Objective.Functional near-infrared spectroscopy (fNIRS) is an emerging technique for brain-computer interfaces (BCIs) due to its advantages in spatial resolution, robustness to artifacts, portability and usability for long-term monitoring, etc. Existing BCI methods take a holistic approach to all signal-collecting channels and corresponding brain regions, while the task-related brain regions and their interactions have not been well explored.Approach.This paper proposes a principal brain-region analysis (PBA) framework to incorporate the functional contribution as well as collaboration of task-specific brain regions (TSBRs) to boost BCI performance. Firstly, the identification of TSBRs is formulated as an optimization problem by maximizing classification accuracy under spatial constraints on brain regions of interest. Then, an evolutionary decomposition algorithm is constructed by combining spatial nondominated operators and genetic iterative computation, identifying TSBRs from the whole brain regions. Afterwards, classifiers are trained by neuroimaging features in the decomposed TSBRs in combination with stacking to generate the final predictions.Main results.The proposed PBA method was evaluated on two public datasets for fNIRS-based BCIs, significantly enhancing the classification accuracy for the sliding slope-based method by 8.91% and 6.03% and the sliding mean concentration change method by 13.62% and 6.15%, respectively.Significance.PBA establishes a pivotal framework to fundamentally advance the accuracy and explainability of BCIs.
Lu et al. (2026) studied this question.