To address the problems of limited information in single-modal ophthalmic imaging recognition and loss of edge information in medical image fusion

a medical image fusion model based on an encoder-decoder is proposed.In the encoder

the efficient channel attention(ECA)and omni-scale block(OSBlock)are integrated to extract shallow multi-scale features.Then

the Transformer-convolutional neural network(CNN)architecture is used to establish context-dependent relationships in parallel

in which the CNN part uses the residual pyramid module to extract deep-level information.The fusion module combines the fast Fourier transform and the residual module

and adds the spatial and channel reconstruction convolution(SCConv)module to reduce redundant computations.The decoder uses a symmetrical structure to reconstruct the fused image.The proposed algorithm demonstrates superior performance in training and testing on private ophthalmic datasets and the public Harvard Whole Brain Atlas dataset.