Bridging scalp and intracranial EEG in BCI via pretrained neural representations and geometric constraint embedding
Yihang Dong, Changhong Jing, Shuqiang Wang · 2026
Read Bridging scalp and intracranial EEG in BCI via pretrained neural representations and geometric constraint embedding by Yihang Dong, Changhong Jing, Shuqiang Wang (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-2604-14202.txt.
Introduction
Dong, Jing, and Wang lift scalp EEG toward intracranial quality for brain-computer interfaces, where surface signals lose detail and implanted recordings stay rare. Guided by the idea that geometric structure dictates function, they map static cortical anatomy onto dynamic limits for neural signal spread. They fold in representations from a pretrained large EEG model. A multidimensional diffusion step then synthesizes enhanced EEG, recovering activity patterns lost in transit through the brain and offering a low-cost path to high-fidelity signals.
Catalog record
| CREATOR | Yihang Dong, Changhong Jing, Shuqiang Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74935 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.14202
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bridging scalp and intracranial EEG in BCI via pretrained neural representations and geometric constraint embedding Yihang Dong, Changhong Jing, Shuqiang Wang Abstract Electroencephalography (EEG) has become one of the key modalities underpinning brain–computer interfaces (BCIs) due to its high temporal resolution, rapid responsiveness, non-invasiveness, low cost, and portability. However, EEG signals are substantially inferior to intracranial EEG (iEEG) in signal-to-noise ratio and local spatial resolution, whereas iEEG suffers from extremely limited clinical accessibility owing to its invasive nature, hindering widespread application. To address this challenge, this study proposes a unified data- and prior knowledge–driven framework for EEG–iEEG representational enhancement. Guided by the principle that “geometric structure dictates function”, the framework maps static cortical anatomy onto dynamic constraints governing neural signal propagation and integrates general-purpose neural representations extracted by a pre-trained large EEG model to explicitly model signal transmission through the brain. Enhanced EEG signals are then synthesized via a multidimensional representation diffusion process. Numerous experimental results demonstrate that the generated enhanced EEG signals effectively recover the neural activity patterns lost during propagation through the brain. This finding indicates that the performance ceiling of BCIs is constrained not only by acquisition hardware but also by the depth to which the generative model resolves the mechanisms of neural signal propagation. Collectively, the proposed framework provides a viable pathway toward acquiring high-fidelity neural signals at low cost. In recent years, BCI technologies have demonstrated substantial application potential in neurorehabilitation, human–machine interaction, and defense domains, a potential that hinges critically on accurate decoding of neural activity chaudhary2016brain ; schwemmer2018meeting ; d2025towards ; lee2025brain ; merk2025invasive . Multiple neuroimaging modalities including functional magnetic resonance imaging (fMRI), functional ultrasound (fUS), and functional near-infrared spectroscopy (fNIRS) have been employed for BCI decoding rybavr2025simultaneous ; griggs2024decoding ; however, these modalities generally suffer from insufficient temporal resolution, slow response latency, high instrumentation costs, and stringent acquisition environments, rendering them ill-suited for real-time, dynamic neural decoding. In contrast, EEG has emerged as the preferred signal source for non-invasive BCI owing to its millisecond-scale temporal resolution, rapid responsiveness, non-invasiveness, low cost, and high portability ding2025eeg ; huang2026robust ; daly2023neural . Intracranial EEG (iEEG) encompassing electrocorticography (ECoG), multielectrode arrays, and stereoelectroencephalography (sEEG) is regarded as the cornerstone modality for invasive BCI due to its direct recording of cortical electrical activity and exceptional signal-to-noise ratio stolk2018integrated ; dipalo2018plasmonic . Nevertheless, EEG signals are inherently limited by volume conduction and skull attenuation, and are highly susceptible to artifacts from multiple sources. Conversely, iEEG requires craniotomy, resulting in extremely limited clinical accessibility parvizi2018human . These constraints severely impede the advancement of non-invasive and invasive BCI in terms of signal fidelity and widespread applicability, respectively. Consequently, enhancing non-invasive EEG signals to approximate the neural activity representational capacity of iEEG, thereby yielding enhanced EEG signals, constitutes a pivotal step toward realizing high-performance, non-invasive, low-cost, and portable BCI systems. However, the key of EEG signal enhancement does not lie in investigating the neural sources or their activation states, but rather in learning the neural activity representations lost during signal propagation. The former, which aims to localize the origin of neural electrical activity, generally falls under EEG source imaging and fundamentally