Study On Post-Processing Of Oxygen Uptake Fraction Measurement Based On BOLD-fMRI

Cerebrovascular disease has always been a major threat to human life and health,and the prevalence rate is high among the elderly population.It is very important to prevent and treat ischemic stroke by evaluating the cerebral hemodynamic state and hemodynamic parameters.Oxygen Extraction Fraction(OEF,Oxygen Extraction Fraction)is one of the important parameters of human brain hemodynamics and tissue metabolism,which can indirectly reflect the physiological condition and lesion degree of brain tissue blood vessels.Auxiliary criteria,and to a certain extent can predict the condition of the lesion and the focus area,it can be seen that OEF has very important significance in medical diagnosis.PET(Position Emission Tomography)technology is currently the gold standard for detecting OEF.Since the radioactive elements used in the measurement process have irreversible damage to the human body,the OEF value can be compensated by MRI(Magnetic Resonance Imaging)to noninvasively Defects of PET technology.However,due to the inevitable influence of system noise and magnetic field inhomogeneity in the imaging process,the detection results are in error compared with the PET gold standard,so it is of great significance to perform image processing on the MRI image to improve the accuracy of OEF measurement.This paper mainly studies and improves the post-processing method of noninvasive measurement of cerebral oxygen uptake by MRI.The main work is as follows:1.Research on the mathematical model of OEF measurement by NMR technology: First,the basic principles of nuclear magnetic resonance and several common scanning sequences are described,and an improved nuclear magnetic resonance scanning sequence-GESSE combining spin echo sequence and gradient echo sequence(Gradient-Echo Sampling of Spin Echo)sequence.Then introduce the principle of functional magnetic resonance imaging(f MRI,functional magnetic resonance imaging)based on blood oxygen level dependent effect(BOLD,Blood oxygenation level dependent)and the OEF two-chamber model to provide theoretical support for the subsequent measurement of OEF.Finally,the OEF post-processing flow of nuclear magnetic resonance images is introduced in detail.The OEF two-compartment model can build a bridge between OEF andhemoglobin magnetization shift,and the GESSE sequence can capture the degree to which the magnetic resonance signal is attenuated by the magnetization shift.In the OEF post-processing model,the gray value of the nuclear magnetic resonance image is quantified by a mathematical formula to extract the transverse relaxation change coefficient of the magnetic resonance signal affected by the magnetization offset,and then gradually calculated through the relationship between the coefficient and the magnetization offset OEF value.2.Using the feedforward noise reduction convolutional neural network(Dn CNN)based on residual learning to perform noise reduction processing on f MRI:First,the noise characteristics of f MRI and the evaluation indexes of denoising performance are described,and the principles of Gaussian filtering and wavelet transform denoising are briefly introduced.Then introduce the basic principles and design ideas of Dn CNN,and explain the principles and advantages of residual learning and batch normalization theory,and how to use it with Dn CNN.Design the structure and parameters of Dn CNN,use the designed network to train the MRI images with different intensity noise,generate multiple sets of optimal training networks for different noise intensity,and input the test image sets with different intensity noise into the corresponding Network to test and output.Finally,according to the evaluation index of image denoising performance,the results of Gaussian denoising and wavelet denoising are compared to verify the advantages of the algorithm in image denoising.3.Explore the role and impact of the algorithm in this paper in improving the accuracy of OEF measurement: MR raw dicom images of the heads of normal subjects are used to input all the trained convolutional neural networks for noise reduction and output.Before and after the treatment,OEF post-processing experiments were performed to obtain the OEF value,and the group with the OEF value closest to the PET gold standard was selected as the final output.Through comparative analysis,it is verified that the OEF measurement value of the image after the noise reduction optimization process is closer to the PET gold standard.Therefore,the noise reduction optimization algorithm in this paper improves the accuracy of OEF measurement and provides beneficial help for the subsequent prevention of cerebrovascular diseases.