Research On Simulation Of Small Animal PET System And Filling Method Of Detection Data

Positron Emission Tomography(PET)is a noninvasive molecular imaging technology that can dynamically reflect the metabolism level of living tissues and organs.Small animal PET technology is of great significance in scientific research such as pharmacokinetics,new drug development,and efficacy evaluation.However,the quantitative accuracy of small animal PET is still limited by the spatial resolution and sensitivity of the detector.Therefore,developing a small animal PET systems with high-performance has always been an important content in molecular imaging science research.In this paper,a prototype of a small animal PET system with high spatial resolution and high detection sensitivity was built using LYSO crystals and SiPMs detectors.The prototype consists of 60 crystal detectors divided into 5 consecutive 12-sided detection rings,with a radial diameter and an axial span of 102 mm and 125.4 mm,respectively,so that it has a maximum photon reception angle of 50.8°.The full text mainly focuses on the performance evaluation of the prototype and the method of filling the detection data.Based on this,the data processing software of the small animal PET imaging system is developed.The specific content is as follows(1)System simulation and expected performance evaluation of proposed small animal PET.The Monte Carlo platform,GATE,was used to build a simulation model of the prototype,and then the spatial resolution,counting performance(scattering fraction and noise equivalent count rate),detection sensitivity,and imaging quality of the prototype were pre-evaluated and analyzed.The simulation results show that the prototype has a spatial resolution of 1.62 mm,a detection sensitivity of 9.26%,a scattering fraction of 20.8,and a noise equivalent count rate of 2,256 kcps.The overall performance is similar to the Siemens Inveon PET system and the detection sensitivity and noise equivalent counting rate performance has been improved by 21.36%and 35.14%,respectively.Therefore,the proposed prototype is expected to further improve the quantitative accuracy of PET application in small animals.(2)Method of filling missing data of sinogram based on deep learning framework.Due to the gap of about 4.5° between adjacent crystal detectors in the prototype,about 15%of the data in the sinogram was lost.Existing detection data filling techniques cannot achieve a good balance between image quality and reconstruction efficiency.In view of this problem,this paper proposes a deep learning-based data filling and reconstruction network(Gap-filling and Reconstruction Net,GapFill-Recon Net),which consists of two sub-modules:Gap-Filling block and Image-Recon block.A convolutional residual networks was used to fill the incomplete sinogram and an automatic encoder-decoder network was established to implement the direct mapping from the projection domain to the image domain.Totally 150 whole-body PET images of mice was randomly generated using MOBY simulation phantoms.And after data enhancement and screening,43,660 2D PET images were obtained as the reference images.Then use the Solid Angle model to generate the system response matrix of the small animal PET prototype as the projector.Then use the projector to forward project the PET reference image to generate the corresponding incomplete sinogram as the input of the network for training.The experimental results are compared with MLEM,Interp-FBP and Gap-Filling FBP algorithms.The results show that GapFill-Recon Net reconstructions achieves the best performance on three performance indicators:average relative root mean square error(rRMSE),structural similarity(SSIM),and peak signal-to-noise ratio(PSNR).Its average reconstruction time is similar to the FBP algorithm,both in the order of 10 ms,which is about 84 times faster than the MLEM algorithm.