Improvement of the Spatial Resolution of a Multi-pinhole SPECT System With a Deep Learning Method

We propose a method to improve the spatial resolution of images obtained with a stationary multi-pinhole single photon emission computed tomography (SPECT) system. A multi-pinhole SPECT system has an advantage of being able to measure dynamic functions of organs, which is not possible with conventional SPECT systems. However, the image quality is very sensitive to a pinhole size. When the pinhole is large, the spatial resolution is significantly degraded, which could not be compensated with the widely applied 7-rays method. In this work, we proposed a deep-learning based method to improve the spatial resolution, in which we converted the projection images to those measured with an infinitesimal pinhole system. The feasibility of this method was investigated by an image reconstruction simulation. We obtained projection images with a stationary multi-pinhole system using a Monte Carlo method. The reconstructed images with our method had significantly higher spatial resolution and reproduced more detailed structures than those with the 7-rays method. These results demonstrated the effectiveness of the proposed method.