Applicability test for reducing noise on PET dynamic images using phantom applying deep image prior
Abstract
Objective Positron emission tomography (PET) allows imaging of patho-physiological information as a form of rate constants from a dynamic image. The rate constant image(s) may be affected from noise on the dynamic image. We introduced an artificial intelligence technique of deep image prior (DIP) to reduce noise on dynamic images. Method We utilized a phantom filled with 18F-F- and 11C-flumazenil solutions in the main and sub-cylinders, respectively. The phantom was scanned by a Biograph mCT and dynamic images were obtained. DIP was applied to all slices involved in the dynamic images while introducing an index for choosing an optimal epoch with minimize the degree of noise. Then, decay rate images were generated and quantitative accuracy and quality were measured in the images. Results The obtained decay rates on images were not significantly different from those of the reference values. Coefficient of variances (CV) were smaller using DIP-based images than those from original images for the reconstructed and decay rate images, both in the 11C and 18F filled regions. Conclusion The present method for choosing a generated image during DIP is feasible for noise reduction in dynamic PET images, and for obtaining rate constant images with less noise.
Cite
@article{arxiv.2308.04707,
title = {Applicability test for reducing noise on PET dynamic images using phantom applying deep image prior},
author = {Nobuyuki Kudomi and Yukito Maeda},
journal= {arXiv preprint arXiv:2308.04707},
year = {2023}
}
Comments
22 pages, 2 tables, and 5 figures, and supplemental material