Consistent MRI Reconstruction with Diffusion Models and Sequential Monte Carlo

Motivation: MRI reconstruction remains unreliable due to high variance in quality, limiting its use in clinical diagnostics. Our research aims to reduce this variability, enhancing reliability for medical applications. Goal(s): Develop a method to reduce reconstruction variance while improve their quality. Approach: We applied Sequential Monte Carlo with a diffusion model prior, integrating a twisting function to reduce reconstruction variance. Results: Our method successfully reduces MRI reconstruction variance, consistently achieving higher PSNR, SSIM, and LPIPS scores than baseline methods. Impact: This method enhances MRI reconstruction consistency, increasing reliability for clinical use and establishing a foundation for broader adoption in diagnostic imaging and medical diagnostics.