Akima Connelly

Akima Connelly

Auto-generated profile, based on publicly available information

Akima Connelly

School of Environment and Society, Institute of Science Tokyo, Tokyo, Japan

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 August 2021 in Journal of Neural Engineering
Verified
Journal Article
Closed access
ELElnaz Lashgari
Elnaz Lashgari
JOJordan Ott
Jordan Ott
ACAkima Connelly
Akima Connelly
PBPierre Baldi
Pierre Baldi
UMUri Maoz
Uri Maoz
Objective. Motor-imagery (MI) classification base on electroencephalography (EEG) has been long studied in neuroscience and more recently widely used in healthcare applications such as mobile assistive robots and neurorehabilitation. In particular, EEG-based MI classification methods that rely on convolutional neural networks (CNNs) have achieved relatively high classification accuracy. However, naively training CNNs to classify raw EEG data from all channels, especially for high-density EEG, is computationally demanding and requires huge training sets. It often also introduces many irrelevant input features, making it difficult for the CNN to extract the informative ones. This problem is compounded by a dearth of training data, which is particularly acute for MI tasks, because these are cognitively demanding and thus fatigue inducing. Approach. To address these issues, we proposed an end-to-end CNN-based neural network with attentional mechanism together with different data augmentation (DA) techniques. We tested it on two benchmark MI datasets, brain–computer interface (BCI) competition IV 2a and 2b. In addition, we collected a new dataset, recorded using high-density EEG, and containing both MI and motor execution (ME) tasks, which we share with the community. Main results. Our proposed neural-network architecture outperformed all state-of-the-art methods that we found in the literature, with and without DA, reaching an average classification accuracy of 93.6% and 87.83% on BCI 2a and 2b, respectively. We also directly compare decoding of MI and ME tasks. Focusing on MI classification, we find optimal channel configurations and the best DA techniques as well as investigate combining data across participants and the role of transfer learning. Significance. Our proposed approach improves the classification accuracy for MI in the benchmark datasets. In addition, collecting our own dataset enables us to compare MI and ME and investigate various aspects of EEG decoding critical for neuroscience and BCI.

14 June 2023
Verified
Conference Paper
Closed access
ACAkima Connelly
Akima Connelly
PLPengcheng Li
Pengcheng Li
PRPhurin Rangpong
Phurin Rangpong
TWTheerawit Wilaiprasitporn
Theerawit Wilaiprasitporn
TYTohru Yagi
Tohru Yagi
Motor imagery (MI) classification based on electroencephalography (EEG) has been extensively studied and recently used more in brain-computer interfaces (BCI). This study uses left and right hand MI tasks for the BCI system. A common obstacle for MI-BCI is the inability of some participants to perform the BCI task, called BCI illiteracy. Various training protocols have been investigated to improve the performance of BCI but are designed with a balanced dataset. Similarly to how people show a bias towards a side (e.g. left or right) for motor execution tasks, it has been seen that participants also show a performance bias in MI tasks as well. To address this MI bias in participants, a novel neurofeedback protocol was designed to adjust the number of trials each condition has. Trials will be adjusted to increase the number of times participants have to perform their weak MI task. This study aims to investigate the overall effect that the trial-adjusted neurofeedback had on participant’s cognitive performance on the MI-BCI system. The effects were investigated through time-frequency and band power analysis. The time-frequency analysis showed improvement in key MI feature and band power analysis results had an improvement on the alpha and beta frequency bands. In the analysis results, trial-adjusted neurofeedback was seen to have an effect on participant’s cognitive performance on the MI-BCI task.

15 June 2026 in IEEE Transactions on Neural Systems and Rehabilitation Engineering
Verified
Journal Article
Open access
ZZZhuohao Zhang
Zhuohao Zhang
HHHaruto Hamada
Haruto Hamada
PRPhurin Rangpong
Phurin Rangpong
ACAkima Connelly
Akima Connelly
TYTohru Yagi
Tohru Yagi
Brain-computer interface (BCI) systems are commonly classified as reactive or active. Reactive BCIs rely on responses to external stimuli, simplifying decoding but limiting direct user control. In contrast, active BCIs enable intuitive control through spontaneous mental activity but impose greater signal-processing challenges. Speech BCIs, a subset of active systems, hold promise for individuals with severe neurological impairments such as locked-in syndrome (LIS), potentially enabling communication through imagined speech. However, their adoption is hindered by the demanding pre-training required to collect imagery data, which often induces fatigue, and the difficulty many users face in generating consistent, high-quality imagery. To address these limitations, we propose a hybrid BCI framework that uses passive listening tasks as a surrogate for auditory imagery tasks during model training. As a first step, this proof-of-concept feasibility study examined neural responses to five Japanese vowels (/a/, /i/, /u/, /e/, /o/) under listening and imagery conditions in a limited experimental setting with healthy male participants, using event-related potentials (ERPs), topographical mapping, and event-related spectral perturbations (ERSPs), followed by statistical analysis of shared and distinct neural patterns. In the primary three-class and binary classification analyses, classification accuracies for imagery signals obtained from models trained solely on listening data were comparable to, and in some settings higher than, those trained directly on imagery data, with both outperforming chance level. These findings indicate that passive listening may provide effective training data for active auditory imagery BCIs within the tested setting, potentially reducing cognitive demands without compromising performance. Further validation in larger and more diverse cohorts is required before clinical application.