Quantitative Myocardial Blood Flow Measurement Of Cardiac PET Images Based On Factor Analysis

Positron Emission Tomography(PET)is one of state-of-the-art technology of nuclear medicine and represents the highest level in the field of nuclear medicine.As a noninvasive medical checkup,PET plays a very important role in the diagnosis of oncology,neurological and cardiovascular diseases.Its working principle is to inject the radionuclide labeled biomolecules into the body of the subject and thus to scan within the effective field of vision.Finally,the biodistribution of the drug in the subject can be visualized by computerized three-dimensional reconstruction.Dynamic PET imaging can provide tracer distribution at successive time points.Later on,kinetic models can be used to obtain the functional parameters of tissues and organs to achieve the quantitative analysis of biochemical changes and dynamic metabolic processes in vivo target organs.Coronary artery disease(CAD)has become a major disease that threatens human health.Myocardial perfusion imaging provides an important imaging method for diagnosis,evaluation of the curative effect and prognosis.Myocardial perfusion imaging currently has single photon emission tomography Single Photon Emission Tomography(SPECT)and Positron Emission Tomography(PET).Because of its higher spatial resolution and sensitivity than SPECT and better tracer tissue uptake,PET has been used in myocardial perfusion imaging to increase its use year by year and has become the gold standard for evaluating whether myocardium is survival.In dynamic PET myocardial imaging,estimating the blood input function and applying it to the kinetic model can provide abnormal myocardial flow(MBF),myocardial flow reserve(MFR)and other valuable clinical diagnostic information.In the application of kinetic model,the estimation of the blood input function is very critical,the accuracy of the results directly affect the quantitative analysis of late.The widely used method for estimating the blood input function is based on the region of interest method.The method is simple and easy to use,but it depends on the region of interest manually drawn by the doctor.The accuracy of the region of interest is influenced by the doctor’s personal experience and partial volume effect influences.As a statistical technique for extracting common factors from a variable population,Factor Analysis has been used to noninvasively extract time-activity curves of blood input functions and tissues from dynamic images.The factor analysis models based on principal component analysis,independent component analysis,least square method and maximum likelihood method have been proposed.However,due to the influence of noise and partial volume effect,the accuracy of the traditional factor analysis model extraction results is lower,which greatly limits the clinical application.There are two problems in the current factor analysis model.First,the method is based on either the least squares measure based on Gaussian noise model or the maximum likelihood measure based on Poisson noise model.However,the dynamic acquisition of dynamic PET image noise is difficult to describe directly by the Gaussian or Poisson model.Second,the result obtained by factor analysis is non-unique.The assumption of traditional method is that the overlap between each factor image is the smallest to solve the problem.Nonetheless,myocardial signals are often mixed with 10%-15%of the blood signal due to PET image resolution and partial volume effects.The fact is not taken into account to assumption of traditional method.In view of the above analysis,we purpose a novel factor analysis method based on kinetic cluster and a-divergence measure to resolve two problems above.By selecting different a values,the noise model of PET image can be more accurately modeled.In terms of uniqueness constraints,we use the pixel kinetic clustering of PET images as a priori constraint,and finally obtain the time activity curve and the blood input function of each tissue.The model was applied to 82Rb PET myocardial perfusion simulation data.The experimental results showed that kinetic cluster and a-divergence measure as priori constraint based on kinetic clustering both had a positive effect on the accuracy of tissue activity curve extraction.Under high noise(SNR = 5)can still accurately extract the time activity curve and blood input function of each tissue,which has outstanding performance in visual evaluation and quantitative evaluation.