Fast Reconstruction Enhances Deep Learning PET Head Motion Correction

Head motion correction poses a major challenge in brain PET imaging because even small-magnitude motion can significantly degrade image quality and introduce artifacts. In this study, we propose a new framework for PET head motion correction that builds upon previous work and takes fast reconstructions as input. The proposed method has three main characteristics: (i) the adoption of a high-resolution short-frame fast reconstruction workflow; (ii) the development of a novel encoder for PET data representation extraction; and (iii) the implementation of data augmentation techniques. Ablation studies are conducted to assess the individual contributions of each of these design choices. Furthermore, we conduct a multi-subject study on an 18F-FPEB dataset and qualitatively and quantitatively evaluate the method performance through comparison of the Motion-compensation OSEM List-mode Algorithm for Resolution-Recovery Reconstruction (MOLAR) reconstruction study and corresponding brain Region of Interest (ROI) Standard Uptake Values (SUV). Additionally, we contrast our method with a conventional intensity-based registration method. Our results demonstrate that the proposed method outperforms other methods on all subjects and can accurately estimate motion for subjects outside of the training set.