PresentationPDF Available

MultipleMS: A distributed workflow for managing and processing neuroimaging data

Authors:

Abstract

In this whitepaper, we describe one such neuroscience project and we present our solution at QMENTA implemented together with the UCSF School of Medicine, Department of Neurology. In the end, we summarize the benefits of our approach and inform you on how you can also use our platform with minimum effort.
A distributed workflow for managing and processing
neuroimaging data
In every large-scale neuroscience project, storing, managing, and
processing a large amount of neuroimaging data can be a challenge
especially when the data is heterogeneous, and sourced from multiple sites.
Besides getting all the acquired data together in
one convenient place and having people from
different centers distributed worldwide working
on the processing and analysis of the data, one
has to deal with the regulations for access
control and de-identification for the different
locations of the contributing sites.
In this whitepaper, we describe one such
neuroscience project and we present our
solution at QMENTA implemented together with
the UCSF School of Medicine, Department of
Neurology. In the end, we summarize the
benefits of our approach and inform you on
how you can also use our platform with
minimum effort.
2
Introduction
For a large multi-site study on patients with Multiple Sclerosis (MS), a large amount of
datasets will need to be uploaded over an extensive period of time (5 years). At the end
of the project, it is estimated for the project to contain a total of 3000-4000 MRI
datasets, uploaded from 16 sites around the world, including sites in USA and
Germany.
The data will be heterogeneous: there is
retrospective data, which consists mostly of
T1W and in some cases T2W or FLAIR MRI
scans, and there will be prospective data, which
is mostly T1W and FLAIR. All this data needs to
be stored and managed in a unified way.
Because the sites providing the data are
located in various countries, we have to adhere
to the laws and restrictions of the countries
involved. To avoid issues, we always apply rules
that are at least as strict as the most strict laws
involved.
After the data that has been uploaded, it is pre-
processed automatically. This pre-processing
may include bias-field correction, registration,
defacing, and conversion to different file
formats such as NIFTI. We use several publicly-
available tools for the pre-processing, which are
listed in the next section. When the pre-
processing is done, a specialist will review the
images and approve or disapprove the dataset
for inclusion in the study. This manual quality
control (QC) step is needed to avoid introducing
incorrect data in the study that was acquired
with, for example, wrong scanning parameters
or if there was any problem during file format
conversion.
3
The Project
For all approved datasets, the MS lesions need
to be segmented, as well as different brain
structures so that they can be later used to
derive measurements of cortical atrophy, brain
degeneration, lesion volume, white matter
integrity, etc.
Datasets can be easily and conveniently
uploaded through the browser, or using our
specialized QMENTA app which will
automatically anonymize the data before it is
actually sent to the platform.
The platform stores all data in a secure way by
applying proper de-identification and
encryption, and complies with the various
privacy and security laws in different countries.
This is elaborated upon in more detail in our
security whitepaper.
The goal is to automate this process as much
as possible. Each step however, will still need a
QC step where a specialist evaluates and
potentially edits the results, and can approve or
disapprove the results.
4
The QMENTA platform allows users from the different sites to collaborate on the
project. The project owner can add users, and can set the permissions per user and per
site.
Implementation
The datasets will be processed using several open source neuroimaging tools:
FSL (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSL) for registration and defacing in the pre-processing
step
MRtrix3 (http://www.mrtrix.org/) to convert DICOM to NIFTI file format in the pre-processing
step
LST: Lesion Segmentation Tool (http://www.statistical-modelling.de/lst.html) for SPM
to automatically segmentation of MS lesions
Lesion filling is by default our own algorithm, but can optionally use FSL or ANTS
Freesurfer image analysis suite (http://surfer.nmr.mgh.harvard.edu/) for skull stripping
and brain parcellation,
ANTS: Advanced Normalization Tools (http://stnava.github.io/ANTs/) for bias-field correction
(in pre-processing), skull stripping and brain parcellation.
5
Each of these tools is specialized in one or more neuroimaging tasks that are performed
automatically. By making these tools available in our system, we remove the overhead for the user
to download, install and configure the tools, and to apply them manually or scripted to many
datasets. The users can set the parameters of each of the tools using our platform, but proper
default values are chosen when the user does not set them.
All the tools and QC steps are combined together in a single MultipleMS Workflow which is depicted
in the diagram below:
Start Preprocess QC Lesion
Segmentation QC QC EndVolumetry
Pre-processing will automatically start after a
dataset has been uploaded, and after pre-
processing is done, the data will be added to
the queue for manual quality control (QC).
Users that have been assigned a QC role will
see a list of datasets to verify when they log
onto the platform. Several of the other
advanced analyses are followed by a QC step
as well.
For the QC after lesion segmentation, we use
MindControl¹, built on top of the Papaya²
medical image viewer for the user to review
and edit the detected MS lesions. Using this
automatic detection followed by manual QC
saves a lot of time compared to full manual
detection of lesions. The QC steps after the
other analyses consist of a simple approval
step where the specialist has to choose PASS
or FAIL to confirm or reject the results of an
analysis.
1. https://www.ncbi.nlm.nih.gov/pubmed/28365419, https://github.com/akeshavan/mindcontrol
2. https://github.com/rii-mango/Papaya
QMENTA offers a cloud-based neuroimaging platform for hospitals, research centers,
and pharmaceutical companies to help accelerate the discovery and development of
new treatments for brain diseases. It provides a unique infrastructure with scalable
computing power to help save valuable time, effort, and resources on intensive R&D
activities.
The QMENTA cloud platform supports the
entire R&D workflow, including data collection,
data management, image processing, 3D
visualization, and sharing. Images and clinical
scores can be easily uploaded to the cloud,
automatically anonymized, and managed from
the browser without the need to install or
maintain any software. All data is securely
stored in a HIPAA compliant environment with
fine-grained access settings. Proprietary, open-
source, and licensed tools such as FSL, MRtrix,
LST, ANTs and FreeSurfer are offered together.
Thus, the standardized computing environment
with these validated tools enables
reproducibility in research.
By using the QMENTA platform, the data is
stored securely in the cloud, so no need to
worry about missing CDs or broken hard
drives. The data is secure, accessible, and easy
to share with other users that have the right
permissions.
QMENTA platform
6
For the MultipleMS project, we have created a
new workflow in the QMENTA platform that
combines various open source neuroimaging
tools and manual quality control steps. Using
this workflow on our platform enables
researchers and specialists worldwide to
effectively and efficiently work together. It
helps them to easily view the data uploaded
from different sites, automatically starts pre-
processing of the data, and ties together a
variety of available neuroimaging tools to
acquire the desired results. Individual
processing steps are followed by manual QC
steps by researchers and specialists to guard
the overall quality of the project. The combined
automatic processing combined with manual
QC saves a lot of time when compared to other
approaches.
Contact
To learn more about our recent
developments, you can regularly check our
website at www.qmenta.com
For more information about MultipleMS go
to www.multiplems.eu
Please do not hesitate to reach out to
info@qmenta.com for any questions,
concerns, or feedback.
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.