pith. sign in

arxiv: 1907.02596 · v1 · pith:CBIUTYAInew · submitted 2019-07-03 · 📡 eess.SP · cs.LG· q-bio.NC· stat.ML

QuPWM: Feature Extraction Method for MEG Epileptic Spike Detection

Abderrazak Chahid , Fahad Albalawi , Turky Nayef Alotaiby , Majed Hamad Al-Hameed , Saleh Alshebeili , Taous-Meriem Laleg-Kirati This is my paper

Pith reviewed 2026-05-25 10:06 UTC · model grok-4.3

classification 📡 eess.SP cs.LGq-bio.NCstat.ML
keywords MEGepileptic spike detectionfeature extractionPosition Weight MatrixquantizationSVMmachine learningneurological signals
0
0 comments X

The pith

QuPWM combines PWM with quantization to extract compact MEG features for SVM-based epileptic spike detection at up to 98% accuracy.

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

The paper introduces QuPWM as a feature extraction step that applies a uniform quantizer to the Position Weight Matrix on sliding windows of MEG recordings. These compact features feed into an SVM classifier to label segments as containing epileptic spikes or not. The method is evaluated on 3104 balanced samples drawn from 16 subjects via 100-point windows stepped by 2 points, yielding 98% average accuracy in 5-fold cross-validation. A sympathetic reader would care because the approach targets the current reliance on slow, subjective visual inspection for localizing epileptogenic zones in epilepsy patients.

Core claim

The authors establish that PWM combined with uniform quantization produces reduced-size feature vectors from MEG time series that an SVM can classify to detect epileptic spikes, reaching an average accuracy of 98% under 5-fold cross-validation on a balanced set of 3104 samples extracted from eight healthy and eight epileptic subjects.

What carries the argument

QuPWM, the quantized Position Weight Matrix that turns 100-sample MEG windows into compact feature vectors for SVM classification.

If this is right

  • The quantized PWM approach reduces feature vector size relative to unquantized methods while preserving high SVM accuracy.
  • Automatic classification via this pipeline can replace manual visual scanning of MEG recordings for spike detection.
  • Performance is demonstrated on balanced data drawn from both healthy and epileptic subjects.
  • Five-fold cross-validation on the extracted samples supports internal consistency of the detection results.

Where Pith is reading between the lines

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

  • The method could shorten the time required for epileptogenic zone localization if it maintains accuracy on larger clinical archives.
  • Subject-independent splitting rather than random cross-validation would be needed to rule out leakage from overlapping windows or individual recording traits.
  • The same quantized-matrix construction might transfer to EEG spike detection or other transient event tasks in brain signals.
  • Integration into existing MEG review software would require testing on continuous, unbalanced recordings rather than pre-extracted balanced frames.

Load-bearing premise

The 3104 sliding-window samples from only 16 subjects form a representative dataset free of labeling bias or overlap-induced artifacts that would inflate cross-validation accuracy.

What would settle it

Accuracy falling below 85% when the trained model is evaluated on an independent set of MEG recordings from new subjects or different scanners would show the reported performance does not hold outside the original cohort.

Figures

Figures reproduced from arXiv: 1907.02596 by Abderrazak Chahid, Fahad Albalawi, Majed Hamad Al-Hameed, Saleh Alshebeili, Taous-Meriem Laleg-Kirati, Turky Nayef Alotaiby.

Figure 1
Figure 1. Figure 1: Classification framework subdivided into three stages: MEG records pre-processing, feature generation and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the brain abnormal activities in different types of epileptic seizures [built based on [12] [13]]. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of healthy and epileptic subject MEG recording. Zoomed plot shows a segment where an epileptic [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example of MEG signal quantization using four levels. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The quantization of the real-valued sequence with a resolution [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Examples of three binary sequences corresponding to mono-mers, di-mers and tri-mers, respectively. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The statistical properties of the epileptic spikes duration for eight epileptic subjects. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
read the original abstract

Epilepsy is a neurological disorder classified as the second most serious neurological disease known to humanity, after stroke. Localization of the epileptogenic zone is an important step for epileptic patient treatment, which starts with epileptic spike detection. The common practice for spike detection of brain signals is via visual scanning of the recordings, which is a subjective and a very time-consuming task. Motivated by that, this paper focuses on using machine learning for automatic detection of epileptic spikes in magnetoencephalography (MEG) signals. First, we used the Position Weight Matrix (PWM) method combined with a uniform quantizer to generate useful features. Second, the extracted features are classified using a Support Vector Machine (SVM) for the purpose of epileptic spikes detection. The proposed technique shows great potential in improving the spike detection accuracy and reducing the feature vector size. Specifically, the proposed technique achieved average accuracy up to 98\% in using 5-folds cross-validation applied to a balanced dataset of 3104 samples. These samples are extracted from 16 subjects where eight are healthy and eight are epileptic subjects using a sliding frame of size of 100 samples-points with a step-size of 2 sample-points

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

2 major / 2 minor

Summary. The manuscript proposes QuPWM, a feature extraction pipeline that applies the Position Weight Matrix (PWM) method followed by uniform quantization to MEG signals, with the resulting features fed to an SVM classifier for epileptic spike detection. The central claim is that this yields an average accuracy of up to 98% under 5-fold cross-validation on a balanced set of 3104 samples drawn from 16 subjects (8 healthy, 8 epileptic) via sliding windows of length 100 with step size 2.

