An empirical comparison of deep learning explainability approaches for EEG using simulated ground truth

Q: What is the clinical evidence from this study?

Population: Simulated EEG data. Intervention: Deep learning explainability approaches (DeepLift, LRP, etc.) vs. Other explainability methods (e.g., Saliency, GradCAM++). Primary outcome: Robustness and sensitivity to detect temporal, spectral, and spatial properties.

Akshay Sujatha Ravindran; José L. Contreras-Vidal

doi:10.1038/s41598-023-43871-8

October 18, 2023Scientific ReportsOpen Access

An empirical comparison of deep learning explainability approaches for EEG using simulated ground truth

View Full Paper

Ask AI

Bookmark

Share

Key result

DeepLift and LRP outperform Saliency and GradCAM++ in reliably detecting simulated EEG features.

Figures

Why the study?

Deep learning neural decoders have improved EEG decoding, but model interpretability remains under-explored, requiring evaluation of explanation methods and when they fail.

Structured PICO

P

Population

Simulated EEG data framework designed to evaluate the robustness and sensitivity of twelve back-propagation-based visualization methods for deep learning models.

I

Intervention

Twelve back-propagation-based visualization methods for deep learning models (including DeepLift, LRP, GradCAM, Saliency, etc.)

C

Comparator

Ground truth features in simulated EEG

O

Outcome

Robustness and sensitivity to detect temporal, spatial, and spectral precision

Limitations

Synthetic EEG is only an approximation to measured EEG and may miss some key physiological and non-physiological signals or artifacts.
The scope is limited to visualization methods that highlight key segments of the input data, making it difficult to ascertain which specific feature caused the correct prediction.
The study is limited to a generic CNN architecture and does not explore variants like attention modules, Transformers, or Recurrent Neural Networks.
GradCAM performance decreases rapidly with lower SNR
GradCAM cannot produce channel relevancy if initial layers spatially mix information

Implication

DeepLift and Layerwise Relevance Propagation are recommended over Saliency for explaining deep learning models in EEG due to superior robustness and sensitivity.

Abstract

Recent advancements in machine learning and deep learning (DL) based neural decoders have significantly improved decoding capabilities using scalp electroencephalography (EEG). However, the interpretability of DL models remains an under-explored area. In this study, we compared multiple model explanation methods to identify the most suitable method for EEG and understand when some of these approaches might fail. A simulation framework was developed to evaluate the robustness and sensitivity of twelve back-propagation-based visualization methods by comparing to ground truth features. Multiple methods tested here showed reliability issues after randomizing either model weights or labels: e.g., the saliency approach, which is the most used visualization technique in EEG, was not class or model-specific. We found that DeepLift was consistently accurate as well as robust to detect the three key attributes tested here (temporal, spatial, and spectral precision). Overall, this study provides a review of model explanation methods for DL-based neural decoders and recommendations to understand when some of these methods fail and what they can capture in EEG.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Ravindran et al. (2023) studied Simulated EEG data. Deep learning explainability approaches (DeepLift, LRP, etc.) vs. Other explainability methods (e.g., Saliency, GradCAM++) was evaluated on Robustness and sensitivity to detect temporal, spectral, and spatial properties. DeepLift and Layerwise Relevance Propagation (LRP) were consistently accurate and robust in detecting temporal, spatial, and spectral features in simulated EEG data, whereas Saliency and GradCAM++ showed reliability issues.

synapsesocial.com/papers/6a20f6b5b9c39a5cf6ecbe44 https://doi.org/10.1038/s41598-023-43871-8

View Full Paper

Ask AI

Bookmark

Share

An empirical comparison of deep learning explainability approaches for EEG using simulated ground truth

Figures

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

An empirical comparison of deep learning explainability approaches for EEG using simulated ground truth

Figures

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading