Bayesian Quantification of Contrast-Enhanced Ultrasound Images with Adaptive Inclusion of an Irreversible Component
Journal article
Rizzo, G, Tonietto, M, Castellaro, M, Raffeiner, B, Coran, A, Fiocco, U, Stramare, R and Grisan, E (2017). Bayesian Quantification of Contrast-Enhanced Ultrasound Images with Adaptive Inclusion of an Irreversible Component. IEEE Transactions on Medical Imaging. 36, pp. 1027-1036. https://doi.org/10.1109/TMI.2016.2637698
Authors | Rizzo, G, Tonietto, M, Castellaro, M, Raffeiner, B, Coran, A, Fiocco, U, Stramare, R and Grisan, E |
---|---|
Abstract | Contrast Enhanced Ultrasound (CEUS) is a sensitive imaging technique to assess tissue vascularity and it can be particularly useful in early detection and grading of arthritis. In a recent study we have shown that a Gamma-variate can accurately quantify synovial perfusion and it is flexible enough to describe many heterogeneous patterns. However, in some cases the heterogeneity of the kinetics can be such that even the Gamma model does not properly describe the curve, with a high number of outliers. In this work we apply to CEUS data the single compartment recirculation model (SCR) which takes explicitly into account the trapping of the microbubbles contrast agent by adding to the single Gamma-variate model its integral. The SCR model, originally proposed for dynamic-susceptibility magnetic resonance imaging, is solved here at pixel level within a Bayesian framework using Variational Bayes (VB). We also include the automatic relevant determination (ARD) algorithm to automatically infer the model complexity (SCR vs. Gamma model) from the data. We demonstrate that the inclusion of trapping best describes the CEUS patterns in 50% of the pixels, with the other 50% best fitted by a single Gamma. Such results highlight the necessity of the use ARD, to automatically exclude the irreversible component where not supported by the data. VB with ARD returns precise estimates in the majority of the kinetics (88% of total percentage of pixels) in a limited computational time (on average, 3.6 min per subject). Moreover, the impact of the additional trapping component has been evaluated for the differentiation of rheumatoid and non-rheumatoid patients, by means of a support vector machine classifier with backward feature selection. The results show that the trapping parameter is always present in the selected feature set, and improves the classification. © 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works |
Keywords | automatic relevance determination; contrast enhanced ultrasound; pixel-wise analysis; quantification; SVM; variational bayes |
Year | 2017 |
Journal | IEEE Transactions on Medical Imaging |
Journal citation | 36, pp. 1027-1036 |
Publisher | Institute of Electrical and Electronics Engineers (IEEE) |
ISSN | 0278-0062 |
Digital Object Identifier (DOI) | https://doi.org/10.1109/TMI.2016.2637698 |
Web address (URL) | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85018520273&doi=10.1109%2fTMI.2016.2637698&partnerID=40&md5=8467547d14b7c2d9081c4aeb98427f62 |
Publication dates | |
03 Apr 2017 | |
Online | 08 Dec 2016 |
Publication process dates | |
Accepted | 06 Dec 2016 |
Deposited | 23 Jan 2020 |
Accepted author manuscript | License File Access Level Open |
https://openresearch.lsbu.ac.uk/item/88xq5
Download files
109
total views110
total downloads0
views this month2
downloads this month