addresses the question “where is the neural activity generated?” michel2004eeg ; kaiboriboon2012eeg . In contrast, this study aims to generate high-fidelity enhanced EEG signals to improve performance on downstream BCI tasks. Recently, Bahman Abdi-Sargezeh and colleagues attempted to map EEG directly to iEEG using deep neural networks in an end-to-end manner abdi2021higher ; abdi2023mapping ; abdi2025eeg . Although these studies preliminarily demonstrate the potential of data-driven approaches in modeling complex nonlinear mappings, their task formulation is inherently limited to signal-level fitting and cannot achieve generalizable EEG signal enhancement. On the other hand, Miao Cao and colleagues proposed reconstructing iEEG from magnetoencephalography (MEG) signals using an LCMV beamformer cao2022virtual . However, this approach remains constrained by linear assumptions and simplified brain models, yielding reconstructions insufficient for high-precision BCI applications. More critically, MEG systems are prohibitively expensive, require stringent magnetically shielded environments, and are impractical for bedside or long-term monitoring scenarios. In comparison, EEG offers superior portability, broad accessibility, and low cost, making it a more viable modality for real-world BCI deployment santhanam2006high . Unlocking this modality’s potential requires overcoming limitations in modeling physical constraints. The brain exhibits intrinsic structure-function coupling, with cortical geometry as the scaffold shaping neural activity and encoding biophysical boundary conditions governing signal generation, propagation, and integration wang2024human ; pun2024measuring . Geometric modal decomposition maps static anatomy into dynamic functional priors, revealing embedded spectral properties and establishing a principled pathway for structure-guided signal enhancement jiang2025gene ; zhang2025dynamic . Incorporating geometric priors compensates for information loss in non-invasive recordings, shifting reconstruction from data-driven toward a structure–function coupled paradigm and laying a theoretical foundation for next-generation high-fidelity EEG enhancement. To address these challenges, we propose a unified EEG–iEEG representation enhancement framework that integrates data-driven learning with neurodynamical priors, as shown in Figure 1 . The framework comprises four core components. First, a pretrained large EEG model is employed to extract general-purpose EEG representations, thereby strengthening the modeling capacity for neural activity patterns jiang2024large ; cui2024neuro . Second, explicit biophysical constraints are incorporated through joint modeling of neural mass models and neural field equations, ensuring that the enhanced signals adhere to the electrophysiological dynamics governing brain activity scheeringa2011neuronal ; kiebel2006dynamic . Third, to account for the complex geometric morphology of the brain, the proposed method processes T1-weighted MRI data via geometric mode decomposition to compute structurally informed priors across frequency bands, which model the attenuation of neural signals during propagation through skull and brain tissues and thereby enable recovery of lost neural representations during enhancement pang2023geometric ; yu2018geometric . Fourth, a multidimensional EEG feature encoder fuses the above three elements—representations, dynamical constraints, and structural priors—to generate features that precisely capture the coupling between long-range neural synchrony and local high-frequency activity; a diffusion model then guides the generation of enhanced EEG signals to improve their intrinsic fidelity to underlying neural activity patterns. Systematic evaluation across 14 diverse EEG datasets spanning multiple task types including emotion recognition shows that the proposed EEG–iEEG unified representational enhancement framework significantly improves the fidelity with which scalp EEG captures intracranial neural activity, without requiring invasive recordings or hardware upgrades. This work thus enables low-cost acquisition of high-fidelity enhanced EEG signals and provides a new technical pathway to overcome the performance bottlenecks of current non-invasive BCI. Figure 1: Schematic of the proposed framework. a, Data- and knowledge-based unified EEG representation enhancement framework. The model is trained with scalp EEG as input and iEEG as the target; during inference, enhanced EEG signals are generated from scalp EEG alone. b, Detailed architecture of the enhancement model, integrating a data-driven module, a neurodynamical–geometric constraint module, and a generative AI–based representation module. Individualized T1-weighted MRI can be incorporated alongside EEG to further improve performance. c, Multidimensional visualization of enhanced EEG signals, including spectral power distribution, spatial maps of band-specific activity, and temporal waveform morphology, collectively demonstrating recovery of neurophysiologically relevant features. d, Applications of the proposed framework, encompassing improved BCI decoding accuracy, enhanced BCI fairness , and support for brain functional mechanism research. Results Consistency evaluation between synthesized iEEG and ground-truth iEEG This study aims to enhance EEG signals using