A Domain-Informed Multi-Objective Framework for EEG Channel Selection in Motor Imagery BCIs
Dekka Muni Kumar, Dhruba Jyoti Kalita, Yogesh Kumar Meena · 2026
Read A Domain-Informed Multi-Objective Framework for EEG Channel Selection in Motor Imagery BCIs by Dekka Muni Kumar, Dhruba Jyoti Kalita, Yogesh Kumar Meena (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-29943.txt.
Introduction
Dekka Muni Kumar, Dhruba Jyoti Kalita, and Yogesh Kumar Meena pick EEG channel sets for motor-imagery links. Their multi-goal search balances channel place, via a Gaussian kernel, against task split, judged by task-linked desynchronization inside trials. Three methods — genetic, swarm, and decomposition searches — hunt together for compact sets near sensorimotor cortex. On four sets it scores high eighties on the best set and mid sixties on the hardest, beating single-goal and fixed-set rivals. Compact sets cut compute load, suiting wearable and live brain-computer use.
Catalog record
| CREATOR | Dekka Muni Kumar, Dhruba Jyoti Kalita, Yogesh Kumar Meena |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46885 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.29943
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Domain-Informed Multi-Objective Framework for EEG Channel Selection in Motor Imagery BCIs Dekka Muni Kumar, Dhruba Jyoti Kalita, Yogesh Kumar Meena Abstract Abstract— Motor imagery (MI) classification using electroencephalography (EEG) signals is essential for advancing brain-computer interfaces (BCIs). Traditional EEG channel selection methods often face limitations, such as dependency on single-objective criteria and susceptibility to local optima. To address these challenges, this work proposes a multi-objective optimisation framework that employs non-dominated sorting genetic algorithm, multiple-objective particle swarm optimisation, and a multi-objective evolutionary algorithm based on decomposition. Our approach effectively balances spatial relevance—using a Gaussian kernel—and functional discriminability, which assesses intratrial task-related desynchronisation, thereby improving performance. We evaluated this framework on four EEG datasets: Physionet, OpenBMI, HighGamma, and BCIIV-2A. The proposed approach successfully identifies compact, relevant channel subsets concentrated around sensorimotor cortex regions linked to MI activity, addressing the prevalent challenges of dimensionality and complexity inherent to traditional techniques. Furthermore, the framework achieved classification performance of 87%, 71%, 75%, and 65% on the Physionet, OpenBMI, HighGamma, and BCIIV-2A datasets, respectively. By outperforming existing single-objective and accuracy-based methods, and those relying on fixed subsets, these findings demonstrate that this new multi-objective optimisation framework can enhance MI-based BCI performance while facilitating compact channel configurations with reduced computational complexity, making them better suited for wearable, portable, and real-time BCI applications. I Introduction Motor imagery (MI) based brain-computer interfaces (BCIs) enable users to control external devices by decoding the electroencephalogram (EEG) signals generated around sensorimotor regions of the brain from the imagined motor movements [ 44 , 45 ] . These systems have a broad range of applications in assistive technologies for paralysed patients [ 40 , 39 ] , neuroprosthetic control of robotic limbs [ 8 , 5 , 7 ] , wheelchair navigation [ 26 , 38 ] , computer cursor control [ 35 , 13 ] , and virtual reality-based neurorehabilitation therapy for stroke recovery [ 41 , 36 ] . A critical factor for developing effective MI-based BCI systems is optimal EEG channel selection, which ensures the capture of the most relevant neural information while minimising noise and redundancy. Traditional channel selection methods in this context often adopt single-objective approaches or heuristic techniques, such as statistical significance-based selection, filter-based ranking, or domain-knowledge-emphasising sensorimotor regions (e.g., C3, C4) [ 32 , 24 , 27 ] . Neurophysiological approaches based on hemispheric activation patterns have been explored for channel selection, including lateralisation index (LI)-based methods for identifying contralateral motor cortex activity during MI tasks [ 22 ] and generalised lateralised readiness potential (LRP)-based channel optimisation frameworks using non-dominated sorting genetic algorithm (NSGA-II) to obtain compact channel subsets across multiple subjects [ 18 ] . Handiru et al. [ 14 ] proposed the iterative multi-objective optimisation for channel selection (IMOCS) framework, which iteratively refines channel subsets using spatial proximity and task discriminability measures. Although effective, IMOCS combines multiple objectives into a single aggregated score, limiting exploration of the true Pareto trade-off surface and increasing sensitivity to initialisation. To address these limitations, previous works have explored evolutionary multi-objective optimisation techniques. Kee et al. [ 16 ] and Esfahani et al. [ 25 ] employed NSGA-II-based frameworks for jointly optimising classification performance and channel reduction. Similarly, Wei et al. [ 42 ] and Moubayed et al. [ 1 ] investigated multiple-objective particle swarm optimisation (MOPSO) and multi-objective evolutionary algorithm based on Decomposition (MOEA/D) strategies for MI channel selection, demonstrating the advantages of Pareto-based optimisation over traditional sequential search methods. More recent studies incorporated multitasking and domain knowledge into the optimisation process. Liu et al. [ 20 ] proposed an evolutionary multitasking-based multi-objective optimisation algorithm (EMMOA), while Liu et al. [ 21 ] introduced a domain-knowledge-assisted multi-objective evolutionary algorithm (DK-MOEA) integrating spatial electrode distance information into the search process. Traditional channel selection methods in this context often adopt single-objective approaches or heuristic techniques, such as statistical significance-based selection, filter-based ranking, or domain-knowledge-emphasising sensorimotor sites (e.g., C3, C4) [ 2 , 32 ] . While these methods can reduce the number of channels and sometimes improve interpretability, they typically optimise classification accuracy and channel reduction or merge spatial and functional objectives into a single criterion or score. This may bias the selection process, overlooking important trade-offs between noise reduction, anatomical relevance, and signal discriminability [ 21 , 4 ] . As a result, such approaches may include noisy or neurophysiologically irrelevant channels while excluding MI-relevant ones. To address the limitations of traditional channel selection methods, this work proposes a multi-objective optimisation framework. This framework recognises spatial and functional relevance as distinct yet complementary objectives. The goal is to select an optimal subset of channels that is both neurophysiologically meaningful and discriminative in nature [ 28 , 23 ] . Fig. 1 : Key components of the proposed domain-informed multi-objective optimisation framework for EEG channel selection. To solve the formulated MOO problem, the proposed framework employs algorithms such as NSGA-II [ 12 ] , MOPSO [ 10 ] , and MOEA/D [ 46 ] , which identify Pareto-optimal solutions (POS) by independently exploring multiple objectives. Unlike hybrid approaches such as IMOCS, which merge criteria into a single score and may obscure important trade-offs, MOO algorithms preserve the independence of objectives, thereby yielding diverse Pareto-optimal channel subsets [ 9 ] . These population-based global optimisation methods extensively explore the solution space, avoid premature convergence to local optima, and generate multiple alternative solutions [ 12 , 46 ] . This flexibility enables researchers and clinicians to adapt channel configurations for different analytical scenarios or practical BCI deployment needs. The major contributions of this work are: 1. Proposing a multi-objective optimisation framework for selecting EEG channels in MI classification, combining spatial relevance and functional discriminability with three advanced optimisation algorithms. 2. Introducing a Gaussian kernel spatial relevance that emphasises electrodes near C3/C4, along with an ITTRD-based functional discriminability to quantify task-related band power changes for EEG channel selection. 3. Implementing optimisation algorithms (i.e., NSGA-II, MOPSO, and MOEA/D)—to preserve independence among the defined objectives, effectively avoiding local optima when finding Pareto-optimal channel subsets. 4. Establishing benchmarks for four public EEG datasets identifies efficient channel subsets in sensorimotor cortex regions linked to motor imagery and enhances classification accuracy, signal quality, interpretability, and computational efficiency. II Proposed Multi-Objective Framework The proposed multi-objective framework (see Fig. 1 ) consists of four major stages: preprocessing, multi-objective channel selection, feature extraction and selection, and classification. Initially, raw EEG signals acquired from multiple electrodes are segmented into trials corresponding to MI tasks. The segmented signals are then preprocessed using a 5th-order Butterworth bandpass filter to suppress noise and preserve task-relevant frequency components associated with MI activity. Following preprocessing, channel selection is formulated as a constrained multi-objective optimisation (MOO) problem. In this stage, the kernel is initialised around the sensorimotor reference regions C3 and C4, enabling the assignment of spatial weights to the channels. Subsequently, objective functions corresponding to spatial relevance and functional discriminability are evaluated for the selected channel subsets. To identify Pareto-optimal channel subsets that are both neurophysiologically relevant and effective for MI classification, we employ multi-objective evolutionary algorithms, including NSGA-II, MOPSO, and MOEA/D. II-A Formulation of Multi-Objective Optimisation Problem In this work, the EEG channel selection problem is formulated as a constrained multi-objective optimisation problem presented in