Various medical image fusion challenges have been addressed by using convolutional neural network (CNN) based generative adversarial techniques. While CNNs are designed for small-scale processing, their inductive bias makes it difficult to learn contextual features. Our solution to this problem is modality-based image analysis, which utilizes vision transformers for their circumstantial compassion. The generator for the vision transformer uses a generative approach to a new cumulative local transformer (CLT) block, which associates with local convolutional and transformer modules as its fundamental bottleneck. The encoder and decoder image techniques are used to analyse the information bottleneck model, and different modality images, such as T1, T2, PD, and FLAIR, are designed to test the image evaluation matrix. Instead of creating separate fusion models for each source-target modality, we have developed a standard procedure. Multi-contrast magnetic resonance imaging (McMRI) relies on MRI scans to ascertain the missing fusion orders. We explored different existing approaches for comparative experimental analysis based on evaluation measures such as PSNR and SSIM. Our models demonstrate that LVT is significantly more successful than competing CNN- and transformer-based methods, making it the ideal choice.