Quantifying Explainable AI-introduced signal noise on ECG data with Spectral Entropy
Summary
The paper proposes using spectral entropy as a metric to quantify noise introduced by explainability techniques in ECG arrhythmia classification, helping to distinguish true model signal from XAI-generated artifacts.
View Cached Full Text
Cached at: 06/25/26, 05:08 AM
# Quantifying Explainable AI-introduced signal noise on ECG data with Spectral Entropy Source: [https://arxiv.org/abs/2606.24974](https://arxiv.org/abs/2606.24974) [View PDF](https://arxiv.org/pdf/2606.24974) > Abstract:Explainability techniques are used to assess the output of various deep learning models\. This is especially true in healthcare, where models need to be trusted and decisions justified\. Explainability \(XAI\) tools use heuristics which often add signal noise to the explanation "core"\. It is not always obvious what is signal from the model and what is noise from the XAI\. We propose the use of spectral entropy as a measure of noise in XAI output\. We demonstrate its usefulness in the context of classifying arrhythmias in an ECG dataset with different post hoc explainability techniques\. ## Submission history From: David Kelly \[[view email](https://arxiv.org/show-email/c62689d4/2606.24974)\] **\[v1\]**Tue, 23 Jun 2026 11:47:29 UTC \(365 KB\)
Similar Articles
Comparing Post-Hoc Explainable AI Methods for Interpreting Black-Box EEG Models in Depression Detection
This paper compares several post-hoc explainability methods applied to an InceptionTime model for EEG-based depression detection, finding partial convergence among methods while highlighting methodological variability and limitations.
Spectral Priors vs. Attention: Investigating the Utility of Attention Mechanisms in EEG-Based Diagnosis
This paper shows that traditional machine learning models using spectral features from EEG signals can match or outperform state-of-the-art attention-based deep learning models for diagnosing neurodegenerative diseases, suggesting fundamental limitations of attention mechanisms in this domain.
Spectral Energy Centroid: a Metric for Improving Performance and Analyzing Spectral Bias in Implicit Neural Representations
This paper introduces the Spectral Energy Centroid (SEC) metric to analyze and improve spectral bias in implicit neural representations, demonstrating its utility for hyperparameter selection, signal complexity measurement, and cross-architecture alignment.
Mechanistic Interpretability of EEG Foundation Models via Sparse Autoencoders
This paper applies TopK Sparse Autoencoders to three EEG foundation models (SleepFM, REVE, LaBraM) to extract interpretable feature dictionaries and introduces a framework for concept steering, revealing representational failures and clinical entanglements.
DeepArrhythmia: Segment-Contextualized ECG Arrhythmia Classification via Selective Evidence Acquisition
DeepArrhythmia is a multimodal framework for beat-level ECG arrhythmia classification that combines raw ECG signals and waveform images, using segment-level confidence to selectively acquire physiological evidence for improved accuracy.