Deep Learning-based Dynamic PET Intra-frame Motion Correction and Integration with Inter-frame Motion Estimation

Résumé fourni par la source

In dynamic PET imaging, intra-frame motion poses a significant challenge, leading to blurred reconstructed frames. Current methods using respiratory gating and conventional image registration can correct intra-frame motion but suffer from computational inefficiency and limitations in integrating with intra-frame motion estimation. To address these challenges, we propose to utilize a deep learning-based spatial-temporal motion estimation framework for both gated image registration and inter-frame registration. On a whole-body FDG dynamic PET dataset, our preliminary results demonstrate the generalization ability of the model on gated images and its efficacy across different reconstruction protocols. With only 1/400 of the inference time, the deep learning-based registration approach significantly improves intra-frame motion correction compared with conventional registration, leading to enhanced spatial alignment in gated images and improved image sharpness and contrast in reconstructed dynamic frames. This deep learning-based work shows promise in accelerating motion estimation processes and improving image quality in dynamic PET imaging.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.