Pith. sign in

REVIEW 2 major objections 1 minor 24 references

Quantifying Explainable AI-introduced signal noise on ECG data with Spectral Entropy

T0 review · 2 major / 1 minor · reviewed 2026-06-26 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Spectral entropy quantifies noise added by explainability techniques to ECG arrhythmia classifications.

desk verdict The paper proposes spectral entropy to measure XAI-introduced noise on ECG explanations but supplies no methods, results, or validation that it actually isolates that noise. read the letter →

arxiv 2606.24974 v1 pith:W6T2Q3RO submitted 2026-06-23 cs.LG cs.AI

David A. Kelly , Nathan Blake This is my paper · ORCID
classification cs.LGcs.AI
keywords spectralentropyexplainableAIECGarrhythmiaclassificationsignalnoisepost-hocXAI
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes using spectral entropy to measure the noise that post-hoc explainability techniques introduce into the outputs of deep learning models. It demonstrates this approach on an ECG dataset for classifying arrhythmias using various XAI methods. A sympathetic reader would care because healthcare models require trustworthy explanations, and distinguishing model-derived signal from heuristic-induced noise improves reliability. If the proposal holds, it offers a concrete way to evaluate and compare the quality of different explainability tools on signal data.

What carries the argument

Spectral entropy applied to XAI-generated explanations to quantify introduced signal noise.

What would settle it

Observing that spectral entropy values remain unchanged across XAI techniques or fail to increase when known noise is artificially added to explanations would challenge the claim.

Watch

Extended reading notes

Core claim

Spectral entropy serves as a measure of noise in XAI output, shown to be useful when applied to explanations from different post hoc explainability techniques in an ECG arrhythmia classification task.

Load-bearing premise

Spectral entropy can isolate and quantify the noise introduced by XAI heuristics separately from the underlying model signal.

Editorial extensions

If this is right

  • XAI methods can be ranked by the amount of noise they add, as measured by spectral entropy on ECG data.
  • Explanations with lower spectral entropy are closer to the core model signal.
  • The method allows assessment of XAI usefulness in medical signal classification tasks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Similar entropy-based measures might apply to other medical imaging or time-series domains.
  • Integration into XAI toolkits could help practitioners select less noisy explainers.
  • This raises the question of whether spectral entropy correlates with human-interpretable explanation quality.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes spectral entropy as a quantitative measure of noise introduced by heuristics in post-hoc XAI explanations. It claims to demonstrate the measure's usefulness for distinguishing signal from noise when applying different XAI techniques to a deep learning model for arrhythmia classification on ECG data.

Significance. A validated, separable noise metric for XAI outputs would be useful in safety-critical domains such as ECG analysis. The approach is conceptually straightforward and could complement existing attribution methods if it can be shown to isolate XAI-specific perturbations rather than reflecting base-signal complexity or model decision boundaries.

major comments (2)
  1. [Abstract] Abstract: the claim that spectral entropy 'demonstrate[s] its usefulness' is unsupported; the text supplies no methods, dataset description, XAI techniques tested, quantitative results, baselines, or statistical comparisons.
  2. The central assumption that spectral entropy differences across XAI methods isolate heuristic-induced noise from the underlying ECG frequency content or model attributions is not tested. No controlled validation (e.g., entropy on raw model outputs, synthetic signals with known perturbations, or comparison to a ground-truth noise metric) is described.
minor comments (1)
  1. Clarify the precise definition of 'signal' versus 'noise' in the XAI output and how spectral entropy is computed (windowing, normalization, frequency range).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed comments. Below we respond point-by-point to the major concerns. We acknowledge the manuscript is concise and agree that revisions are needed to strengthen the presentation and validation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that spectral entropy 'demonstrate[s] its usefulness' is unsupported; the text supplies no methods, dataset description, XAI techniques tested, quantitative results, baselines, or statistical comparisons.

    Authors: We agree the abstract is high-level and does not contain these specifics. The full manuscript describes the ECG arrhythmia classification task, the dataset, the post-hoc XAI methods applied, and reports spectral entropy values across techniques. To address the concern directly, we will expand the abstract to summarize the dataset, methods, and key quantitative comparisons. revision: yes

  2. Referee: The central assumption that spectral entropy differences across XAI methods isolate heuristic-induced noise from the underlying ECG frequency content or model attributions is not tested. No controlled validation (e.g., entropy on raw model outputs, synthetic signals with known perturbations, or comparison to a ground-truth noise metric) is described.

