Accurate Image Reconstruction with Computed System Response Matrix for a High-Sensitivity Dual-Head PET Scanner

We have recently proposed a compact, dual-head PET scanner configuration for providing high-sensitivity imaging of small animals. Although the scanner is able to reach a sensitivity of about 30% at the center of imaging field-of-view, its compact configuration produces substantial depth-of-interaction (DOI) blurring and results in significantly degraded spatial resolution. It is known that DOI blurring can be reduced if the system response matrix (SRM), which describes the individual sensitivity of the detection channels of the scanner to image voxels defined in its field of view, can be incorporated in reconstruction. In practice, however, the huge dimension of the SRM found in modern PET scanners make its computation and storage extremely challenging. In this paper, we show that for the dualhead scanner configuration one can employ cubic image voxels having a proper size to create substantial symmetries in the SRM, as a result producing drastic reductions in computation and storage. We have applied this strategy to the proposed highsensitivity small-animal PET scanner (μPET) to enable accurate computations of its SRM by using Monte-Carlo simulation. Our results with simulated data indicate that the proposed μPET scanner can achieve an isotropic and uniform spatial resolution of ~1.2 mm after incorporating the SRM in reconstruction. In contrast, the image resolution deteriorates significantly and becomes non-isotropic when not employing the SRM: at the center of the scanner, its resolution is ~1.8 mm in directions parallel to the detectors and becomes ~3.2 mm in the direction normal to the detectors. Furthermore, the resolution in the latter direction deteriorates considerably when moving away from the scanner’s center. Images generated by with employing the SRM in reconstruction also show substantially better noise properties than those generated without. When applied to a real dataset, considerable enhancement to the image resolution and contrast is obtained when using the SRM in reconstruction.