Brain PET Image Partial Volume Correction

Positron Emission Tomography（PET）is a medical imaging device with extremely high biochemical sensitivity and has become a key tool in the research of brain diseases.However,due to the limited spatial resolution,the measured value of small region with high activity is lower than the actual one,and the measured value of small region with low activity is higher than the actual one.This phenomenon is known as Partial Volume Effect（PVE）.As the brain structure is delicate and complex,the influence of PVE in PET imaging degradation is more severe,which consequently restricts the development of quantitative research in the field of brain PET images.In this article,the causes of PET partial volume effect,as well as the partial volume effect correction models and methods had been thoroughly studied.One commonly used method,Region Based Voxel-wised Correction（RBV）guided by anatomical images could process images with high accuracy.Howerver,registration and segmentation errors in the process is often difficult to avoid,and regions parcellated by structural imaging data do not necessarily represent regions of uniform uptake in the PET data.In order to solve this problem,this article innovatively proposed Mask Evolution Partial Volume Correction,ME-PVC.The algorithm is based on the PET image,and uses the minimization of Region-Scalable Fitting Energy to evolve the region boundary progressively,eventually obtaining a template that is more consistent with the template based on the assumption of uniformity for correction.The validity and reliability of ME-PVC are further proved by using Monte Carlo simulation data and the actual data obtained from the real brain PET system.After theoretical experiment verification,the proposed method ME-PVC was applied to the study of Alzheimer’s disease（AD）.The ¹⁸F-AV45 and ¹⁸F-FDG PET images of each subject,including both AD patients and cognitively normal subjects,were selected from ADNI database.ME-PVC and RBV corrections were then performed on these data.In the analysis of ¹⁸F-AV45 PET images,the cerebellum was selected as the reference region,and the Cohen’d effect of the cortical standard uptake ratio of the two groups of images corrected by ME-PVC was the largest,indicating that the correction is helpful to the diagnosis of AD using ¹⁸F-AV45 images.In the ¹⁸F-FDG PET data analysis,the uncorrected image showed the largest difference in SUVR between groups.After correction,the difference between groups still exists,which proves that it is AD not just the effect of brain atrophy that causes a decrease in the metabolism of some brain areas.The ME-PVC proposed in this article can still obtain accurate results in the presence of segmentation errors.The algorithm only requires the approximate location information of each target area,and it does not necessarily have to incorporate corresponding anatomical imaging.It is believed that this discovery can be applied to a wide range of imaging situation,making it competitive in the clinical use.Modular design makes the algorithm have good scalability.In the current iterative framework,the regional evolution method can be replaced with other suitable PET image segmentation methods to explore its application in other brain diseases or cancer scenarios.