    Authors: The current experiments apply spectral entropy to explanations produced by different XAI methods on identical ECG inputs and model outputs, with the observed differences attributed to the distinct heuristics. We acknowledge that explicit controls (raw outputs without XAI, synthetic signals, or ground-truth noise metrics) are not included. We will add a controlled validation subsection using synthetic perturbations to isolate XAI-specific effects. revision: partial

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical proposal with no derivation chain or fitted predictions

full rationale

The paper proposes spectral entropy as a noise metric for XAI explanations on ECG arrhythmia classifiers and demonstrates it empirically across post-hoc methods. No equations, parameter fits, uniqueness theorems, or self-citations appear in the provided text that would reduce any claimed result to its own inputs by construction. The central claim is a measurement heuristic validated on data rather than a closed mathematical derivation, so the work is self-contained.

Assumptions & free parameters 0 free parameters · 0 assumptions · 0 invented entities

Abstract only; no information provided on free parameters, axioms, or invented entities.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantifying Explainable AI-introduced signal noise on ECG data with Spectral Entropy." pith.science (2026). https://pith.science/paper/W6T2Q3RO

@misc{pith2026260624974,
  author       = {Pith},
  title        = {Pith review of: Quantifying Explainable AI-introduced signal noise on ECG data with Spectral Entropy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W6T2Q3RO}},
  note         = {Machine review of arXiv:2606.24974}
}
read the original 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.

Figures

Figures reproduced from arXiv: 2606.24974 by the authors.

Figure 1
Figure 1. A single ECG complex. Schematic taken from [1] [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Example power spectral density (y-axis) plots for some of the XAI tools. The spread of noise is surprisingly wide, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. PVC ECG with approximate location marked between [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Explanations from different XAI tools for the PVC in Figure 3. The approximate ground truth is indicated between [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 3 canonical work pages

  1. [1]

    Electrocardiogram 1: purpose, physiology and practicalities,

    S. Jarvis, “Electrocardiogram 1: purpose, physiology and practicalities,”Nurs Times, vol. 117, no. 6, pp. 22–6, 2021

    2021

  2. [2]

    A unified approach to interpreting model predictions,

    S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” inNIPS, pp. 4765–4774

  3. [3]

    Grad-CAM: Visual explana- tions from deep networks via gradient-based localiza- tion,

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-CAM: Visual explana- tions from deep networks via gradient-based localiza- tion,” inInternational Conference on Computer Vision (ICCV). IEEE, 2017, pp. 618–626

    2017

  4. [4]

    A review of evaluation approaches for explainable ai with applications in cardiology,

    A. M. Salih, I. B. Galazzo, P. Gkontra, E. Rauseo, A. M. Lee, K. Lekadir, P. Radeva, S. E. Petersen, and G. Menegaz, “A review of evaluation approaches for explainable ai with applications in cardiology,”Artificial Intelligence Review, vol. 57, no. 9, p. 240, 2024

    2024

  5. [5]

    Essential proper- ties and explanation effectiveness of explainable artificial intelligence in healthcare: A systematic review,

    J. Jung, H. Lee, H. Jung, and H. Kim, “Essential proper- ties and explanation effectiveness of explainable artificial intelligence in healthcare: A systematic review,”Heliyon, vol. 9, no. 5, 2023

    2023

  6. [6]

    Future-ai: International con- sensus guideline for trustworthy and deployable artificial intelligence in healthcare,

    K. Lekadir, A. F. Frangi, A. R. Porras, B. Glocker, C. Cintas, C. P. Langlotz, E. Weicken, F. W. Asselbergs, F. Prior, G. S. Collinset al., “Future-ai: International con- sensus guideline for trustworthy and deployable artificial intelligence in healthcare,”bmj, vol. 388, 2025

    2025

  7. [7]

    Causal explanations for image classifiers,

    H. Chockler, D. Kelly, D. Kroening, and Y . Sun, “Causal explanations for image classifiers,”Journal of Artificial Intelligence Research, Mar. 2026

    2026

  8. [8]

    Sufficient, necessary and complete causal explanations in image classification,

    D. A. Kelly and H. Chockler, “Sufficient, necessary and complete causal explanations in image classification,” arXiv preprint arXiv:2507.23497, 2025

    work page arXiv 2025

Show all 24 references
  1. [9]

    Why Should I Trust You?: Explaining the predictions of any classifier,

    M. T. Ribeiro, S. Singh, and C. Guestrin, “Why Should I Trust You?: Explaining the predictions of any classifier,” inProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ser. KDD ’16, 2016, p. 1135–1144

    2016

  2. [10]

    Face: Faithful automatic concept extraction,

    D. Bhusal, M. Clifford, S. Rampazzi, and N. Rastogi, “Face: Faithful automatic concept extraction,” inThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

  3. [11]

