Rest-to-task transfer of sparse-to-full EEG reconstruction for cognitive-load decoding

OBJECTIVE: Sparse-to-full electroencephalography (EEG) reconstruction holds great promise for reducing electrode burden while maintaining compatibility with multichannel analysis pipelines. This study examined whether a convolutional neural network trained exclusively on resting-state EEG could preserve task-relevant information when transferred to previously unseen N-back working memory tasks without task-specific fine-tuning. Approach. A 4-to-21-channel reconstruction model was evaluated using four different sparse electrode configurations (CROSS, DSI, EMOT, MUSE) in participants with major depressive disorder, who served as a clinically heterogeneous test cohort rather than a disorder-specific classification target. Reconstruction was comprehensively assessed based on complementary signal-level metrics, the preservation of both overall power spectral density and workload-related spectral contrasts, and within-subject cognitive-load classification across task-repetition groups. Main results. The preferred electrode configuration depended on the evaluation endpoint. In generic waveform reconstruction, EMOT had the best composite rank for generic reconstruction fidelity and a tie-adjusted winner frequency of 61.4%, although it did not lead all individual metrics. In contrast, CROSS showed the strongest preservation of overall spectral structure and 2-back versus 0-back spectral contrast and had the highest binary classification performance among the reconstructed configurations. However, the relationship between generic reconstruction fidelity and spectral preservation reversed from positive in pooled analysis to negative within participants, and spectral preservation was not significantly associated with binary classification accuracy at the participant level. Significance. These results demonstrate that resting-state-trained sparse-to-full EEG reconstruction can partially preserve task-related spectral representation without task-specific fine-tuning. However, our findings reveal a critical caveat: high-fidelity waveform-level reconstruction does not necessarily correspond to superior preservation of cognitive-task features. Ultimately, these findings show that reconstruction quality is endpoint dependent, underscoring the necessity of a multidimensional validation combining generic signal fidelity, task-specific spectral preservation, and downstream model behavior. TRIAL REGISTRATION: KCT0012206.

Authors

Institutions

Publication Details

Journal
Journal of Neural Engineering
Published
2026-11-18
DOI
https://doi.org/10.1088/1741-2552/aec3d8
Primary Topic
EEG and Brain-Computer Interfaces
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Rest-to-task transfer of sparse-to-full EEG reconstruction for cognitive-load decoding

Keum‐Shik Hong, So-Hyeon Yoo, Seung-Jae Kim, Hyang‐Sook Hoe et al.
2 links
Journal of Neural Engineering
EEG and Brain-Computer Interfaces
article

Rest-to-task transfer of sparse-to-full EEG reconstruction for cognitive-load decoding

Keum‐Shik Hong, So-Hyeon Yoo, Seung-Jae Kim, Hyang‐Sook Hoe, Ji Hoon Jang, Jeong-Heon Song, Jin-Yeop Park, Jun-Su Kim
article en

Abstract

OBJECTIVE: Sparse-to-full electroencephalography (EEG) reconstruction holds great promise for reducing electrode burden while maintaining compatibility with multichannel analysis pipelines. This study examined whether a convolutional neural network trained exclusively on resting-state EEG could preserve task-relevant information when transferred to previously unseen N-back working memory tasks without task-specific fine-tuning. Approach. A 4-to-21-channel reconstruction model was evaluated using four different sparse electrode configurations (CROSS, DSI, EMOT, MUSE) in participants with major depressive disorder, who served as a clinically heterogeneous test cohort rather than a disorder-specific classification target. Reconstruction was comprehensively assessed based on complementary signal-level metrics, the preservation of both overall power spectral density and workload-related spectral contrasts, and within-subject cognitive-load classification across task-repetition groups. Main results. The preferred electrode configuration depended on the evaluation endpoint. In generic waveform reconstruction, EMOT had the best composite rank for generic reconstruction fidelity and a tie-adjusted winner frequency of 61.4%, although it did not lead all individual metrics. In contrast, CROSS showed the strongest preservation of overall spectral structure and 2-back versus 0-back spectral contrast and had the highest binary classification performance among the reconstructed configurations. However, the relationship between generic reconstruction fidelity and spectral preservation reversed from positive in pooled analysis to negative within participants, and spectral preservation was not significantly associated with binary classification accuracy at the participant level. Significance. These results demonstrate that resting-state-trained sparse-to-full EEG reconstruction can partially preserve task-related spectral representation without task-specific fine-tuning. However, our findings reveal a critical caveat: high-fidelity waveform-level reconstruction does not necessarily correspond to superior preservation of cognitive-task features. Ultimately, these findings show that reconstruction quality is endpoint dependent, underscoring the necessity of a multidimensional validation combining generic signal fidelity, task-specific spectral preservation, and downstream model behavior. TRIAL REGISTRATION: KCT0012206.

Journal of Neural Engineering
Korea Brain Research Institute (KR), Pusan National University (KR)
Ministry of Science, ICT and Future Planning, National Research Foundation of Korea
Good health and well-being
EEG and Brain-Computer Interfaces
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon