Identifying And Restoring Respiratory Motion Blur In PET/CT Images

Motion blur in ungated-PET/CT images induced by respiratory motion mayinfluence the accuracy of tumor diagnosis and treatment in the quantitative analysis.In this paper, a method of restoring the motion blur, which combines high-frequencysinusoidal vibration model and deconvolution technology, was proposed. This was aimage post-processing method. It corrected the motion blur by identifying thedirection and extent of movement from the blurred PET image, obtaining the pointspread function, and restoring the motion blur by deconvolution. Compared withgated-PETs, this method required neither additional tracking devices nor significantamount of computing resources, so it is more cost-effective.The blur direction was identified by performing Radon transform on thequasi-cepstrum of motion blurred images. The blur extent was identified frombispectrum of motion blurred image which has been rotated to make the motiondirection vertical. Based on this, Richardson-Lucy iterative algorithm was used torestore respiratory motion blurs. The Shannon entropy, tumor volume and meanstandard uptake value (SUV) were used to quantificationally evaluate the effect ofimage restoration. In order to improve the restoring effect, pre-segmentation andnoise-reducing processes were used before restoring the blurred image. The proposedmethod has been tested by simulation images, phantom and clinic PET images. Theblur direction and extent identified from blued images were found to be able to followthe true values very well. After image restoration, the estimation errors on the tumorvolume and mean standard uptake value (SUV) have been reduced obviously. Thisresult shows that the method proposed in this paper may restore the motion blur PETimages effectively, and will be helpful for tumor diagnosis.