{"dataset":{"id":"45266","dataset_id":"on007720","name":"BCI-FIT: A customization protocol for communication brain-computer interface systems.","description":"BCI-FIT is a customization protocol dataset for non-implantable communication brain-computer interface (cBCI) systems, comprising EEG recordings from five participants with speech and/or physical impairments due to amyotrophic lateral sclerosis (ALS). Participants completed calibration and copy-spelling tasks using both customized and non-customized versions of a cBCI system across multiple visits, with iterative system adjustments based on performance, preferences, and signal characteristics. The dataset includes EEG data collected using dry electrode caps (DSI-Flex or DSI-24) sampled at 300 Hz during Matrix and RSVP typing paradigms.","owner_user_id":15,"status":"active","github_repo":"nemarDatasets/on007720","concept_doi":"10.82901/nemar.on007720","latest_version_doi":"10.82901/nemar.on007720.v1.0.0","created_at":"2026-06-19 04:01:42","updated_at":"2026-07-22 02:37:07","zenodo_concept_id":null,"is_sandbox":0,"visibility":"public","ezid_status":"public","enrichment_json":"{\n \"version\": \"2.0\",\n \"pipeline_stage\": \"validated\",\n \"title\": \"BCI-FIT: A customization protocol for communication brain-computer interface systems.\",\n \"description\": \"BCI-FIT is a customization protocol dataset for non-implantable communication brain-computer interface (cBCI) systems, comprising EEG recordings from five participants with speech and/or physical impairments due to amyotrophic lateral sclerosis (ALS). Participants completed calibration and copy-spelling tasks using both customized and non-customized versions of a cBCI system across multiple visits, with iterative system adjustments based on performance, preferences, and signal characteristics. The dataset includes EEG data collected using dry electrode caps (DSI-Flex or DSI-24) sampled at 300 Hz during Matrix and RSVP typing paradigms.\",\n \"methods_description\": \"EEG activity was recorded using dry electrode caps (DSI-Flex with 7 channels or DSI-24 with 24 channels) with hardware filtering (0.003-150 Hz bandwidth), sampled at 300 Hz with linked-ear reference and active common-mode follower. Data were downsampled to 150 Hz and filtered between 1-20 Hz (2nd-order) with 60 Hz notch filtering. Participants completed calibration tasks (55 inquiries per task) and copy-spelling tasks using Matrix or RSVP layouts presented at 5 Hz on a 24\\\" monitor. Testing was conducted on an MSI Trident 3 desktop with Nvidia RTX 3060 GPU and 12th-gen Intel i7 processor.\",\n \"license\": \"CC0\",\n \"dataset_type\": \"raw\",\n \"authors\": {\n \"Betts Peters\": {},\n \"Daniel Klee\": {},\n \"Michelle Kinsella\": {},\n \"Yonah Hendin\": {},\n \"Tab Memmott\": {},\n \"Matthew Lawhead\": {},\n \"Srikar Ananthoju\": {},\n \"Basak Celik\": {},\n \"Scott Spaulding\": {},\n \"Barry Oken\": {},\n \"Melanie Fried-Oken\": {}\n },\n \"keywords\": [\n {\n \"term\": \"Brain-Computer Interfaces\",\n \"subject_scheme\": \"MeSH\",\n \"value_uri\": \"http://id.nlm.nih.gov/mesh/D062207\"\n },\n {\n \"term\": \"EEG\"\n },\n {\n \"term\": \"amyotrophic lateral sclerosis\"\n },\n {\n \"term\": \"motor imagery\"\n },\n {\n \"term\": \"P300\"\n },\n {\n \"term\": \"communication disorders\"\n },\n {\n \"term\": \"assistive technology\"\n }\n ],\n \"related_identifiers\": [\n {\n \"identifier\": \"https://github.com/nemarDatasets/on007720\",\n \"identifier_type\": \"URL\",\n \"relation_type\": \"IsDescribedBy\"\n },\n {\n \"identifier\": \"10.18112/openneuro.ds007720.v1.0.1\",\n \"identifier_type\": \"DOI\",\n \"relation_type\": \"IsDerivedFrom\"\n },\n {\n \"identifier\": \"https://nemar.org/dataset/on007720\",\n \"identifier_type\": \"URL\",\n \"relation_type\": \"IsDescribedBy\"\n }\n ],\n \"funding_references\": [\n {\n \"funder_name\": \"National Institutes of Health\",\n \"award_number\": \"R01DC009834\",\n \"award_title\": \"National Institute on Deafness and Other Communication Disorders\"\n }\n ],\n \"resource_type_general\": \"Dataset\",\n \"resource_type_specific\": \"EEG Dataset\",\n \"modalities\": [\n \"eeg\"\n ],\n \"sizes\": [\n \"2.3 GB (1114 files)\"\n ],\n \"formats\": [\n \".eeg\",\n \".json\",\n \".md\",\n \".tsv\",\n \".vhdr\",\n \".vmrk\",\n \".yml\"\n ],\n \"source_hash\": \"fc9fa6e15836ba0f5c4cdfae55022b0d4549664e4b84496127c59e76b5e33dec\"\n}","last_activity_at":"2026-06-19 04:01:42","source":"openneuro","source_id":"ds007720","subject_count":5,"modalities":"eeg","age_min":null,"age_max":null,"file_size":2327377271,"total_files":2727,"tasks":"MatrixCalibration,MatrixCopyPhrase,RSVPCalibration,RSVPCopyPhrase","metadata_columns_error":null,"staleness_warn_stage":null,"staleness_admin_notified_at":null,"authors":"Betts Peters, Daniel Klee, Michelle Kinsella, Yonah Hendin, Tab Memmott, Matthew Lawhead, Srikar Ananthoju, Basak Celik, Scott Spaulding, Barry Oken, Melanie Fried-Oken","license":"CC0","readme":"[![DOI](https://img.shields.io/badge/DOI-10.82901%2Fnemar.on007720-blue)](https://doi.org/10.82901/nemar.on007720)\n\nBCI-FIT: A customization protocol for communication brain-computer interface systems.