PET image reconstruction based on multi-kernel fusion

Positron emission tomography (PET) is a commonly used medical imaging modality, but it has high noise at low doses. To further optimize PET image reconstruction, this paper proposes a novel approach that incorporates magnetic resonance (MR) image information into kernel-based PET image reconstruction. This method integrates MR image texture information into the kernel matrix of the PET prior information; the integrated information is subsequently incorporated into the forward projection model of PET. First, Haralick texture features from the MR images are obtained. Then, a kernel function is constructed using prior information from PET images and texture information from MR images. Finally, the kernel-based PET image reconstruction method is utilized to generate the reconstructed images. Experimental results demonstrate that the proposed method exhibits superior reconstruction performance compared to other existing kernel-based PET image reconstruction methods.