Data-driven Gating for PET Using a Population-based Encoder

Le résumé fourni par la source

Respiratory motion is one of the main sources of motion artifacts in positron emission tomography (PET) imaging. In a previous study, we proposed a data-driven approach using an unsupervised deep clustering network that employs an autoencoder (AE) trained with patient-specific short-frame images to extract latent features for respiratory gating. The extracted latent vectors could subsequently be clustered for automatic gating, resulting in better performance than traditional phase gating. However, the patient-specific training is time consuming. In this work, we investigate a new, population-based AE that is pre-trained using an ensemble of patient data and compared it with the patient-specific AE. Once pre-trained, the population-based AE does not require any patient-specific training and thus reduce the computation time for respiratory gating. Experiment results show that a pre-trained AE with $N=42$ patients training dataset can achieve the same performance as the on-the-fly AE

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