Section: New Results

Research axis 3: Management of Information in Neuroimaging

In the context of population imaging, we have made progress in three main areas this year. First we were involved in the development of infrastructures for open science with OpenAIRE, we also participated in the collaborative definition of standards that will ensure that infrastructures remain interoperable. Finally, we started a research new axis looking at how variations in analytical pipelines impact neuroimaging results (i.e. analytic variability).

Infrastructures

Open research: linking the bits and pieces with OpenAIRE-connect

Participants : Camille Maumet, Christian Barillot, Xavier Rolland.

Open research is growing in neuroimaging. The community — supported by funders who want best use of public funding but also by the general public who wants more transparent and participatory research practices — is constantly expanding online resources including: data, code, materials , tutorials, etc. This trend will likely amplify in the future and is also observed in other areas of experimental sciences. Open resources are typically deposited in dedicated repositories that are tailored to a particular type of artefact. While this is best practice, it makes it difficult to get the big picture: artefacts are scattered across the web in a multitude of databases. Although one could claim that the publication is here to link all related artefacts together, it is not machine-readable and does not me toallow searching for artefacts using filters (e.g. all datasets created in relation with a given funder). We presented OpenAIRE-connect, an overlay platform that links together research resources stored on the web: https://beta.ni.openaire.eu/ [45].

This work was done in collaboration with Dr. Sorina Caramasu-Pop and Axel Bonnet from Creatis in Lyon and with collaborators of the OpenAIRE-Connect project.

Standardisation and interoperability

The best of both worlds: using semantic web with JSON-LD. An example with NIDM-Results & Datalad

Participant : Camille Maumet.

The Neuroimaging data model (NIDM-Results) provides a harmonised representation for fMRI results reporting using Semantic Web technologies. While those technologies are particularly well suited for aggregation across complex datasets, using them can be costly in terms of initial development time to generate and read the corresponding serialisations. While the technology is machine accessible, it can be difficult to comprehend by humans. This hinders adoption by scientific communities and by software developers used to more-lightweight data-exchange formats, such as JSON. JSON-LD: a JSON representation for semantic graphs (“JSON-LD 1.1” n.d.) was created to address this limitation and recent extensions to the specification allow creating JSON-LD documents that are structured more similar to simple JSON. This representation is simultaneously readable by a large number of JSON-based applications and by Semantic Web tools. Here we review our work on building a JSON-LD representation for NIDM-Results data and exposing it to Datalad, a data-management tool suitable for neuroimaging datasets with built-in support for metadata extraction and search [44].

This work was done in collaboration with Prof. Michael Hanke from Institute of Neuroscience and Medicine in Julich and with members of the INCF.

Tools for FAIR Neuroimaging Experiment Metadata Annotation with NIDM Experiment

Participant : Camille Maumet.

Acceleration of scientific discovery relies on our ability to effectively use data acquired by consortiums and/or across multiple domains to generate robust and replicable findings. Efficient use of existing data relies on metadata being FAIR1 - Findable, Accessible, Interoperable and Reusable. Typically, data are shared using formats appropriate for the specific data types with little contextual information. Therefore, scientists looking to reuse data must contend with data originating from multiple sources, lacking complete acquisition information and often basic participant information (e.g. sex, age). What is required is a rich metadata standard that allows annotation of participant and data information throughout the experiment workflow, thereby allowing consumers easy discovery of suitable data. The Neuroimaging Data Model (NIDM)2 is an ongoing effort to represent, in a single core technology, the different components of a research activity, their relations, and derived data provenance3. NIDM-Experiment (NIDM-E) is focused on experiment design, source data descriptions, and information on the participants and acquisition information. In this work we report on annotation tools developed as part of the PyNIDM4 application programming interface (API) and their application to annotating and extending the BIDS5 versions of ADHD2006 and ABIDE7 datasets hosted in DataLad[40].

This work was led by Dr David Keator from UCI Irvine and done in collaboration with members of the INCF.

Quantifying analytic variability

Exploring the impact of analysis software on task fMRI results

Participant : Camille Maumet.

A wealth of analysis tools are available to fMRI researchers in order to extract patterns of task variation and, ultimately, understand cognitive function. However, this 'methodological plurality' comes with a drawback. While conceptually similar, two different analysis pipelines applied on the same dataset may not produce the same scientific results. Differences in methods, implementations across software packages, and even operating systems or software versions all contribute to this variability. Consequently, attention in the field has recently been directed to reproducibility and data sharing. Neuroimaging is currently experiencing a surge in initiatives to improve research practices and ensure that all conclusions inferred from an fMRI study are replicable. In this work, our goal is to understand how choice of software package impacts on analysis results. We use publically shared data from three published task fMRI neuroimaging studies, reanalyzing each study using the three main neuroimaging software packages, AFNI, FSL and SPM, using parametric and nonparametric inference. We obtain all information on how to process, analyze, and model each dataset from the publications. We make quantitative and qualitative comparisons between our replications to gauge the scale of variability in our results and assess the fundamental differences between each software package. While qualitatively we find broad similarities between packages, we also discover marked differences, such as Dice similarity coefficients ranging from 0.000-0.743 in comparisons of thresholded statistic maps between software. We discuss the challenges involved in trying to reanalyse the published studies, and highlight our own efforts to make this research reproducible [6].

This work was done in collaboration with Alexander Bowring and Prof. Thomas Nichols from the Oxford Big Data Institute in the UK.