pith. sign in

arxiv: 2506.09255 · v1 · submitted 2025-05-23 · 📡 eess.SP · cs.LG

AI-Driven SEEG Channel Ranking for Epileptogenic Zone Localization

Saeed Hashemi , Genchang Peng , Mehrdad Nourani , Omar Nofal , Jay Harvey This is my paper

Pith reviewed 2026-05-19 12:55 UTC · model grok-4.3

classification 📡 eess.SP cs.LG
keywords SEEGepileptogenic zoneXGBoostSHAPchannel rankingictal periodsmachine learningepilepsy surgery
0
0 comments X

The pith

XGBoost with SHAP ranks SEEG channels by their contribution to seizures and extends clinician selections to flag more potential epileptogenic zones.

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

The paper trains an XGBoost classification model on ictal periods of SEEG recordings to learn which signal features distinguish important channels. It then applies SHAP scoring to produce an ordered ranking of channels according to how much each one contributes to seizure activity. A channel extension step widens the search beyond the channels initially chosen by clinicians. This approach is tested on data from five patients and is presented as delivering accurate, consistent, and explainable rankings that could reduce the burden of manual visual review of hundreds of channels.

Core claim

Training an XGBoost model on ictal SEEG data allows SHAP values to rank channels by their contribution to seizures; adding a channel extension strategy then identifies suspicious epileptogenic zones outside the set selected by clinicians, with the combined method showing promising accuracy, consistency, and explainability on recordings from five patients.

What carries the argument

XGBoost classifier trained on ictal periods, followed by SHAP scoring to rank channel contributions and a channel extension strategy to expand the search space beyond initial clinical selections.

If this is right

  • Clinicians receive an ordered list of channels prioritized by seizure contribution to guide focused review.
  • The extension strategy surfaces additional suspicious zones that lie outside the initial clinical selection.
  • SHAP-based rankings supply explicit explanations for why each channel is considered important.
  • Validation on five patients indicates the rankings achieve high accuracy and consistency with clinical judgment.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same ranking pipeline could be applied to larger patient cohorts to test whether it shortens pre-surgical evaluation time.
  • Integration with other imaging modalities might further refine the identification of multi-focal zones.
  • If the extension step proves reliable, it could help detect epileptogenic tissue in complex cases where clinicians initially select limited channels.

Load-bearing premise

The SHAP scores produced by the XGBoost model trained on ictal data correctly identify channels that belong to the epileptogenic zone rather than merely correlating with seizure activity.

What would settle it

A follow-up study that compares the top-ranked and extended channels against the actual tissue resected in patients who achieved seizure freedom after surgery, checking whether the ranked channels match the removed zones that stopped seizures.

Figures

Figures reproduced from arXiv: 2506.09255 by Genchang Peng, Jay Harvey, Mehrdad Nourani, Omar Nofal, Saeed Hashemi.

Figure 1
Figure 1. Figure 1: Flowchart of proposed methodology [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: SEEG example: (a) Implantation map showing 11 electrodes, with the LA and LB electrodes (clinician [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Defining two classes for SEEG data [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of electrode extension (blue) and zone extension (black). [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: SHAP-based ranking for patient 1000. (a) Electrode extension, where LA9 is identified as a new finding. (b) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: SHAP-based ranking for patient 1500. (a) Electrode extension, where LB9 is identified as a new finding. (b) [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: SHAP-based ranking for patient 1300, 1600, and 1700. Sub-figures labeled (i) refer to electrode extension [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

Stereo-electroencephalography (SEEG) is an invasive technique to implant depth electrodes and collect data for pre-surgery evaluation. Visual inspection of signals recorded from hundreds of channels is time consuming and inefficient. We propose a machine learning approach to rank the impactful channels by incorporating clinician's selection and computational finding. A classification model using XGBoost is trained to learn the discriminative features of each channel during ictal periods. Then, the SHapley Additive exPlanations (SHAP) scoring is utilized to rank SEEG channels based on their contribution to seizures. A channel extension strategy is also incorporated to expand the search space and identify suspicious epileptogenic zones beyond those selected by clinicians. For validation, SEEG data for five patients were analyzed showing promising results in terms of accuracy, consistency, and explainability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript proposes a machine learning framework for ranking SEEG channels to aid in epileptogenic zone localization. An XGBoost classifier is trained on discriminative features from ictal periods, SHAP values are used to rank channels by their contribution to seizure classification, and a channel extension strategy is applied to identify additional suspicious zones. The approach is tested on SEEG recordings from five patients, with claims of promising accuracy, consistency, and explainability.

