Image Processing Techniques in Emission Tomography - Correction Techniques in Emission Tomography - page 144

(a) Original (b) RL (c) BD

FIGURE 7.14: (a) The original PET image of a liver tumor. (b) and (c)
Deconvolution results using Richardson{Lucy and blind deconvolution respec-
tively.

Multiresolution
The inclusion of information from high-resolution images, e.g., CT or MRI, is
a promising approach for PVC. The information of functional imaging, i.e., of
PET data, is combined with the anatomical information of high resolution
CT data in a reasonable way. Besides a registration of PET data and the
respective modality, a positive correlation of intensity values in the region-of-
interest is required. The additionally used imaging modality is downsampled
to the resolution of PET, for example, by using wavelets. Thus, it is possible
to reconstruct lost details in the low resolution image using the additional
resolution levels from the wavelet transform. A comparison of the registered
images results in partial volume corrected signal intensities.
In [5] a multiresolution technique is proposed using 2D wavelet transforms
and a linear model for scaling the transformed images. Based on this approach
a method using Markov-modeling is proposed in [50] to compare the two image
modalities in order to avoid the introduction of PVC artifacts. An improve-
ment compared to [5] is reported in [50]. The advantage of these methods is
that no prior knowledge about the PSF is required.

7.6 Super-resolution
Compared to CT with a maximal spatial resolution below 100 m, PET
and SPECT obtain a relatively poor resolution of approximately 4 mm. Thus,
there is a demand to increase resolution of PET. Besides improving the PET
scanner's hardware, resolution is to a certain extent improvable through image
processing methods. These methods estimate a high resolution image (HR)
from several low resolution images (LR). The LR images are acquired at
slightly shifted positions, thus supplying complementary image information.
In the reconstruction of PET images the resolution obtained is usually lower
than the scanner's resolution due to a trade-o between noise and resolu-
tion [25].