Skip to content

Z-bros/HackhathonBR41N.IO26

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Silence Is Not Absence

Replication and critical reassessment of a vibrotactile P300 BCI pipeline for patients with Unresponsive Wakefulness Syndrome.

Built for BR41N.IO 2026 by Group 45.

What this is

This project replicates the published vibrotactile P300 BCI pipeline (Spataro et al. 2018; Guger et al. 2018) on a dataset of 8 EEG recordings from 2 UWS patients, then critically reassesses whether the dataset's pre-assigned high/low accuracy labels survive faithful reanalysis.

The data comes pre-labeled by the original researchers: each recording is tagged as either "high" (>95% accuracy in the original mindBEAGLE classifier) or "low" (<5%). We tested whether these labels would hold up when the published pipeline is re-implemented from scratch.

Headline result

Only 3 of 8 files cleared the 95% significance threshold (23%) in our re-implementation, scattered across both labels and patients. One file labeled as a >95% accuracy run came back at 8.2% — below the 12.5% chance level — in our hands.

No spectral biomarker (pre-stimulus alpha, event-related power) and no alternative classifier (shrinkage LDA, xDAWN spatial filtering, Riemannian MDM) substantially exceeded the LDA baseline. The bottleneck is signal strength, not algorithm choice.

Pipeline

The analysis is split across five MATLAB scripts, run in order:

Script Purpose Key output
uws_preprocessing.m Filter (0.1–30 Hz + 50 Hz notch), epoch (−100 to +600 ms), baseline-correct, reject ±100 µV. Extract three feature sets. processed_data.mat
uws_classification.m Single-trial LDA + group-of-8 LDA classification with leave-one-group-out CV. Permutation testing. classification_results.mat
uws_spectral.m Welch PSD, pre-stimulus band power per trial, event-related spectrograms. spectral_results.mat
uws_classifier_inspection.m LDA weights, Haufe-corrected activation patterns, scalp topographies. classifier_inspection.pdf
uws_classifier_comparison.m Comparison of four classifier variants on the same task. classifier_comparison.pdf

Reproducing the analysis

Requirements

  • MATLAB R2020a or newer (for exportgraphics)
  • Signal Processing Toolbox
  • Statistics and Machine Learning Toolbox

Data

The 8 .mat files (P1_low1.mat, P1_low2.mat, ..., P2_high2.mat) are provided by the BR41N.IO 2025 organizers and are not included in this repository. To request access, contact BR41N.IO.

Each file contains:

  • y: EEG data, samples × 8 channels (Fz, C3, Cz, C4, CP1, CPz, CP2, Pz)
  • trig: trigger channel (−1 = distractor, 1 = non-target, 2 = target)
  • fs: sampling rate (256 Hz)

Place the .mat files in the project root, then run the scripts in scripts/ in the order listed above.

Results

All output figures are in results/. Highlights:

  • group_classification_perfile.pdf — per-file group-of-8 accuracy with chance and significance thresholds (the headline figure)
  • classifier_comparison.pdf — four classifier variants compared on the same task
  • spectral_summary.pdf — PSD, pre-stimulus alpha, and event-related spectrograms

Limitations

A few important caveats on what this project does and doesn't show:

  • Two patients only. Findings on signal-level fragility cannot be generalized to UWS as a population.
  • Replication does not equal critique. The original Spataro and Guger papers report group-level findings (e.g., BCI accuracy correlates with 6-month CRS-R recovery, r = 0.89) that are not addressed here. Our work focuses on a narrower question: whether single-session run-level labels survive faithful reanalysis on this specific dataset.
  • Outcome data was not available. A more complete reanalysis would correlate replication failure with clinical outcome.
  • Per-file analysis only. We did not pool trials across files, since pooling introduces session-level confounds that can inflate accuracy artifactually.

References

  • Giacino JT, Fins JJ, Laureys S, Schiff ND. (2014). Disorders of consciousness after acquired brain injury: the state of the science. Nat. Rev. Neurol. 10:99–114.
  • Spataro R, Heilinger A, Allison B, et al. (2018). Preserved somatosensory discrimination predicts consciousness recovery in unresponsive wakefulness syndrome. Clin. Neurophysiol. 129:1130–1136.
  • Guger C, Spataro R, Pellas F, et al. (2018). Assessing command-following and communication with vibro-tactile P300 brain-computer interface tools in patients with unresponsive wakefulness syndrome. Front. Neurosci. 12:423.
  • Wannez S, Heine L, Thonnard M, et al. (2017). The repetition of behavioral assessments in diagnosis of disorders of consciousness. Ann. Neurol. 81:883–889.

Team

  • Zidane Fatuna — GitHub | LinkedIn
  • Timoleon Fourfaro
  • Johana Dominguez
  • Neha Sharma

License

Code: MIT License (see LICENSE). Original publications and dataset: subject to their respective licenses.

About

EEG-based consciousness assessment in UWS patients — replication and reassessment of a vibrotactile P300 BCI pipeline. BR41N.IO 2026

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages