2024-02-22 update

1. New data

1.1 Synchronized physio data tables

Added new tables (fnirs_sync, gsr_sync, eeg_sync, and ekg_sync) to the database to store preprocessed signals synchronized to a main clock for four modalities (fNIRS, GSR, EEG, and EKG) across experiments, with filtering for artifact removal. For each experiment, a main clock was defined with start time at 5 seconds before the beginning of the first task (rest state), end time at 5 seconds after the end of last task (Minecraft), and frequency at 200 Hz.

The changes are introduced in this PR: https://github.com/ml4ai/tomcat/pull/558.

The pipeline for synchronization that produces the data in the *_sync tables differs from the pipeline described in our NeurIPS 2023 paper (that produced the derived data products from 2023-08-28) in the following ways:

  • In the 2023-08-28 version of the derived data products, we used the same functions for filtering the GSR and EKG signals as we did for the EEG data. However, this was an error on our part. We now use the proper GSR and EKG filtering functions from the neurokit library.
  • In the 2023-08-28 version of the derived data products, we used the earliest timestamp for the modality among the participants as the starting point. However, this creates a different start time for each modality, which makes it challenging if a user wants to work with multiple modalities. So instead, we defined a shared main clock for each group session which starts 30 seconds before the first task and ends ends 30 seconds after the final task. All modalities are now mapped to the ticks in this clock (in the *_sync tables in the database).
  • In the 2023-08-28 version of the derived data products, synchronized signals were generated for 100 Hz, 500 Hz, and 1000 Hz. However, the new *_sync tables in the database only contain synchronized signals at 200 Hz, based on recent discussions with experts on brain data processing (100 Hz may lose important information, and 500/1000 Hz is excessive computationally expensive).

1.2 Vocalic features table

Added a table called audio_vocalics that contains vocalic features extracted using OpenSMILE.

1.3 Updated fnirs_raw table

Expanded fNIRS data by adding raw signals for 760 nm and 850 nm wavelengths to the fnirs_raw table, in case users wish to transform those into HbO and HbR themselves.

1.4 Added advisor information

Added a column to the group_session table containing information about which advisor was present during the trials.

2. Timestamp corrections

2.1 Timestamp corrections for exp_2023_01_31_14

Summary: Corrected Minecraft start/end timestamps for exp_2023_01_31_14 which had inconsistent dates in the future.

Explanation: exp_2023_01_31_14 had dates for trial start/stop and mission stop (same as trial stop due to the absence of a mission stop message in the corresponding .metadata file) in the past due to inconsistency. We noticed that the code was not taking the most recent trial dates into consideration in case of multiple trial start/stop. We fixed the data processing code and updated the dates manually. The wrong dates were causing the main clock not to have any ticks because rest state (with correct timestamp) had timestamp after the Minecraft task which is used to compute the final timestamp of the main clock when synchronizing signals.

2.2 Timestamps for rest state task in exp_2022_12_05_12.

Summary: Added start and end timestamps for the rest state task in exp_2022_12_05_12.

Explanation: If there are no start and end timestamps saved in the raw data file for the rest state task (prior to 2023-04, the timestamps for the rest state task were being saved in a CSV file), we use the file modification date as an approximation to the start timestamp when it started and add 300 seconds to the start timestamp to obtain the end timestamp. The file is created when the task start so the offset should be minimal. So far, only experiment exp_2022_12_05_12 has this issue. We use the start time of the rest state when defining the main clock, so having that information is important for us not to lose any data.

3. Fixed swapped signals for EKG and GSR for some experiments.

Addressed an issue where EKG and GSR signals were swapped in specific experiments:

  • exp_2022_09_30_10
  • exp_2022_10_04_09 for the lion station
  • exp_2022_10_04_09 for the tiger station) by updating the raw EEG data processing processing to swap the signals for these specific instances.

Other updates

  • Updated funding acknowledgment section on the landing page to include new grant from ARO.
  • Updated entity-relationship diagram.