An improved PET image reconstruction method based on super-resolution

Positron emission tomography (PET) is a non-invasive high-end examination that can quantitatively detect early disease stages. It complements information provided by functional and anatomical imaging. Therefore, PET is widely used clinically early on in the process of diagnosing malignant tumors or lesions. Fast and accurate reconstruction of PET images has been the subject of ongoing research. Patch-based regularization penalty likelihood reconstruction can reconstruct PET images more accurately, but it is sensitive to its algorithm's parameter values and requires a great deal of time to adjust parameters to achieve the best reconstruction. In this paper, we propose a novel method that uses random forests to improve PET imaging resolution at each iteration reconstruction step in the sinogram domain and the image domain; we refer to this method as patch-based super-resolution random forests reconstruction (patch-SRF). The patch-SRF algorithm allows the reconstruction to converge in advance and avoids the free-time adjustment process, achieving better reconstruction results despite relatively poor parameter settings.