Ensemble CycleGAN for Retrospective Rigid Motion Correction in MRI

Magnetic Resonance Imaging (MRI) is crucial for diagnosis but is vulnerable to patient movement, leading to artifacts and lower diagnostic accuracy. This paper introduces a novel Retrospective Motion Correction (RMC) approach using an Ensemble Cycle-consistent Generative Adversarial Network (CycleGAN). It combines three distinct CycleGAN models, each trained on a specific motion type, boosting its ability to correct that particular artifact. Combining these models into an ensemble leverages their strengths, offering a more effective solution for motion artifact correction in MRI. The motion-corrupted data is generated using a comprehensive Motion Simulation Tool. The method's effectiveness is proven on brain slices, measured by the Structural Similarity Index Measure (SSIM) and Normalized Mean Squared Error (NMSE). Results show that our Ensemble CycleGAN approach surpasses the performance of a single model, demonstrating enhanced capability in correcting motion artifacts.