Significance. If substantiated with rigorous validation, this method could offer a valuable tool for clinicians by combining data-driven insights with expert selections to streamline the analysis of high-channel-count SEEG data. The emphasis on explainability via SHAP is particularly relevant for medical decision support. However, the preliminary nature of the validation limits its immediate clinical significance.

major comments (3)
  1. [Abstract] The abstract states that analysis of five patients showed 'promising results in terms of accuracy, consistency, and explainability' but provides no quantitative metrics, such as accuracy percentages, overlap scores with clinician selections, or statistical measures, which are necessary to support the central claim.
  2. [Methods] The training procedure for the XGBoost model, including the definition of positive/negative labels for channels (e.g., based on clinician selection or post-surgical outcomes), feature engineering for ictal periods, and any cross-validation or hold-out testing, is not detailed, undermining the reliability of the derived SHAP rankings.
  3. [Results] Validation is limited to five patients without specifying the ground truth for epileptogenic zones (such as resected tissue or seizure-free outcomes) or comparisons to clinician-only selections, making it unclear if the SHAP-based ranking and channel extension add value beyond the small cohort.
minor comments (1)
  1. [Abstract] The term 'channel extension strategy' could be clarified earlier to explain how it expands the search space without introducing excessive false positives.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript proposing an XGBoost and SHAP-based framework for SEEG channel ranking in epileptogenic zone localization. We address each major comment point by point below, with revisions planned to improve clarity and detail where appropriate.

read point-by-point responses

  1. Referee: [Abstract] The abstract states that analysis of five patients showed 'promising results in terms of accuracy, consistency, and explainability' but provides no quantitative metrics, such as accuracy percentages, overlap scores with clinician selections, or statistical measures, which are necessary to support the central claim.

    Authors: We agree that the abstract would benefit from quantitative support. In the revised version, we will add specific metrics including classification accuracy, overlap scores (e.g., Dice coefficient) with clinician selections, and relevant statistical measures from the five-patient analysis to substantiate the claims of promising results. revision: yes

  2. Referee: [Methods] The training procedure for the XGBoost model, including the definition of positive/negative labels for channels (e.g., based on clinician selection or post-surgical outcomes), feature engineering for ictal periods, and any cross-validation or hold-out testing, is not detailed, undermining the reliability of the derived SHAP rankings.

    Authors: We acknowledge the need for greater methodological transparency. We will revise the Methods section to fully detail the label assignment process (incorporating clinician selections and post-surgical outcomes), the ictal feature engineering steps, and the cross-validation or testing procedures used to support the SHAP-based rankings. revision: yes

  3. Referee: [Results] Validation is limited to five patients without specifying the ground truth for epileptogenic zones (such as resected tissue or seizure-free outcomes) or comparisons to clinician-only selections, making it unclear if the SHAP-based ranking and channel extension add value beyond the small cohort.

    Authors: We will clarify the ground truth definition in the Results section, specifying reliance on clinician selections supplemented by available post-surgical outcomes. We will also add explicit comparisons demonstrating the added value of SHAP rankings and channel extensions over clinician-only selections. The five-patient cohort is preliminary by design for this initial study, and we will note this limitation more explicitly while highlighting consistency across cases. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation uses standard ML training and post-hoc explanation

full rationale

The paper's core chain trains an XGBoost classifier on ictal-period channel features (incorporating clinician selections), applies SHAP for ranking, and adds a channel-extension step to flag additional zones. None of these steps reduce the output ranking to the input labels or fitted parameters by construction, nor invoke self-citations, uniqueness theorems, or ansatzes from prior author work. Validation on five patients is presented as an independent empirical check rather than a definitional tautology. The method is a conventional supervised learning plus explainability pipeline applied to SEEG data; its claims rest on the model's learned discriminative power and external patient outcomes, not on re-labeling its own training targets as predictions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach depends on standard assumptions in machine learning for signal classification and the validity of clinician-selected channels as a starting point for extension.

axioms (1)
  • domain assumption Ictal periods in SEEG signals contain features discriminative for epileptogenic zones
    This is used to train the XGBoost model on channel data during seizure periods.

pith-pipeline@v0.9.0 · 5676 in / 1190 out tokens · 38643 ms · 2026-05-19T12:55:01.204739+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages · 2 internal anchors

  1. [1]

    Early identification of candidates for epilepsy surgery: a multicenter, machine learning, prospective validation study,

    B. D. Wissel et al. , “Early identification of candidates for epilepsy surgery: a multicenter, machine learning, prospective validation study,”Neurology, vol. 102, no. 4, p. e208048, 2024

    work page 2024

  2. [2]

    Stereotactic EEG practices: a survey of United States tertiary referral epilepsy centers,

    J. Gavvala et al. , “Stereotactic EEG practices: a survey of United States tertiary referral epilepsy centers,” J. Clin. Neurophysiol., vol. 39, no. 6, pp. 474–480, 2022

    work page 2022

  3. [3]

