Medical Image Fusion And Its Application Of Brain Imaging

BackgroundToday,clinical technicians are increasingly applying medical image registration and fusion techniques because images from multiple sources can provide doctors with more comprehensive and accurate information in diagnosing the condition.Magnetic resonance imaging(MRI)and positron emission tomography(PET)are two commonly used brain imaging techniques.Magnetic resonance imaging(MRI)is anatomical imaging,which has a high spatial resolution.And positron emission tomography(PET)is a functional imaging,which provides information on human physiological metabolism,but the organizational structure display effect is poor.They have their own advantages,and doctors hope that the two can be combined.ObjectiveThis project aims to design an improved PET/MRI brain image fusion system based on neural network,which mainly solves the problem of feature registration and information fusion of multi-modality images acquired by different imaging devices PET and MRI.MethodsIn this field,the commonly used related methods of medical image registration and fusion were carefully studied,and the results of relevant methods were compared and analyzed.Then an improved neural network-based PET/MRI brain was designed.Image information fusion method.The specific implementation steps are:1.The PET image to be merged(index image)is subjected to IHS conversion and color conversion and converted into IHS and RGB channel information.2.using PET,MRI rigid registration based on the outline of the brain pixels,through translation,rotation,scaling,so that the brain anatomical position alignment.3.Using NSCT transform to obtain frequency characteristics in different directions in different directions,designing low-frequency information fusion rules based on local-domain weighted average method and PCNN high-frequency information fusion rules for spatial frequency excitation to achieve RGB three-channel PET image and MRI grayscale Image information fusion.4.Using other common image fusion methods to fuse the same source image,compare the results of various fusions.ResultIn order to have a relatively intuitive comparison of the effects of the various methods,the quality of the fused image was measured using standard deviation,entropy,sharpness,average gradient,and Qabf.Standard deviation reflects the degree of dispersion of gray levels;entropy represents the amount of information in the image,the greater the entropy,the greater the amount of information contained;sharpness reflects the ability of the image to express small details;the average gradient reflects the edge details of the image;Qabf Measure the fusion effect of the edge,and combine the structural similarity of the result with the source image.The parameters of this design reached 150.2410,8.5421,15.320,0.0926,and 0.5714,respectively,higher than the other methods.ConclusionThe fusion method based on artificial neural network makes full use of the neural network's feature recognition ability and self-learning ability.In combination with the multi-scale transformation method,image fusion is achieved through the selection of contrast ratio,and the functional information provided by the PET image and the anatomy provided by the MRI image are fully utilized.The information of the soft tissue greatly improved the doctor's diagnostic efficiency and accuracy.The fusion effect retained the outline,texture and other characteristics to the greatest extent,the details were more prominent,and the subjective effect was the best.