Medical image fusion under diffusion-inspired multiframe super-resolution framework

Medical doctors prefer high-quality images to provide more accurate and reliable treatments to patients. In achieving this pressing demand, scholars have proposed a range of image enhancement and restoration methods to improve the information density and quality of medical images. For decades, multimodal fusion has become a widely accepted method to serve the purpose because it allows integration of useful information from medical imaging modalities, such as computed tomography, magnetic resonance imaging, single-photon emission computed tomography, and positron emission tomography. Despite the promising achievements, the current multimodal fusion methods generate images with spectral degradation that suffer from noise and unnecessary artifacts. These problems limit the usefulness of such methods in generating quality medical images for accurate diagnosis. This work presents a method, powered by the multiframe super-resolution (MSR) framework, to generate content-rich images with detailed features that can intuitively be analyzed and interpreted by radiologists. The proposed approach proceeds in three stages: estimation of motion parameters (rotation and translations) between pairs of input images, alignment, and fusion. In addition, the proposed method integrates a robust diffusion-driven regularization functional with multiple roles: addressing the MSR ill-posedness problem, recovering critical image features, suppressing noise, and removing spurious effects from input images. Extensive range of experiments demonstrate that the proposed method generates competitive values of entropy, peak signal-to-noise ratio (PSNR), and structural similarity (SSIM). Considering a specific case of the input image pair, for instance, the proposed method outperforms the closest competing approach by 5.76, 11.74, and 11.51% in terms of entropy, PSNR, and SSIM.