    Modeling and analysis of seizure network using SEEG for pre-surgery evaluation,

    G. Peng, M. Nourani, H. Dave, and J. Harvey, “Modeling and analysis of seizure network using SEEG for pre-surgery evaluation,” in Proc. IEEE Int. Conf. Bioinformatics and Bioengineering (BIBE) , Nov. 2022, pp. 316–321

    work page 2022

  4. [4]

    French guidelines on stereoelectroencephalography (SEEG),

    J. Isnard et al., “French guidelines on stereoelectroencephalography (SEEG),” Neurophysiol. Clin., vol. 48, no. 1, pp. 5–13, 2018

    work page 2018

  5. [5]

    Defining epileptogenic networks: contribution of SEEG and signal analysis,

    F. Bartolomei et al. , “Defining epileptogenic networks: contribution of SEEG and signal analysis,” Epilepsia, vol. 58, no. 7, pp. 1131–1147, 2017

    work page 2017

  6. [6]

    ILAE official report: a practical clinical definition of epilepsy,

    R. S. Fisher et al., “ILAE official report: a practical clinical definition of epilepsy,”Epilepsia, vol. 55, no. 4, pp. 475–482, 2014

    work page 2014

  7. [7]

    Increased nationwide use of stereoencephalography for intracranial epilepsy electroen- cephalography recordings,

    H. Abou-Al-Shaar et al., “Increased nationwide use of stereoencephalography for intracranial epilepsy electroen- cephalography recordings,” J. Clin. Neurosci., vol. 53, pp. 132–134, 2018

    work page 2018

  8. [8]

    Is SEEG safe? A systematic review and meta-analysis of stereo- electroencephalography–related complications,

    J. P. Mullin et al. , “Is SEEG safe? A systematic review and meta-analysis of stereo- electroencephalography–related complications,” Epilepsia, vol. 57, no. 3, pp. 386–401, 2016

    work page 2016

  9. [9]

    Stereoelectroencephalography in epilepsy, cognitive neurophysiology, and psychiatric disease: safety, efficacy, and place in therapy,

    B. E. Youngerman, F. A. Khan, and G. M. McKhann, “Stereoelectroencephalography in epilepsy, cognitive neurophysiology, and psychiatric disease: safety, efficacy, and place in therapy,”Neuropsychiatr . Dis. Treat., vol. 15, 2019

    work page 2019

  10. [10]

    Automatic localization of seizure onset zone from high-frequency SEEG signals: A preliminary study,

    L. Xiao et al., “Automatic localization of seizure onset zone from high-frequency SEEG signals: A preliminary study,”IEEE J. Transl. Eng. Health Med. , vol. 9, pp. 1–10, 2021

    work page 2021

  11. [11]

    SEEG functional connectivity measures to identify epileptogenic zones: stability, medication influence, and recording condition,

    D. L. Paulo et al., “SEEG functional connectivity measures to identify epileptogenic zones: stability, medication influence, and recording condition,”Neurology, vol. 98, no. 20, pp. e2060–e2072, 2022

    work page 2022

  12. [12]

    Seizure prediction using directed transfer function and convolution neural network on intracra- nial EEG,

    G. Wang et al., “Seizure prediction using directed transfer function and convolution neural network on intracra- nial EEG,” IEEE Trans. Neural Syst. Rehabil. Eng. , vol. 28, no. 12, pp. 2711–2720, 2020

    work page 2020

  13. [13]

    SEEG-based epileptic seizure network modeling and analysis for pre-surgery evaluation,

    G. Peng, M. Nourani, H. Dave, and J. Harvey, “SEEG-based epileptic seizure network modeling and analysis for pre-surgery evaluation,”Comput. Biol. Med., vol. 167, p. 107692, 2023

    work page 2023

  14. [14]

    XGBoost: A scalable tree boosting system,

    T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., 2016, pp. 785–794

    work page 2016

  15. [15]

    A Unified Approach to Interpreting Model Predictions

    S. Lundberg, “A unified approach to interpreting model predictions,” arXiv preprint arXiv:1705.07874, 2017. Accepted to be presented at the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2025). This version is submitted to arXiv prior to final IEEE formatting and publication

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  16. [16]

    A deep reinforcement learning model for predictive maintenance planning of road assets: Integrating LCA and LCCA,

    M. Latifi, F. G. Darvishvand, O. Khandel, and M. L. Nowsoud, “A deep reinforcement learning model for predictive maintenance planning of road assets: Integrating LCA and LCCA,” 2023. [Online]. Available: https://arxiv.org/abs/2112.12589

    work page arXiv 2023

  17. [17]

    Consistent Individualized Feature Attribution for Tree Ensembles

    S. M. Lundberg, G. G. Erion, and S.-I. Lee, “Consistent individualized feature attribution for tree ensembles,” arXiv preprint arXiv:1802.03888, 2018. Accepted to be presented at the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2025). This version is submitted to arXiv prior to final IEEE formatting an...

    work page internal anchor Pith review Pith/arXiv arXiv 2018