Improving the Quality of Brain PET Images Using MR-Guided PET Reconstruction Technique With Integrated PET/MR

To evaluate the performance of an MR-guided PET reconstruction method in enhancing the image quality of brain PET images acquired with an integrated PET/MR system. We implemented an MR-guided PET reconstruction (MRg) algorithm in an integrated PET/MR system and conducted both phantom and clinical studies. The PET images of a Hoffman phantom were reconstructed using the MRg algorithm and routine Ordered Subsets Expectation Maximization (OSEM) algorithm. Image quality metrics were calculated to investigate the dose effects, number of iterations, and penalization factor (β). In the clinical study, the PET images of 30 patients without brain lesions, two patients with diffuse large B cell lymphoma (DLBCL), 18 patients with Parkinson's disease (PD) and one patient with cavernous angioma were reconstructed using MRg and OSEM, respectively. Qualitative and quantitative assessments were performed. In the phantom study, MRg images demonstrated improved image quality metrics (higher contrast and lower variability) compared to conventional OSEM, particularly at lower doses. The penalization factor $\beta =2$ maximized the advantages of MRg. In the clinical study of patients without brain lesions, MRg significantly outperformed OSEM in image quality evaluation. For patients with DLBCL, the lesions' signal/contrast-to-background/noise ratios exhibited significant improvements using the MRg. For patients with PD, the standardized uptake value ratio and contrast-to-noise ratio of bilateral putamen were significantly improved by MRg. The MRg has convincingly demonstrated significant advantages over the conventional OSEM in brain PET images, indicating great potential in clinical routines for brain PET/MR examinations.