eScience Infrastructure for running validated image analysis pipelines:
how to best compare MRI scans from different medical centers
-
1
Radboud University Nijmegen, Dept. of Neuroinformatics, Donders Institute for Brain, Cognition and Behaviour, Netherlands
-
2
Erasmus MC, Biomedical Imaging Group Rotterdam, Depts. Medical Informatics and Radiology, Netherlands
-
3
Netherlands eScience Center, Netherlands
-
4
Radboudumc, Dept. of Geriatric Medicine/Radboud Alzheimer Centre, Donders Institute for Brain, Cognition and Behaviour, Netherlands
-
5
Maastricht University, Dept. of Psychiatry and Neuropsychology/Alzheimer Center Limburg, Netherlands
-
6
VU University Medical Center, Dept. of Radiology and Nuclear Medicine, Neuroscience Campus Amsterdam, Netherlands
-
7
Treparel, Netherlands
We recently introduced an eScience infrastructure for the secure sharing of neuroimaging data and running validated analysis pipelines on a high performance cloud [1]. We have populated this infrastructure with two thousand structural MR images from four Dutch medical centers. As a pilot project, we are segmenting the hippocampus for each of these images, thereby running into a number of practical issues. The most prominent question is whether the pipeline that we use, which has been tuned to perform optimally on data from a single MR scanner, can be directly applied to the four datasets, which differ in resolution, scanner type, and acquisition protocol. The most prominent step of the pipeline [2] is the registration of a set of twenty reference segmentations to the target scan in order to create a probabilistic atlas in target space. This is then combined with an intensity model, and the energy function is minimized via graph cuts. Ideally the pipeline would be able to accept new scans of unknown source, and use a standard set of manual segmentations for registration. We have however observed that the (nonlinear) registration performs worse when the source and target scans have dissimilar tissue intensity scales, which leads to an increased bias and variance of derived results such as the hippocampal volume.
An alternative approach is not to use a single set of manual segmentations for all data, but use separate segmentations for each cohort that is added to the platform. This introduces another type of bias when the manual segmentations have been carried out by different investigators using different criteria. We investigate whether the improved statistical power of combining cohorts outweighs the bias and variance introduced by the different scan parameters.
Two neuroinformatics tools are presented as components of the infrastructure:
1. A Java-based upload tool that takes care of client-side pseudomisation and subsequent upload to a central XNAT [3] server.
2. Fancylog (https://github.com/rbakker/fancylog), a Python-based logging system that presents, as the pipeline runs, all executed steps of the pipeline and their intermediate results in a dynamic webpage, as illustrated in Figure 1. It uses the XTK viewer [4] to display volumetric images in the browser.
Acknowledgements
Supported by the Netherlands eScience Center, grant 027.011.304
References
1. De Boer P, Ranguelova E, Ivanova M, Koek M, Van Der Lijn F, Niessen W, Versteeg A, Vrenken H, Burgmans S, Van Boxtel M, Meulenbroek O, De Leeuw F, Bakker R and Tiesinga P (2013). eScience Infrastructure for sharing neuroimaging data and running validated analysis pipelines on a high performance cloud. Front. Neuroinform. Conference Abstract: Neuroinformatics 2013. doi: 10.3389/conf.fninf.2013.09.00085
2. Van der Lijn F, den Heijer T, Breteler MM, Niessen WJ (2008) Hippocampus segmentation in MR images using atlas registration, voxel classification, and graph cuts. Neuroimage 43, 708-20.
3. Marcus, DS, Olsen T, Ramaratnam M, and Buckner, RL (2007) The Extensible Neuroimaging Archive Toolkit (XNAT): An informatics platform for managing, exploring, and sharing neuroimaging data. Neuroinformatics 5(1), 11-34.
4. Haehn D, Rannou N, Ahtam B, Grant E and Pienaar R (2014). Neuroimaging in the Browser using the X Toolkit. Front. Neuroinform. Conference Abstract: 5th INCF Congress of Neuroinformatics. doi: 10.3389/conf.fninf.2014.08.00101
Keywords:
XNAT,
python,
Neuroimaging methods,
atlas-based segmentation,
Multi-center,
Hippocampus
Conference:
Neuroinformatics 2014, Leiden, Netherlands, 25 Aug - 27 Aug, 2014.
Presentation Type:
Demo, not to be considered for oral presentation
Topic:
Neuroimaging
Citation:
Bakker
R,
Koek
M,
Ranguelova
E,
De Boer
PT,
Meulenbroek
O,
Burgmans
S,
Versteeg
A,
Vrenken
H,
Van Der Lijn
F,
Niessen
WJ and
Tiesinga
PH
(2014). eScience Infrastructure for running validated image analysis pipelines:
how to best compare MRI scans from different medical centers.
Front. Neuroinform.
Conference Abstract:
Neuroinformatics 2014.
doi: 10.3389/conf.fninf.2014.18.00088
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
27 Apr 2014;
Published Online:
04 Jun 2014.
*
Correspondence:
Dr. Rembrandt Bakker, Radboud University Nijmegen, Dept. of Neuroinformatics, Donders Institute for Brain, Cognition and Behaviour, Nijmegen, Netherlands, r.bakker@donders.ru.nl