Conditional Variational Autoencoder for Learned Image Reconstruction

Learned image reconstruction techniques using deep neural networks have recently gained popularity and have delivered promising empirical results. However, most approaches focus on one single recovery for each observation, and thus neglect information uncertainty. In this work, we develop a novel co...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Chen Zhang, Riccardo Barbano, Bangti Jin
Formato:	article
Lenguaje:	EN
Publicado:	MDPI AG 2021
Materias:	conditional variational autoencoder uncertainty quantification deep learning image reconstruction Electronic computers. Computer science QA75.5-76.95
Acceso en línea:	https://doaj.org/article/09695f2192ce4215aa4579534d43eb9d
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:09695f2192ce4215aa4579534d43eb9d
record_format	dspace
spelling	oai:doaj.org-article:09695f2192ce4215aa4579534d43eb9d2021-11-25T17:17:12ZConditional Variational Autoencoder for Learned Image Reconstruction10.3390/computation91101142079-3197https://doaj.org/article/09695f2192ce4215aa4579534d43eb9d2021-10-01T00:00:00Zhttps://www.mdpi.com/2079-3197/9/11/114https://doaj.org/toc/2079-3197Learned image reconstruction techniques using deep neural networks have recently gained popularity and have delivered promising empirical results. However, most approaches focus on one single recovery for each observation, and thus neglect information uncertainty. In this work, we develop a novel computational framework that approximates the posterior distribution of the unknown image at each query observation. The proposed framework is very flexible: it handles implicit noise models and priors, it incorporates the data formation process (i.e., the forward operator), and the learned reconstructive properties are transferable between different datasets. Once the network is trained using the conditional variational autoencoder loss, it provides a computationally efficient sampler for the approximate posterior distribution via feed-forward propagation, and the summarizing statistics of the generated samples are used for both point-estimation and uncertainty quantification. We illustrate the proposed framework with extensive numerical experiments on positron emission tomography (with both moderate and low-count levels) showing that the framework generates high-quality samples when compared with state-of-the-art methods.Chen ZhangRiccardo BarbanoBangti JinMDPI AGarticleconditional variational autoencoderuncertainty quantificationdeep learningimage reconstructionElectronic computers. Computer scienceQA75.5-76.95ENComputation, Vol 9, Iss 114, p 114 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	conditional variational autoencoder uncertainty quantification deep learning image reconstruction Electronic computers. Computer science QA75.5-76.95
spellingShingle	conditional variational autoencoder uncertainty quantification deep learning image reconstruction Electronic computers. Computer science QA75.5-76.95 Chen Zhang Riccardo Barbano Bangti Jin Conditional Variational Autoencoder for Learned Image Reconstruction
description	Learned image reconstruction techniques using deep neural networks have recently gained popularity and have delivered promising empirical results. However, most approaches focus on one single recovery for each observation, and thus neglect information uncertainty. In this work, we develop a novel computational framework that approximates the posterior distribution of the unknown image at each query observation. The proposed framework is very flexible: it handles implicit noise models and priors, it incorporates the data formation process (i.e., the forward operator), and the learned reconstructive properties are transferable between different datasets. Once the network is trained using the conditional variational autoencoder loss, it provides a computationally efficient sampler for the approximate posterior distribution via feed-forward propagation, and the summarizing statistics of the generated samples are used for both point-estimation and uncertainty quantification. We illustrate the proposed framework with extensive numerical experiments on positron emission tomography (with both moderate and low-count levels) showing that the framework generates high-quality samples when compared with state-of-the-art methods.
format	article
author	Chen Zhang Riccardo Barbano Bangti Jin
author_facet	Chen Zhang Riccardo Barbano Bangti Jin
author_sort	Chen Zhang
title	Conditional Variational Autoencoder for Learned Image Reconstruction
title_short	Conditional Variational Autoencoder for Learned Image Reconstruction
title_full	Conditional Variational Autoencoder for Learned Image Reconstruction
title_fullStr	Conditional Variational Autoencoder for Learned Image Reconstruction
title_full_unstemmed	Conditional Variational Autoencoder for Learned Image Reconstruction
title_sort	conditional variational autoencoder for learned image reconstruction
publisher	MDPI AG
publishDate	2021
url	https://doaj.org/article/09695f2192ce4215aa4579534d43eb9d
work_keys_str_mv	AT chenzhang conditionalvariationalautoencoderforlearnedimagereconstruction AT riccardobarbano conditionalvariationalautoencoderforlearnedimagereconstruction AT bangtijin conditionalvariationalautoencoderforlearnedimagereconstruction
_version_	1718412510227333120

Conditional Variational Autoencoder for Learned Image Reconstruction

Ejemplares similares