Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Swap-Adversarial Framework for Improving Domain Generalization in Electroencephalography-Based Parkinson's Disease Prediction

Seongwon Jin, Hanseul Choi, Sunggu Yang, Sungho Park, Jibum Kim

TL;DR

This paper tackles domain generalization in brain-signal PD prediction by introducing MOCOP, a reproducible ECoG benchmark, and a Swap-Adversarial Framework (SAF) that combines robust preprocessing, Inter-Subject Balanced Channel Swap (ISBCS) augmentation, and domain-adversarial learning. ISBCS reduces inter-subject bias by counterfactually exchanging channels across subjects, while a Gradient Reversal Layer-based domain classifier encourages subject-invariant features, complemented by a mutual information penalty. Across cross-subject, cross-session, cross-environment, and cross-dataset evaluations, SAF outperforms baselines and demonstrates notable gains, including up to ~41% macro-accuracy improvement in ECoG cross-subject tests and strong cross-dataset performance on EEG benchmarks, indicating broad generalization across modalities. The work provides public data and code, enabling further research into domain-generalizable brain-signal analysis for neurological disorder prediction and beyond.

Abstract

Electroencephalography (ECoG) offers a promising alternative to conventional electrocorticography (EEG) for the early prediction of Parkinson's disease (PD), providing higher spatial resolution and a broader frequency range. However, reproducible comparisons has been limited by ethical constraints in human studies and the lack of open benchmark datasets. To address this gap, we introduce a new dataset, the first reproducible benchmark for PD prediction. It is constructed from long-term ECoG recordings of 6-hydroxydopamine (6-OHDA)-induced rat models and annotated with neural responses measured before and after electrical stimulation. In addition, we propose a Swap-Adversarial Framework (SAF) that mitigates high inter-subject variability and the high-dimensional low-sample-size (HDLSS) problem in ECoG data, while achieving robust domain generalization across ECoG and EEG-based Brain-Computer Interface (BCI) datasets. The framework integrates (1) robust preprocessing, (2) Inter-Subject Balanced Channel Swap (ISBCS) for cross-subject augmentation, and (3) domain-adversarial training to suppress subject-specific bias. ISBCS randomly swaps channels between subjects to reduce inter-subject variability, and domain-adversarial training jointly encourages the model to learn task-relevant shared features. We validated the effectiveness of the proposed method through extensive experiments under cross-subject, cross-session, and cross-dataset settings. Our method consistently outperformed all baselines across all settings, showing the most significant improvements in highly variable environments. Furthermore, the proposed method achieved superior cross-dataset performance between public EEG benchmarks, demonstrating strong generalization capability not only within ECoG but to EEG data. The new dataset and source code will be made publicly available upon publication.

A Swap-Adversarial Framework for Improving Domain Generalization in Electroencephalography-Based Parkinson's Disease Prediction

TL;DR

This paper tackles domain generalization in brain-signal PD prediction by introducing MOCOP, a reproducible ECoG benchmark, and a Swap-Adversarial Framework (SAF) that combines robust preprocessing, Inter-Subject Balanced Channel Swap (ISBCS) augmentation, and domain-adversarial learning. ISBCS reduces inter-subject bias by counterfactually exchanging channels across subjects, while a Gradient Reversal Layer-based domain classifier encourages subject-invariant features, complemented by a mutual information penalty. Across cross-subject, cross-session, cross-environment, and cross-dataset evaluations, SAF outperforms baselines and demonstrates notable gains, including up to ~41% macro-accuracy improvement in ECoG cross-subject tests and strong cross-dataset performance on EEG benchmarks, indicating broad generalization across modalities. The work provides public data and code, enabling further research into domain-generalizable brain-signal analysis for neurological disorder prediction and beyond.

Abstract

Electroencephalography (ECoG) offers a promising alternative to conventional electrocorticography (EEG) for the early prediction of Parkinson's disease (PD), providing higher spatial resolution and a broader frequency range. However, reproducible comparisons has been limited by ethical constraints in human studies and the lack of open benchmark datasets. To address this gap, we introduce a new dataset, the first reproducible benchmark for PD prediction. It is constructed from long-term ECoG recordings of 6-hydroxydopamine (6-OHDA)-induced rat models and annotated with neural responses measured before and after electrical stimulation. In addition, we propose a Swap-Adversarial Framework (SAF) that mitigates high inter-subject variability and the high-dimensional low-sample-size (HDLSS) problem in ECoG data, while achieving robust domain generalization across ECoG and EEG-based Brain-Computer Interface (BCI) datasets. The framework integrates (1) robust preprocessing, (2) Inter-Subject Balanced Channel Swap (ISBCS) for cross-subject augmentation, and (3) domain-adversarial training to suppress subject-specific bias. ISBCS randomly swaps channels between subjects to reduce inter-subject variability, and domain-adversarial training jointly encourages the model to learn task-relevant shared features. We validated the effectiveness of the proposed method through extensive experiments under cross-subject, cross-session, and cross-dataset settings. Our method consistently outperformed all baselines across all settings, showing the most significant improvements in highly variable environments. Furthermore, the proposed method achieved superior cross-dataset performance between public EEG benchmarks, demonstrating strong generalization capability not only within ECoG but to EEG data. The new dataset and source code will be made publicly available upon publication.
Paper Structure (34 sections, 5 equations, 17 figures, 4 tables)

This paper contains 34 sections, 5 equations, 17 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Early Prediction of Parkinson’s Disease
  4. ECoG-based Prediction
  5. General Approaches to Domain Generalization
  6. Domain Generalization in EEG Analysis
  7. Method
  8. Preliminary
  9. Problem formulation
  10. Backbone Network
  11. Pre-processing
  12. Domain generalization
  13. Counterfactual Data Augmentation
  14. Domain Adversarial Learning
  15. Datasets
  16. ...and 19 more sections

Figures (17)

  • Figure 1: The overall architecture of Swap-Adversarial Framework. (a) The overall pipeline consists of four main stages: ECoG recording, preprocessing, counterfactual data augmentation, and domain adversarial learning. (b) Detailed illustration of each stage: (1) ECoG recording from subjects, (2) preprocessing including band-pass filtering, notch filtering, and ASR, (3) counterfactual data augmentation through Inter-Subject Balanced Channel Swapping (ISBCS) between Subject A and Subject B, (4) domain adversarial learning using EEGNet with task and domain classifiers.
  • Figure 2: Visualization of the preprocessing pipeline. A representative 15-second segment demonstrating the signal changes during noise removal is presented. Noise removal is applied sequentially using a band-pass filter, a notch filter, and artifact subspace reconstruction (ASR).
  • Figure 3: Inter-Subject Balanced Channel Swap (ISBCS): A counterfactual data augmentation method randomly swaps channel data between subject A anb B with the same class. The t-SNE visualization shows that ISBCS induces overlapping distributions between subjects, reducing the inter-subject distributional separability. Here, the red and green circles represent different subjects, respectively.
  • Figure 4: Sensitivity analysis of the ISBCS channel-swap probability $p$. (Left) Wireless (train) → transfer to wired (cross-subject); (Right) Wired (train) → transfer to wireless (cross-subject).
  • Figure 5: t-SNE visualization of the ECoG data after power spectral density (PSD) transformation. (a), (b) Distributions of all subjects in the wired and wireless datasets, respectively. (c) Distributions of individual subjects in the wired dataset. (d) Distributions of individual subjects in the wireless dataset. PS denotes the Post-Stimulation state, while PD denotes Parkinson's Disease.
  • ...and 12 more figures