Research On PET-MRI Joint Reconstruction Based On Anatomical And Functional Joint Prior

The hybrid PET/MR system is the up-to-date multi-modality medical imaging equipment combining the positron emission tomography(PET)and the magnetic resonance imaging(MRI).It can achieve anatomical MR images with high resolution and functional PET images with high sensitivity synchronously,and can acquire both PET data and MRI data simultaneously.However,PET imaging and MR imaging are still conducted separately and independently,and the quality of reconstructed images still depend on single-modality reconstruction algorithms,which in fact fail to adequately utilize complementary information between multi-modality images.Traditionally,some scholars have utilized Bayesian or Maximum a posteriori(MAP)methods to incorporate high resolution structural imaging modality(e.g.MR images)for PET reconstruction to improve the quantitation level of PET images.However,these methods are usually utilized for PET reconstruction by incorporating the acquired MR images,and it is difficult to register the PET and MR images from different equipment at different times.The PET and MRI data acquired simultaneously from PET/MR system can avoid problems of image registration,and provide high quality complementary information.Therefore,how to utilize the complementary information for PET-MRI joint reconstruction is currently a key research field.Recently,the structural information between PET and MRI modalities utilized for PET and MRI joint reconstruction based on MAP algorithm has been proposed by Ehrhardt.M.J.et al.They designed the joint prior terms including joint total variation(JTV)and parallel level sets(PLS)for extent phantom experiments,which could suppress noise and improve the quality of images.However,due to the severe artifacts in the initial under-sampled MR images,it could also easily cause cross artifacts about features between PET and MRI images,which therefore degrades the reconstructed images.In order to solve this problem,we propose a MAP reconstruction method for dual-modality joint reconstruction of PET-MRI to enhance reconstructed images via incorporation of a novel cross-guided(CG)prior model.Two nonlinear potential functions based on the gradient information of the PET and MR images within iteration was utilized in the CG prior.With the reasonable weighted parameters in the reconstruction model,the weighted penalties on the regions of boundary and non-boundary were assigned by the nonlinear functions to preserve edges and smooth noisy regions.We designed multiple experiments for phantom and brain data,and optimized the related weighted parameters of reconstruction model.Subsequently,we compared the performance of the proposed reconstruction algorithm and conventional separated reconstruction algorithm without prior(no prior),separated total variation(TV),joint total variation(JTV)and linear parallel level sets(LPLS)methods.As the experimental results showed,the proposed joint reconstruction algorithm can considerably suppress noise and artifacts while retaining the unique features,and so visually outperform other algorithms.Quantitatively,the proposed algorithm can achieve lower normalized root mean square error(NRMSE)level and higher structural similarity index measurement(SSIM)and signal to noise ratio(SNR)level.In conclusion,through extensive experiments,it was demonstrated that the proposed joint reconstruction algorithm resulted in clearer visual and quantitatively enhanced reconstructed images with more complete anatomical structure information and edge details,which helps advance the clinical application and development of PET/MR systems with great clinical significance and application value.