    Guidelines and evaluation of clinical explainable ai in medical image analysis,

    W. Jin, X. Li, M. Fatehi, and G. Hamarneh, “Guidelines and evaluation of clinical explainable ai in medical image analysis,”Medical image analysis, vol. 84, p. 102684, 2023

    2023

  4. [12]

    Integrating explainable ai in medical devices: Technical, clinical and regu- latory insights and recommendations,

    D. Alattal, A. K. Azar, P. Myles, R. Branson, H. Ab- dulhussein, and A. Tucker, “Integrating explainable ai in medical devices: Technical, clinical and regu- latory insights and recommendations,”arXiv preprint arXiv:2505.06620, 2025

    2025

  5. [13]

    Solving the explainable ai conundrum by bridging clinicians’ needs and developers’ goals,

    N. Bienefeld, J. M. Boss, R. L ¨uthy, D. Brodbeck, J. Az- zati, M. Blaser, J. Willms, and E. Keller, “Solving the explainable ai conundrum by bridging clinicians’ needs and developers’ goals,”NPJ Digital Medicine, vol. 6, no. 1, p. 94, 2023

    2023

  6. [14]

    T. M. Cover and J. A. Thomas,Elements of Information Theory (Wiley Series in Telecommunications and Signal Processing). USA: Wiley-Interscience, 2006

    2006

  7. [15]

    Arrhythmia classifier using convolutional neural network with adaptive loss-aware multi-bit net- works quantization,

    H. Sun, A. Wang, N. Pu, Z. Li, J. Huang, H. Liu, and Z. Qi, “Arrhythmia classifier using convolutional neural network with adaptive loss-aware multi-bit net- works quantization,” in2021 2nd International Confer- ence on Artificial Intelligence and Computer Engineering (ICAICE)...

    2021

  8. [16]

    The impact of the mit- bih arrhythmia database,

    G. B. Moody and R. G. Mark, “The impact of the mit- bih arrhythmia database,”IEEE engineering in medicine and biology magazine, vol. 20, no. 3, pp. 45–50, 2001

    2001

  9. [17]

    Learning important features through propagating activation differ- ences,

    A. Shrikumar, P. Greenside, and A. Kundaje, “Learning important features through propagating activation differ- ences,” inProceedings of the 34th International Confer- ence on Machine Learning - Volume 70, ser. ICML’17. JMLR.org, 2017, p. 3145–3153

    2017

  10. [18]

    Mrxai: Black-box explainability for image classifiers in a medical setting,

    N. Blake, D. Kelly, H. Chockler, A. Chanchal, and S. Pe ˜na, “Mrxai: Black-box explainability for image classifiers in a medical setting,”CEUR Workshop Pro- ceedings, vol. 4059, 2025

    2025

  11. [19]

    Limesegment: Meaningful, realis- tic time series explanations,

    T. Sivill and P. Flach, “Limesegment: Meaningful, realis- tic time series explanations,” inInternational Conference on Artificial Intelligence and Statistics. PMLR, 2022, pp. 3418–3433

    2022

  12. [20]

    Multiple different black box explanations for image classifiers,

    H. Chockler, D. A. Kelly, and D. Kroening, “Multiple different black box explanations for image classifiers,” inECAI 2025 - 28th European Conference on Artificial Intelligence, ser. Frontiers in Artificial Intelligence and Applications, vol. 413, 2025, pp. 699 – 706

    2025

  13. [21]

    RISE: randomized input sampling for explanation of black-box models,

    V . Petsiuk, A. Das, and K. Saenko, “RISE: randomized input sampling for explanation of black-box models,” in British Machine Vision Conference (BMVC). BMV A Press, 2018

    2018

  14. [22]

    Yeh, C.-Y

    C.-K. Yeh, C.-Y . Hsieh, A. S. Suggala, D. I. Inouye, and P. Ravikumar,On the (in)fidelity and sensitivity of explanations. Red Hook, NY , USA: Curran Associates Inc., 2019

    2019

  15. [23]

    Explanation in artificial intelligence: Insights from the social sciences,

    T. Miller, “Explanation in artificial intelligence: Insights from the social sciences,”Artificial intelligence, vol. 267, pp. 1–38, 2019

    2019

  16. [24]

    The meta- evaluation problem in explainable ai: identifying re- liable estimators with metaquantus,

    A. Hedstr ¨om, P. Bommer, K. K. Wickstrøm, W. Samek, S. Lapuschkin, and M. M.-C. H ¨ohne, “The meta- evaluation problem in explainable ai: identifying re- liable estimators with metaquantus,”arXiv preprint arXiv:2302.07265, 2023

    2023

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.