The use of Hierarchical Event Descriptors with fMRI

Data-sharing is becoming a more common practice in neuroscience. Currently, the majority of shared neuroimaging data consist of fMRI recordings. Many of these datasets include taskbased fMRI, where participants engage with controlled sensory stimulation following task instructions. A major flaw is that these datasets are often minimally annotated. While these experiments can include complex events and event combinations, descriptions are often generalized on the level of conditions, at best allowing users to replicate the original analysis but leaving little opportunity to use more of the information about brain dynamics supporting cognition contained in the data to its full extent. The Hierarchical Event Descriptor (HED) system provides detailed and machine-actionable descriptions originally developed to describe neuroimaging experiments recording M/EEG data. HED has been applied successfully to EEG data to facilitate analysis across multiple datasets. Its current release (3G) provides a userfriendly environment for annotation of events using a sparse, human-readable tagging system (K. Robbins, this meeting), and several available tools facilitate annotation (D. Truong, this meeting). Applying HED to fMRI data is straightforward as fMRI and EEG experiments follow a similar structure. Despite the differences in the measured signals, both have similar requirements for analysis. By applying HED annotation to fMRI datasets researchers cannot only improve its reusability, but also benefit from the continuous development of HED-enabled tools that aid in developing and/or automatize parts of the data-analysis. Methods for fMRI analysis are developing rapidly; the use of multivariate rather than standard univariate analysis methods is increasingly common. Such experiments are particularly well suited for reanalysis, since they engage many aspects of cognition that can be linked to brain activity. Detailed information about these experiments, experiment events, and time structure is needed to use these valuable datasets to their full potential, and is also vital for helping researchers find relevant datasets in continuously growing databases. Applying the same HED standard to describe events and structure of fMRI and M/EEG experiments will also allow researchers to better connect results from these datasets on the basis of the underlying brain processes they record. Standardized, machine-actionable descriptors for data are vital to accelerating knowledge accumulation. HED- 3G provides a framework for detailed description that is not only useful and necessary for M/EEG datasets, but equally so for fMRI data.