\n============================================================\n\nDataset Overview\n-----------------\n\nThe BCI Functional Implementation Toolkit (BCI-FIT) is a flexible, user-centered, interdisciplinary customization protocol for non-implantable communication brain-computer interface (cBCI) systems. Five participants with speech and/or physical impairments resulting from amyotrophic lateral sclerosis (ALS) completed calibration and copy-spelling tasks with both customized and non-customized versions of a non-implantable cBCI over multiple visits. Starting from a predetermined default, system customization was iteratively adjusted after each visit based on typing performance, participant preferences, clinical observations, brain signal characteristics, and results of simulations and other offline exploration. \n\n- N=5\n- Participants had speech and/or physical impairments\n- All tasks were proctored using BciPy [1]\n- The dataset is organized in accordance with the Brain Imaging Data Structure (BIDS) specification (version 1.7.0).\n\nMethodology\n--------------------------------\n\nTesting was conducted on an MSI Trident 3 desktop computer running Windows 11 Professional (64-bit) with an Nvidia GeForce RTX 3060 GPU, 12th-gen Intel i7-12700 CPU (2.10 GHz), and 16 GB RAM. Participants viewed tasks on a 24” monitor with 1920x1080 resolution cycling at 144 Hz. Participants were seated (in a recliner chair or wheelchair, according to their preference) at 50-60 cm from the computer screen, which was mounted on an adjustable table or a floor mount (Rehadapt, Kassel, Germany). Screen distance was approximated using calibration positioning with a Tobii Nano eye-tracking system (Tobii, Stockholm, Sweden). For switch input, participants activated a Jelly Bean switch (AbleNet, Roseville, MN, USA) that they either held in their hand or rested on their lap. EEG activity was recorded from one of two different dry electrode caps (Wearable Sensing, San Diego, CA, USA): DSI-Flex (FCz, Pz, Oz, P3, P4, PO7, and PO8) or DSI-24 (Fp1/2, Fz, F3/4, F7/8, Cz, C3/4, T3/T4 [T7/T8], T5/T6 [P7/P8], Pz, P3/P4, O1/2), accessed in BciPy via Lab Streaming Layer. The cap was chosen based on user preference and a review of signal quality related to cap fit and comfort. Both caps employ a hardware filter permitting a collection bandwidth of 0.003-150 Hz. Recordings were collected with an averaged linked-ear reference ((A1 + A2) / 2) and an active common-mode follower; the ground electrode was located at FCz on the DSI-Flex and at FPz on the DSI-24. All recordings were sampled at 300 Hz. All data were downsampled to 150 Hz and filtered between 1 and 20 Hz (2nd-order) with a 60 Hz notch. \nParticipants used Matrix and RSVP typing layouts included in BciPy 2.0 [1]. For each calibration task, participants completed 55 test inquiries, each consisting of a target character prompt, a presentation cue, and a sequence of 14 characters. The prompt target characters were displayed in yellow text within the matrix, or in the center of the screen for RSVP, for 1 second. For the presentation cue, the entire matrix was flashed in red, or a red fixation cross was displayed in the center of the screen for RSVP, for 0.5 seconds to alert participants that the flashing letter sequence was about to begin. The 14 characters in the letter stream were then intensified one at a time in the matrix, or flashed in the center of the screen for RSVP, at 5 Hz. Participants were instructed to watch for the intensification or presentation of the target character (in 10% of inquiries, the target was not intensified). A blank interval of 4 seconds separated each inquiry. Presentation timing, color, and number of characters listed above are default values; these parameters were adjusted as part of system customization.\n\nCopy-spelling was similar to calibration, with the key difference that there were no target prompts in test inquiries. Rather, participants were presented with a partially completed phrase at the top of the screen and asked to copy the final word. All target words were five letters long, and phrases were drawn randomly, without replacement, from a preselected list. Participants were asked to watch for the letter they wanted to type and keep noticing that character until the system made a typing decision. If the decision was correct, they were to begin watching the next letter in the word. If it was incorrect, they were to look for the “<” character to select a backspace and delete the error, then try again to type the next letter in the target word. Each character selection was made after a series of inquiries. In the first inquiry of a series, characters were presented in random order with uniform likelihood probability. In subsequent inquiries, character probabilities were updated based on EEG evidence from classifier-adjusted likelihood estimates. The system showed a minimum of two and a maximum of eight inquiries per series before making a selection, which occurred either when the classifier surpassed a confidence threshold of 0.8 probability or if the maximum inquiry count was reached (in which case the character with the highest probability was selected). Participants attempted to copy four five-letter words per condition per visit. A copy-spelling session would end when the target word was successfully completed, when 10 selections were made, when five erroneous selections occurred in a row, or after 10 minutes had passed.\n\nDirectory Structure\n------------------------------------------\n\nThe dataset follows the BIDS convention with the following structure: `sub-