Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Nested Deep Learning Model Towards A Foundation Model for Brain Signal Data

Fangyi Wei, Jiajie Mo, Kai Zhang, Haipeng Shen, Srikantan Nagarajan, Fei Jiang

TL;DR

The paper tackles the challenge of detecting epileptic spikes in EEG/MEG data while accommodating varying channel configurations and localizing spike-generating channels. It introduces Nested Deep Learning (NDL), a framework that learns channel-wise weights and aggregates multi-channel signals through a weighted representation, enabling montage-invariant spike detection and cross-modality transfer. The authors provide a probabilistic model, neural-network-based estimators for the aggregation function g and the channel weights α, and prove identifiability and consistency under standard assumptions. Empirical validation on TUH EEG, UCSF MEG, and BTH EEG datasets demonstrates strong spike-detection performance, accurate channel localization, and cross-modality generalization, with a clear pathway toward using NDL as a foundation model for brain-signal analysis.

Abstract

Epilepsy affects around 50 million people globally. Electroencephalography (EEG) or Magnetoencephalography (MEG) based spike detection plays a crucial role in diagnosis and treatment. Manual spike identification is time-consuming and requires specialized training that further limits the number of qualified professionals. To ease the difficulty, various algorithmic approaches have been developed. However, the existing methods face challenges in handling varying channel configurations and in identifying the specific channels where the spikes originate. A novel Nested Deep Learning (NDL) framework is proposed to overcome these limitations. NDL applies a weighted combination of signals across all channels, ensuring adaptability to different channel setups, and allows clinicians to identify key channels more accurately. Through theoretical analysis and empirical validation on real EEG/MEG datasets, NDL is shown to improve prediction accuracy, achieve channel localization, support cross-modality data integration, and adapt to various neurophysiological applications.

Nested Deep Learning Model Towards A Foundation Model for Brain Signal Data

TL;DR

The paper tackles the challenge of detecting epileptic spikes in EEG/MEG data while accommodating varying channel configurations and localizing spike-generating channels. It introduces Nested Deep Learning (NDL), a framework that learns channel-wise weights and aggregates multi-channel signals through a weighted representation, enabling montage-invariant spike detection and cross-modality transfer. The authors provide a probabilistic model, neural-network-based estimators for the aggregation function g and the channel weights α, and prove identifiability and consistency under standard assumptions. Empirical validation on TUH EEG, UCSF MEG, and BTH EEG datasets demonstrates strong spike-detection performance, accurate channel localization, and cross-modality generalization, with a clear pathway toward using NDL as a foundation model for brain-signal analysis.

Abstract

Epilepsy affects around 50 million people globally. Electroencephalography (EEG) or Magnetoencephalography (MEG) based spike detection plays a crucial role in diagnosis and treatment. Manual spike identification is time-consuming and requires specialized training that further limits the number of qualified professionals. To ease the difficulty, various algorithmic approaches have been developed. However, the existing methods face challenges in handling varying channel configurations and in identifying the specific channels where the spikes originate. A novel Nested Deep Learning (NDL) framework is proposed to overcome these limitations. NDL applies a weighted combination of signals across all channels, ensuring adaptability to different channel setups, and allows clinicians to identify key channels more accurately. Through theoretical analysis and empirical validation on real EEG/MEG datasets, NDL is shown to improve prediction accuracy, achieve channel localization, support cross-modality data integration, and adapt to various neurophysiological applications.
Paper Structure (21 sections, 1 theorem, 61 equations, 11 figures, 4 tables, 1 algorithm)

This paper contains 21 sections, 1 theorem, 61 equations, 11 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Model
  3. Deep Learning Estimation
  4. Real Data Analysis
  5. Data and preprocessing
  6. Results from the TUH Data
  7. Results from the UCSF MEG Data
  8. Results from the BTH EEG Data
  9. Procedures to adopt to continuous time series data
  10. Simluation
  11. Theoretical Results
  12. Necessary definitions and conditions
  13. Theoretical properties
  14. Conclusion and Discussion
  15. Additional Notations
  16. ...and 6 more sections

Key Result

Proposition 1

Assume $g(\cdot)$ is continuously differentiable and $\sum_{l=1}^{d} \boldsymbol\alpha({\bf X}_{il}, {\bf X}_i) = \bf 1$. Then $\boldsymbol\alpha({\bf X}_{il}, {\bf X}_i)$ and $g(\cdot)$ are identifiable.

Figures (11)

  • Figure 1: Illustration of $\{Y_i=0, {\bf X}_i, {\bf Z}_i\}$ (left) and $\{Y_j=1, {\bf X}_j, {\bf Z}_j\}$ (right). Spike signals are annotated in red and bold while background signals are shown in black. In the TUH data of Section \ref{['subsec:tuh']}, we set $d=22$, $T=p=250$.
  • Figure 2: The deep neural network diagram to approximate the true functions $g^*, \boldsymbol\alpha^*$.
  • Figure 3: TUH: Two-second segments of the six classes of events with 22 ACNS TCP montages, where the annotated channels are shown in red.
  • Figure 4: TUH: Results from NDL with fine-tuned hyperparameters.
  • Figure 5: TUH: Side-by-side boxplots comparing the top channels selected by NDL ranking with random ranking through 100 repetitions.
  • ...and 6 more figures

Theorems & Definitions (3)

  • Proposition 1
  • Definition 1
  • Definition 2