Extending the field of view of a multi-pinhole cardiac SPECT camera

Objectives Multi-pinhole cardiac SPECT cameras offer several advantages over traditional parallel-hole collimated systems. They have increased sensitivity allowing a reduction in image acquisition time or patient radiation exposure. The system also contains multiple heads in an arc and so provides sufficient acquisition angles to reconstruct 3D images without the need to rotate the camera. This feature facilitates dynamic imaging and measurement of absolute myocardial blood flow. However, a limitation of the camera is that it has a small field of view (FOV). This raises concerns for the positioning of very large patients and restricts application of this technology to other organ systems that have a larger tracer distribution. The objective of this study is to demonstrate the feasibility of extending the FOV of a multi-pinhole cardiac camera through the use of multiple acquisitions.

Methods The Discovery NM530c camera (GE Healthcare) is built on a rotating gantry system just as is a standard gamma camera. Thus it should be possible to treat it as a traditional multi-head camera, acquire projection data from additional angles and reconstruct the whole projection data set. Obtaining data with the detector at more than one angle provides the additional views needed to support reconstruction over an expanded FOV. Reconstruction in this study was done off-line using in-house developed MLEM software that is based on a system matrix approach similar to the vendor supplied software. The reconstructed image space was expanded from 70x70x50 voxels to 140x140x50 voxels, maintaining a voxel size of 4mm. The center of rotation (COR) for the camera was located by using a point source at a fixed location and acquiring multiple images with the camera positioned at different radii and angles. The system matrix fixes the location of the reconstructed object space with respect to the camera head. Thus as the camera moves around the subject, the subject appears to both rotate and translate. If the rotation angle of the camera head, its radial distance from the COR and the bed position are known, then the object can be rotated and translated, allowing a single system matrix to be used for reconstruction of all projection sets. The feasibility of the approach was then tested by reconstructing an image of a Deluxe Jaszczak phantom using a combination of two projection data sets acquired at 315 and 135 deg. The phantom was filled with 1mCi of 99mTc and data acquired for 3 min at each projection angle using a 140 +/- 14 keV energy window. The images were reconstructed without attenuation or scatter correction.

Results Separate reconstruction of the two projection sets into the expanded (140x140x50) object space shows the expected image truncation and artifacts. That is, extension of the object space to encompass the full object is insufficient to allow an accurate reconstruction. When both projection sets are combined, the truncation and artifacts are removed and the reconstructed image provides a much more accurate representation of the true object. Because the reconstruction code employs a system matrix, reconstruction times remain acceptable with 100 MLEM iterations requiring less than 2 minutes on a single CPU core.

Conclusions It is feasible to expand the reconstructed FOV of a multi-pinhole cardiac camera by integrating the projection data from two different camera angles. This approach has the potential to expand the applications possible with this type of technology.