Separation of Overlapped Projection Images in Multi-pinhole SPECT System With a Neural Network

We propose an image reconstruction method for the overlapping projection images that occur when using a multi-pinhole single photon emission computed tomography (SPECT) system. Generally, data acquisition in SPECT systems is performed with rotating gamma cameras. Therefore, data acquisition is time-consuming and dynamic study of organs is not possible. In addition, a patient must remain in a certain posture during data acquisition, which is a significant burden. On the other hand, system with multi-pinhole collimators from a limited direction range, they enable stationary data acquisition in multi-head SPECT systems and have many advantages such as dynamic data acquisition and reduced patient stress due to shorter data acquisition time. However, one of the disadvantages of this system is its low sensitivity. When the number of pinholes increases to improve the data acquisition angles and sensitivity, the projection images of the pinholes overlap, resulting in artifacts in a reconstructed image. To address this issue, when we avoid the overlap of the projection images, the spatial resolution of the reconstructed image is reduced. In this paper, we propose a method to separate overlapped projection images in a multi-pinhole SPECT system using a neural network. We confirmed that the overlapped projection images can be removed.