Content area

Abstract

Functional near-infrared spectroscopy (fNIRS) is a promising modality for brain-computer interfaces (BCIs), offering a convenient and non-intrusive way to measure real-time brain activity. Despite its potential, the widespread adoption of fNIRS-based BCIs faces challenges, including the difficulty of collecting and labeling sufficient data, the difficulty of interpreting the data due to noise and cross-subject variation, and the need for extensive user-specific calibration. To address these issues, we present a series of contributions aimed at advancing the use of fNIRS in interactive BCIs. First, we introduce new datasets, including the largest open-access dataset of its kind, one containing multivariate fNIRS recordings from 87 participants performing standard visual n-back tasks, and the other containing such recordings from 53 participants performing standard audio n-back tasks to induce cognitive workload levels. These datasets enable benchmarking, standardizing evaluation protocols, and promoting generalization across unseen users. Second, we propose a novel multi-stage supervised machine learning pipeline that combines group-level data with subject-specific fine-tuning, demonstrating considerable accuracy gains compared to baselines that use only subject-specific data or only group-level data, even when our approach is given much less subject-specific data. Third, we introduce NIRS-X, an adaptive learning framework leveraging unlabeled fNIRS data. NIRS-X includes NIRSiam, which extracts generalized brain activity representations, and NIRSformer, a neural network designed to capture both spatial and temporal relationships in multi-channel fNIRS signals. Although our machine learning experiments were conducted offline, all of the techniques we developed - including the supervised pipeline and the NIRS-X framework - are designed to be compatible with real-time adaptive BCI systems, enabling practical deployment in interactive scenarios. Using both supervised and unsupervised approaches, our methods demonstrate robust individual classification of cognitive workload levels and effective calibration on new users and tasks. These contributions highlight the potential of fNIRS for real-world BCI applications, reducing barriers to entry and paving the way for widespread exploration of fNIRS-based interactive systems.

Details

1010268
Business indexing term
Title
Towards Real-World Applicability of BCIs: Improving Calibration, Generalization, and Adaptive Learning With fNIRS
Number of pages
149
Publication year
2025
Degree date
2025
School code
0234
Source
DAI-B 86/11(E), Dissertation Abstracts International
ISBN
9798315741558
Committee member
Chang, Remco; Fantini, Sergio; Dogar, Fahad; Brizan, David Guy
University/institution
Tufts University
Department
Computer Science
University location
United States -- Massachusetts
Degree
Ph.D.
Source type
Dissertation or Thesis
Language
English
Document type
Dissertation/Thesis
Dissertation/thesis number
31994535
ProQuest document ID
3207188363
Document URL
https://www.proquest.com/dissertations-theses/towards-real-world-applicability-bcis-improving/docview/3207188363/se-2?accountid=208611
Copyright
Database copyright ProQuest LLC; ProQuest does not claim copyright in the individual underlying works.
Database
ProQuest One Academic