Anatomical And Functional Information-Guided PET Images Partial Volume Correction

Positron Emission Tomography (PET) is the best advanced clinical examination technology of nuclear medicine. PET can provides early diagnosis of the disease at the molecular level by exploiting the tracer techniques compared with CT/MRI. Due to the limitation of of imaging theory and detection technics, the partial volume effect (PVE) of PET is more prominent than CT/MRI. The partial volume effect can be described that the measured value is falling into the average of these matters when there are two or more different density matters in the same scan level. It can blur the image edge information, enlarge the lesions size, and lead to image distortion and impair the accuracy of the clinical diagnosis. Therefore, the partial volume correction method is one of the most chanllege research topic in PET imaging.Various methods have been proposed to correct the PVE in PET images. Hoffman proposed the recovery coefficient (RC) method for description and correcting the PVE of PET. This method is implemented to measure recovery coefficient in various diameter regions of interest (ROIs) of standard image quality phantom. However, it is difficult to apply the method in clinical because it need manually define the lesion regions. Nevertheless, recovery coefficient is used as the quantitative evaluation metric. In recent years, there is an approach incorporating the PSF into the system matrix to enhance the spatial resolution in reconstructed PET images, namely "Resolution Model (RM)". The RM can reduce the noise in the reconstruction and improve noise properties. However, Recovery is limited at high iterations since there will still be residual partial volume effects.Another approach can be performed by using a deconvolution operation for image enhancement by post-processing the PET image. The deconvolution methods are simple and easy to implemente, and can be applied to whole body PVC. Teo firstly applied the Van Cittert (VC) deconvolution algorithm to correct PET tumor image, then Tohka and Reihac applied the VC and Richardson-Lucy (RL) algorithm to correct PET brain image. Unfortunately, the VC and RL can lead to high noise with iteration increasing. Therefore, we foucus on PVC methos that can recovery the intensity while decrease the noise level.Later, methods incorporating anatomical image prior information have been proved theoretically correct and practically effective compared to other methods. They can be divided into two cagegories:the first category is reconstruction-based method (e.g. Bayesian or Maximum a Posteriori (MAP)) by incorporating anatomical image edge or region prior information for partial volume correction (PVC). The second category is implemented in post-reconstruction step. Inparticular, methods incorporated MRI/CT information for restore the real activity concentration in region or pixel level. The most classic method is the geometric transfer matrix (GTM) method, which has been a standard PVC method. However, the two categories methods are mainly used to study the PET brain image and required the anatomical image to accurate segmentation. Therefore, how to exploit the anatomy superior edge structure and reduce the bias inducing in segmentation step are the focus in this paper.In this paper, we analysis and research the correction methods of partial volume effect. We have done following work on partial volume correction:1. We proposed partial volume correction in PET imaging incorporating total variation (TV) regularization in deconvolution alogrithm. To deal with edge blur, activity attenuation characterics in PVE and VC and RL introducing high levels of noise, we applied the TV regularization procedure in VC and RL in postreconstruction situation. Then they are tested in simulated NCAT images and images of NEMA NU4-2008 IQ phantom and tumor mouse scanned by Simens Invoen microPET. The simulated experiment and tumor mouse experiment results show the algorithms using TV regularization provides superior qualitative and quantitative appearance compared with traditional VC and RL algorithms in terms of image visualization, recovery coefficient, standard deviation (SD). Experimental results show that the present methods can enhance image concentrations and achieve significant goals in terms of noise suppression and edge information preservation. Besides, deconvolution methods can change the image noise distribution and noise distribution has important influence for tumor detection. We use prewhitened matched filter (PWMF) and non-prewhitened matched filter (NPWMF), for Simenz biograph HR PET/CT clinical data parameters, to discuss different deconvolution methods in detection tumor under various PSF values. The experimental results show the increasing inter-voxel correlations with increasing values of PSF degrade detection task performance in the case of NPWMF for RL-TV and VC-TV. Only when in PWMF case, to remove the inter-voxel correlations, RL-TV and VC-TV will improve the effect of lesion recognition and detection.2. We proposed anatomy guided partial volume correction for PET images. Based on the development of incorporating the MR anatomical information into PET imaging, we proposed MR image-guided, pixel-based partial volume correction method for PET images. The PET images are corrected by iterative deconvolution with edge preserving smoothness constraint which is constructed from coregistered MR image without segmentation. The correction is implemented in a Bayesian deconvolution framework and is solved by a steepest descent method. The method is tested in simulated images and Hoffman brain images. Experiment results show the proposed algorithm provides superior quantitative appearance compared with uncorrected images. The proposed algorithm is an efficient method for reducing PVEs in PET.