Enhancing Multimodal Medical Image Fusion Using a Markov Discriminator in Generative Adversarial Networks

Multimodal medical images, comprising anatomical and functional images, offer complementary insights into organ structure and metabolism. Anatomical images depict internal organ structures, whereas functional images illustrate metabolic activity but lack detailed structural information. Multimodal image fusion integrates data from different sensors to create images enriched with diverse semantic content, overcoming the limitations of single-modality imaging. Current fusion methods based on generative adversarial networks (GANs) use discriminators that convolve the entire input image, which can reduce efficiency and result in detail loss. To address this, we propose a GAN framework with a Markov discriminator that leverages local (Markov) properties. By redesigning the discriminator and formulating the loss function based on Markov correlation principles, our method focuses on local areas, thereby enhancing network performance and preserving finer details in the fusion images.Experimental results demonstrate that our approach produces fusion images with significantly improved detail retention and superior performance compared to conventional methods.