Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FAIR-ESI: Feature Adaptive Importance Refinement for Electrophysiological Source Imaging

Linyong Zou, Liang Zhang, Xiongfei Wang, Jia-Hong Gao, Yi Sun, Shurong Sheng, Kuntao Xiao, Wanli Yang, Pengfei Teng, Guoming Luan, Zhao Lv, Zikang Xu

TL;DR

FAIR-ESI tackles the challenging ill-posed ESI problem by adaptively refining feature importance across spectral, temporal, and patch-wise representations to improve source reconstruction from noninvasive scalp signals. It combines FFT-based spectral refinement, weighted temporal feature fusion, and self-attention-based patch-wise refinement, trained with synthetic paired data generated by neural mass models. Across two simulation datasets and two real-world clinical datasets, FAIR-ESI achieves superior localization precision, lower LE and nMSE, and robust extendability across head models and modalities. This approach promises more accurate noninvasive brain source localization and potential improvements in brain disorder diagnosis, including epilepsy.

Abstract

An essential technique for diagnosing brain disorders is electrophysiological source imaging (ESI). While model-based optimization and deep learning methods have achieved promising results in this field, the accurate selection and refinement of features remains a central challenge for precise ESI. This paper proposes FAIR-ESI, a novel framework that adaptively refines feature importance across different views, including FFT-based spectral feature refinement, weighted temporal feature refinement, and self-attention-based patch-wise feature refinement. Extensive experiments on two simulation datasets with diverse configurations and two real-world clinical datasets validate our framework's efficacy, highlighting its potential to advance brain disorder diagnosis and offer new insights into brain function.

FAIR-ESI: Feature Adaptive Importance Refinement for Electrophysiological Source Imaging

TL;DR

FAIR-ESI tackles the challenging ill-posed ESI problem by adaptively refining feature importance across spectral, temporal, and patch-wise representations to improve source reconstruction from noninvasive scalp signals. It combines FFT-based spectral refinement, weighted temporal feature fusion, and self-attention-based patch-wise refinement, trained with synthetic paired data generated by neural mass models. Across two simulation datasets and two real-world clinical datasets, FAIR-ESI achieves superior localization precision, lower LE and nMSE, and robust extendability across head models and modalities. This approach promises more accurate noninvasive brain source localization and potential improvements in brain disorder diagnosis, including epilepsy.

Abstract

An essential technique for diagnosing brain disorders is electrophysiological source imaging (ESI). While model-based optimization and deep learning methods have achieved promising results in this field, the accurate selection and refinement of features remains a central challenge for precise ESI. This paper proposes FAIR-ESI, a novel framework that adaptively refines feature importance across different views, including FFT-based spectral feature refinement, weighted temporal feature refinement, and self-attention-based patch-wise feature refinement. Extensive experiments on two simulation datasets with diverse configurations and two real-world clinical datasets validate our framework's efficacy, highlighting its potential to advance brain disorder diagnosis and offer new insights into brain function.
Paper Structure (19 sections, 3 equations, 5 figures, 4 tables)

This paper contains 19 sections, 3 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminaries
  4. Paired Data Simulation & Patch Data Generation
  5. FAIR-ESI: Feature Adaptive Importance Refinement
  6. FFT-based Spectral Feature Refinement
  7. Weighted Temporal Feature Refinement
  8. Self-Attention-based Patch-wise Feature Refinement
  9. Source Activity Reconstruction
  10. Loss Function
  11. Experiments and Results
  12. Datasets
  13. Simulation Datasets
  14. Real-world Clinical Datasets
  15. Experiment Configurations
  16. ...and 4 more sections

Figures (5)

  • Figure 1: The forward propagation and inverse reconstruction problem of ESI.
  • Figure 2: Overview of FAIR-ESI. (a) The pipeline of the proposed method, which contains three views of feature refinement. (b) Details about spectral-temporal feature refinement. Note that $P^{*}_{\text{RE}}$ and $P^{*}_{\text{IM}}$ are in log-scale for better visualization. (c) Details about patch-wise feature refinement. (d) Details about source activity reconstruction. $P^{O*}$ and $X$ are the output of the feature refinement module and the input scalp signal, respectively.
  • Figure 3: Visualization of SimMEG and SimEEG datasets, the simulated source is located at region 121. Upper: SimMEG with $[r, N_{s}, N_{n}] = [5, 1, 2]$; Lower: SimEEG with $[r, N_{s}, N_{n}] = [5, 1, 2]$. Colors represent the amplitude of the estimated source activation.
  • Figure 4: Visualization of CMR and Localize-MI datasets. Upper: CMR dataset; Lower: Localize-MI dataset. Colors represent the amplitude of the estimated source activation.
  • Figure 5: (a) Visualization of spectral-temporal feature refinement; (b)Visualization of key-patch selection, E dnotes the energy of the patch.