter-weighting.
This was one of the first major lessons to be learnt in MR image-processing: if one has more than one known factor influencing the contrast of an image, and if the change in contrast is perceivable by the human eye, it is not worthwhile to extract such a factor to create a parametric image. This holds in particular if this factor cannot be quantified exactly. In the case of relaxation times, only an estimation is possible in vivo. In 1985, it was finally realized that even carefully performed in vivo T2 measurements cannot be used as a diagnostic method in cancer detection, characterization, or typing [⇒ Rinck 1985].
Synthetic or simulated images. For specific applications, pure relaxation time maps can be used to create synthetic MR images and to simulate image contrast behavior. Such techniques were proposed very early in MR imaging in the first half of the 1980s to allow fast retrospective optimization of image contrast. A number of publications dealt with this [⇒ Bielke, ⇒ Bobman, ⇒ Riederer, ⇒ Torheim] and dedicated software programs, e.g., MR Image Expert [⇒ Torheim 1994 +1996] were developed for educational and research purposes. Some examples are shown in Chapter 10.
The procedure leading to synthetic images requires several steps. High quality, low noise simulations are based on true T1, T2, and proton density maps of the same slice or volume. Then pixel-by-pixel signal intensities can be calculated with equations: the operator-selected variables are, for instance, TR, TE, TI, and FA.
Simulated images have substantially less noise than images acquired directly on an MR machine. Applying computer simulation for sequence optimization is time and cost efficient compared to in vivo experiments. They can be used when looking for specific anatomical or pathological features or to evaluate best pulse-sequence parameters for contrast agent enhancement or comparison of contrast at different field strength. MR Image Expert, the simulator used for this textbook (Figure 15-09), could also be employed, e.g., for clinical imaging if integrated into a suitable MR equipment. The drawbacks of such image-processing programs are their dependence on specific "clean" data acquisition sequences with known contributing components where signal intensities can be exactly and reproducibly calculated.
In addition to the factors mentioned above (exclusion of many parameters influencing image contents and contrast, e.g., multi-exponential decays, diffusion, and flow), multispectral processing and feature extraction for the creation of synthetic images are cumbersome and prone to substantial mistakes (see also Multispectral Analysis and Relaxation time values and proton density calculation). More so, in the brain, for instance, absolute signal amplitudes are proportional to the water content, not to "proton density" because myelin lipids do not contribute to the signal [⇒ Fischer 1990], another of the many features that cannot be simulated. The sometimes proposed fingerprinting based on multi-parametric data collection is unreliable and impracticable in diagnostic routine.
T1 maps are also used as the basis for calculating tissue concentrations of contrast agents in dynamic imaging. Here, two measurements are necessary, one before injection of the contrast agent, a second one after injection together with drawing a blood sample to determine the blood concentration of the contrast agent. It is rarely used in clinical routine.