Inhomogeneity Correction in Magnetic Resonance Images Using Deep Image Priors

Shuo Han, Jerry L. Prince, Aaron Carass

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Intensity inhomogeneity in magnetic resonance (MR) images can decrease the performance of image processing, such as segmentation and registration. In this work, we propose an unsupervised learning approach to correct the inhomogeneity of an MR image based on deep image priors (DIPs). In DIPs, the structure of the convolutional neural networks was previously shown to capture the prior probability of an image, which has been demonstrated in several applications such as image denoising, segmentation, and super resolution. To obtain an inhomogeneity-free MR image, the problem was formulated in a Bayesian inference framework. The priors of the image and inhomogeneity field were captured by two DIPs and their likelihood was modeled based on the observed image. The approximated expectation of the posterior was calculated to get the corrected image using a stochastic gradient Langevin dynamics algorithm. Since we modeled the noise distribution, the proposed method is simultaneously capable of denoising to some extent. We compared our method with N4, a popular inhomogeneity correction method, in a simulated data set and a couple of real data sets, statistically showing that it has comparable or even superior performance than N4 when the inhomogeneity is severe or noise is high.

Original languageEnglish (US)
Title of host publicationMachine Learning in Medical Imaging - 11th International Workshop, MLMI 2020, Held in Conjunction with MICCAI 2020, Proceedings
EditorsMingxia Liu, Chunfeng Lian, Pingkun Yan, Xiaohuan Cao
PublisherSpringer Science and Business Media Deutschland GmbH
Pages404-413
Number of pages10
ISBN (Print)9783030598600
DOIs
StatePublished - 2020
Event11th International Workshop on Machine Learning in Medical Imaging, MLMI 2020, held in conjunction with the 23rd International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2020 - Lima, Peru
Duration: Oct 4 2020Oct 4 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12436 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference11th International Workshop on Machine Learning in Medical Imaging, MLMI 2020, held in conjunction with the 23rd International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2020
CountryPeru
CityLima
Period10/4/2010/4/20

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'Inhomogeneity Correction in Magnetic Resonance Images Using Deep Image Priors'. Together they form a unique fingerprint.

Cite this