CT-to-MRI translation of medical volume data based on an enhanced diffusion model

Abstract

Abstract In clinical practice, medical inter-modality imaging results can assist doctors in making better decisions, as different modalities imaging results can provide complementary information. Traditionally, obtaining these imaging results requires using various medical devices to scan patients, which can be time-consuming, costly, and potentially harmful to the patient. Motivated by the need to address these limitations, we propose an alternative method that facilitates the conversion of volume CT into volume MRI. The method is based on a Diffusion model and incorporates a post-processing approach to enhance the model’s output. To validate our approach, we conduct experiments and achieve good results on brain and pelvic datasets obtained from clinical practice, despite approximately 6% of the slices being incompletely paired. We also compare our method with state-of-the-art techniques, both qualitatively and quantitatively. Our experimental results show that our method outperforms state-of-the-art techniques, including MedSynthesisV1, CycleGAN, Pix2Pix and Diffusion, when using ground truth as a reference. Finally, we conduct an experiment to select the optimal hyperparameters, including the number of epochs and the parameters $$cutoffPercentage\_left$$ and $$cutoffPercentage\_right$$ .