Motion Estimation - Motion Correction in Thoracic Positron Emission Tomography - page 55

results with comparable accuracy for PET motion correction, see [ 37 , 51 ] and the
evaluation at the end of this section. It is thus difficult to give preference to one
particular method in practice for all applications. The discussion above is intended to
provide some orientation in the reader's decision-making process for one method. In
order to give more assistance we perform a quantitative evaluation of both methods
in the following.

2.3.2 Evaluation

The following evaluation is carried out on software phantom data and patient
data . While the analysis based on the software phantom data is performed to
quantitatively compare the image registration and optical flow based approaches,
the patient study shows the general clinical applicability of the proposed methods.

2.3.2.1 Software Phantom Data

The images of the software phantom evaluation are based on the results presented
in Fig. 2.8 for V AMPIRE and Fig. 2.9 for MPOF. More information about the XCAT
software phantom data and in particular the image generation process can be found
in Sect. 2.1.6 .

Results

For a visual comparison we show some results of the proposed methods for motion
estimation in Fig. 2.10 . Cardiac slices are shown for a single gate in Fig. 2.10 aforthe
reference gate

R

, showing a high noise level. The combination of all available gates,
representing a reconstruction without any motion correction, is given in Fig. 2.10 b.
This image features severe motion induced artifacts compared to the reference
gate

R

, thus demanding for motion correction techniques. However, the SNR is
significantly higher compared to the single gate in Fig. 2.10 a. The combination
of all single gates using the ground-truth motion information, provided by XCAT,
can be found in Fig. 2.10 c. Accordingly, this image represents the best possible,
and thus desired, image quality as it combines the high SNR of Fig. 2.10 b and the
reduced motion of Fig. 2.10 a. The resulting images based on estimated motion can
be found in Fig. 2.10 d for the image registration based SAD V AMPIRE approach
and in Fig. 2.10 e for the MPOF approach. The image quality of both approaches is
comparable with the image based on the ground-truth motion in Fig. 2.10 c.