Application Of Improved PCNN In Multimodal Medical Image Fusion

Generally speaking,single-modal medical images contain comprehensive information that cannot reflect the lesion area well.In order to better observe the condition of the lesion area,medical image fusion technology is widely used in the field of medical images as a common technical means.Multimodal Medical Image Fusion(MMIF)mainly fuses multiple singlemodality medical images.In the process of fusion,not only can effective information be retained,complementary information can be added,and redundant information can be removed,but also the medical images after fusion processing are clearer.Doctors can obtain a variety of effective information in a fused medical image to provide a more reliable basis for their medical diagnosis.At present,medical image fusion methods are divided into spatial domain,transform domain and deep learning methods.The methods based on the spatial domain easily lead to distortion of the fused image.The method based on deep learning requires a large amount of data training to achieve the purpose of fusion,and the implementation process is complicated.In contrast,the transform domain-based method not only has good fusion effect,but also is simple and easy to implement.Pulse Coupled Neural Network(PCNN),as a single-layer network,does not require data training,and is widely used in the field of MMIF due to its excellent feature extraction ability.However,PCNN parameters need to be manually set based on experience,and poor parameter values will directly affect the fusion effect,resulting in incomplete retention of fusion image information and poor visual effects.Therefore,this thesis proposes to improve the image fusion algorithm of PCNN in the Non-Subsample Shearlet Transform(NSST)domain.The main research contents are as follows:(1)Aiming at the problem of complex parameter setting of PCNN in the process of traditional medical image fusion,a method is proposed that combines the variation part of Particle Swarm Optimization(PSO)and Differential Evolution(DE)to optimize Methods of PCNN parameters.First,the source image is decomposed into several high frequency subband and one low frequency subband of the same size in the NSST domain.Second,Improved Sum Modified Laplacian(ISML)is used as input to excite the PCNN model.The amplitude constant Vθ,attenuation constant αθ and link strength β of PCNN are optimized by the PSO-DE model,and the optimized parameter values are substituted into PCNN to realize the fusion of high frequency subband.Next,the low-frequency components are fused using a method combining Weighted Local Energy(WLE)and Average Gradient(AG).Finally,the fused high-frequency components and low-frequency components are subjected to NSST inverse transformation to obtain the final fused image.In this thesis,the fusion experiments of different modalities of brain anatomy medical images are compared with the results of other five algorithms,which proves the superiority of the proposed method in subjective visual effects and objective evaluation indicators.(2)In view of the problems that the edge preservation effect of organs and tissues is not good when the functional image and the anatomical image are fused,and the link strength β of PCNN is manually set to a fixed value,which affects the fusion effect,this thesis proposes a fusion method that choosing Multi-Scale Morphological Gradient(MSMG)as link strength βadjusting PCNN.First,the functional image is decomposed into intensity,chroma,and saturation components by IHS transform.Second,the NSST transform is used to decompose the intensity components from the high and low frequency subband of the anatomical image.The fusion of high-frequency sub-bands adopts the sum of difference of two squares(SDS)excitation and MSMG self-adaption PCNN as the fusion rule.The low-frequency components are fused using the regional ISML coefficient weighting method.Next,the fused highfrequency and low-frequency sub-bands undergo inverse NSST transformation to obtain fused intensity components.Finally,the fused intensity component,chrominance and saturation components undergo IHS inverse transformation to obtain the final fused image.Through the simulation experiment of fusion of functional image and anatomical image,compared with other five fusion algorithms,it is proved that the algorithm proposed in this thesis can effectively preserve the edge information of the source image and the fused image is clearer.