BEP: BIDS-CND Extension for Continuous Sensory Tasks

[satwik-chandra]

BIDS-CND Extension for Continuous Sensory Tasks

The proposed extension to the BIDS specification aims to address the need for a domain-specific data structure for the study of natural sensory perception. Relevant scenarios involve the recording of neural data during tasks involving, for example, natural speech listening, music listening, and movie-watching tasks, where the stimulus can be seen as a continuous sensory stream. This type of scenario has become progressively more used in the last decade or so, as it allows for the study of cognition with tasks that are close to being ecologically valid. However, our field lacks detailed specifications on how data (neural data and stimuli) from these experiments should be stored and shared, making both replication of findings and re-analyses laborious and prone to errors.

Here we propose a BEP that would tackle these issues. This particular BEP is peculiar in that it specializes in a particular set of experimental scenarios (across sensory modalities) while being applicable to various recording modalities (e.g., EEG, MEG, iEEG). The first version of the specifications has already been formulated as part of the Cognition and Natural Sensory Perception initiative (CNSP) and has been discussed as part of the CNSP-Workshop 2022 (https://cnspworkshop.net/). As a result, the CNSP community converged on a set of resources for storing and analyzing neural data (tested on EEG, MEG, and iEEG) for this particular scenario. We referred to the first version of the data structure as Continuous-Event Neural Data (CND).

Here, we propose the development of a CND-BIDS (Continuous-Event Neural Data BIDS) structure, with the goal of adapting the CND structure to the principles of BIDS. CND-BIDS would be immediately compatible with toolboxes specific to the natural sensory perception domain, such as the mTRF-Toolbox, the Eelbrain Toolkit, and other tools that we are designing as part of the CNSP initiative.

In sum, CND-BIDS would be suitable for multiple neural recording methodologies (e.g., EEG, MEG, iEEG) but domain/task-specific (natural sensory processing), providing just enough constraints to make data sharing easy and consistent across research labs, facilitating replication and re-analysis.

We will provide conversion scripts between CND-BIDS and other data structures. Please find below a design concept for an EEG dataset where participants listened to an audio book (https://cnspworkshop.net/resources.html):

---

---

We look forward to your feedback and contributions to the development of this extension. @satwik-chandra @diliberg @arnndffr @mickcrosse

[sappelhoff]

Hi @satwik-chandra, thanks for your proposal. What aspects are currently missing in BIDS that prevent you from organizing your data using the existing standard? What key extensions are needed (conceptually)?

[satwik-chandra]

Hi @sappelhoff, thank you for your response!

The current BIDS specification can be used for what we are describing, as it is a general purpose. However, there are currently no guidelines on how to save this kind of data, meaning that two different labs may save their data using BIDS but obtain different results. What we are proposing is innovative and necessary in two main ways:

1) CND-BIDS would be a constrained version of BIDS, with guidelines that are specific to the particular domain of research involving natural sensory experiments. This would guarantee consistency across experiments and labs, facilitating the development of standardized code and the rapid replication of results.

2) CND-BIDS would also provide specifications for stimulus storage and preprocessing.

CND-BIDS is specific to experimental scenarios involving natural sensory tasks, which is an additional constraint compared with BIDS, but it remains applicable to various recording modalities, such as EEG, MEG, and iEEG/sEEG (and potentially eye-tracking/pupillometry in the future).

We are happy to discuss why all of this is useful, of course. While our initial idea was to use BIDS with this particular set of guidelines as a way to store and use data within our lab (Di Liberto-lab), we discussed this standard with the CNSP community at the CNSP workshops 2021 and 2022 (with overall about 200 participants so far) and concluded that pushing this a step further could be useful to many colleagues.

[sappelhoff]

I see. So my understanding is that you currently can accommodate all your data under existing BIDS principles, rules, and features -- but that the rules are "loose enough" so that different curators may end up with two different BIDS datasets given the same source data.

It'd be interesting to know how big this deviation would be, but my initial reaction to this is that that's not a problem, as both datasets would/should be well documented and understandable by humans / readable by software.

I think writing guidelines on how BIDS principles are to be interpreted in specific situations are great and important -- but I don't think that it would qualify as BEP material, i.e., not content that should be part of the specification.

What do others think? @bids-standard/maintainers @bids-standard/steering

[Remi-Gau]

I think before starting another BEP I would be curious how much what is being proposed cannot already be leveraged by using Hierarchichal events descriptors that are supported by BIDS.

I am pretty sure that HED could be used to tag the stimuli part of what you suggest and also the neural part. But I may be wrong.

Pinging @VisLab

[satwik-chandra]

Hi @sappelhoff,

Thank you for your suggestion regarding the adaptation of BIDS guidelines for this issue. We are open to exploring this approach and would appreciate your insights on the best way to proceed. Would you be able to refer to an existing example of this process that we can look into for reference?

Our main concern is ensuring that the specific guidelines developed for the CND case are easily searchable and accessible to the community focused on natural sensory tasks.

Our goal is to ensure that these specific guidelines remain easily accessible and not lost amidst the larger collection of general guidelines.