2. Algebraic 2D PET Image Reconstruction Using Depth-of-Interaction Information
Taiga Yamaya, Takashi Obi*, Masahiro Yamaguchi*, Nagaaki Ohyama* and Hideo Murayama (* Tokyo Institute of Technology)
Keywords: image reconstruction, depth-of-interaction (DOI), positron emission tomography (PET), nuclear medicine
Development of a new generation positron emission tomography (PET) system with depth-of-interaction (DOI) capable detectors is in progress at the National Institute of Radiological Sciences. In current 3-dimensional (3D) PET scanners, the length of the detector crystals is about ten times as long as their width in order to improve detection efficiency. Therefore the PET measurement system exhibits shift-variant characteristics, such as broadened sensitivity functions of each detector pair from center to edge of field-of-view (FOV) and/or from small to large ring differences. However, using DOI information can narrow the broadened sensitivity functions while maintaining system sensitivity.
In this paper, we applied an algebraic image reconstruction method, such as natural pixel decomposition (NPD), to a 2-dimensional (2D) DOI-PET scanner, in order to evaluate the effects of using DOI information on PET image quality. Algebraic reconstruction methods have been successfully used to improve quality of PET images by accurate modeling of the measurement system, while the conventional filtered backprojection (FBP) method is based on an inaccurate system model. The measurement system model for the DOI-PET scanner was defined in consideration of geometrical arrangement and penetration of crystals. At this stage, we supposed that scatter coincidences, random coincidences and attenuation were corrected completely.
We applied NPD to simulated data for a small animal DOI-PET scanner. The scanner had a ring of 144 BGO crystals of 3.8 x 3.8 x 10 mm3 arranged in 2 DOI layers. For comparison, a non-DOI PET scanner, which had a ring of 144 BGO crystals of 3.8 x 3.8 x 20 mm3, was also simulated. The detector ring had a diameter of 187 mm, and the FOV had a diameter of 140 mm. Reconstructed images were obtained using NPD and FBP. Two figures of merit (FOMs), background noise and spatial resolution, were used to evaluate the image quality. The spatial resolution was measured as the average of radial and tangential full widths at half maximum (FWHM) of the point spread function at the center, and 20, 40 and 60 mm off center. A warm phantom of 100 mm diameter was used to measure the background noise as the normalized standard deviation (NSD). First the trade-off between the background noise and the spatial resolution was investigated, using NPD with different values of the regularization parameter and FBP with ramp filters of different cut-off frequencies. Plots of radial and tangential resolution at the same background noise levels (NSD=0.09) are shown in Fig. 2. Comparison between NPD and FBP shows the improvement of image quality by using the accurate system model. Also comparison between DOI-PET and non-DOI PET shows the improvement of resolution uniformity by using DOI information.
In summary, the numerical simulation results show that accurate system modeling improves spatial resolution without noise emphasis, and that DOI information improves uniformity of spatial resolution.
Publications:
Yamaya, T., Obi, T., Yamaguchi, M., Kita, K., Ohyama, N., Hasegawa, T., Haneishi, H. and Murayama, H. : Jpn. J. Med. Phys., 21, 223-231, 2002. (in Japanese)
|