Significance. If the accuracy result survives proper subject-wise or block-wise validation, the approach could provide a compact, interpretable feature representation that reduces dimensionality relative to raw MEG segments while supporting automated spike detection. The paper does not, however, supply baseline comparisons, feature counts, or error bars, so the incremental value over existing PWM or wavelet-based pipelines remains unclear even if the numerical claim is upheld.

major comments (2)
  1. [Abstract] Abstract: the reported 98% accuracy rests on 5-fold CV performed on 3104 pooled samples obtained with a step size of 2 (98-sample overlap). No statement indicates that folds were formed at the subject level or via contiguous temporal blocks; random splitting on overlapping windows therefore permits near-identical signal content to appear in both train and test folds, rendering the accuracy figure non-diagnostic of feature quality.
  2. [Abstract] Dataset construction paragraph (implied by abstract description): the 3104-sample collection is formed from only 16 subjects with heavy temporal overlap and no reported check for labeling consistency or independence across windows. This construction directly determines the validity of the cross-validation results that constitute the paper's primary empirical support.
minor comments (2)
  1. [Abstract] Abstract: 'sample-points' should read 'sample points'.
  2. [Abstract] Abstract: the quantizer bin count, final feature dimensionality after QuPWM, and SVM hyperparameter settings are not stated, although they are listed among the free parameters.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and for highlighting the critical issue of data leakage in our cross-validation procedure. The comments correctly identify that our current evaluation does not ensure independence between training and test sets given the overlapping windows. We address each point below and will revise the manuscript to implement subject-wise validation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the reported 98% accuracy rests on 5-fold CV performed on 3104 pooled samples obtained with a step size of 2 (98-sample overlap). No statement indicates that folds were formed at the subject level or via contiguous temporal blocks; random splitting on overlapping windows therefore permits near-identical signal content to appear in both train and test folds, rendering the accuracy figure non-diagnostic of feature quality.

    Authors: We agree that random 5-fold cross-validation on pooled overlapping windows allows near-identical content to leak across folds, which undermines the reported accuracy. In the revised manuscript we will replace this with subject-wise cross-validation: all windows from any given subject will be assigned entirely to either the training or test fold. Updated accuracy, sensitivity, and specificity figures under this protocol will be reported, along with the number of features used. revision: yes

  2. Referee: [Abstract] Dataset construction paragraph (implied by abstract description): the 3104-sample collection is formed from only 16 subjects with heavy temporal overlap and no reported check for labeling consistency or independence across windows. This construction directly determines the validity of the cross-validation results that constitute the paper's primary empirical support.

    Authors: The referee is correct that the manuscript provides no explicit verification of labeling consistency across overlapping windows or of temporal independence. We will add a description of the labeling process (expert visual inspection of each window) and will adopt the subject-wise partitioning described above. In addition, we will report the total number of original MEG recordings per subject and the resulting feature dimensionality after quantization. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical pipeline with independent CV evaluation

full rationale

The paper presents a straightforward feature extraction pipeline (PWM + uniform quantization) followed by SVM classification and reports an empirical accuracy from 5-fold cross-validation on a constructed dataset. No equations, fitted parameters, or self-citations are shown that reduce the reported accuracy or method to a definitional identity or input by construction. The derivation chain consists of independent steps (feature generation then classification) whose performance is measured externally on the dataset; no load-bearing premise collapses to a self-citation or renaming. This is the normal case of a self-contained empirical method.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

The claim rests on treating MEG time series as sequence data amenable to PWM motif analysis, on the uniform quantizer preserving discriminative information, and on the 16-subject balanced set being sufficient for generalization. No free parameters are numerically specified; the quantizer bin count and SVM hyperparameters are implicit but unstated.

free parameters (2)
  • quantizer bin count
    Uniform quantizer requires a choice of discrete levels; value not reported.
  • SVM hyperparameters
    Kernel type, regularization parameter C, and gamma (if RBF) must be chosen or tuned; not specified.
axioms (1)
  • domain assumption MEG signals contain local statistical patterns that can be captured by a position weight matrix after quantization
    The method directly imports PWM from sequence analysis to time-series windows without additional justification in the abstract.
invented entities (1)
  • QuPWM no independent evidence
    purpose: Name for the quantized PWM feature extractor
    New label introduced for the combination of PWM and uniform quantization; no independent evidence outside the paper.

pith-pipeline@v0.9.0 · 5785 in / 1522 out tokens · 29669 ms · 2026-05-25T10:06:16.517027+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

  1. [1]

    Fisher, Walter van Emde Boas, Warren Blume, Christian Elger, Pierre Genton, Phillip Lee, and Jerome Engel Jr

    Robert S. Fisher, Walter van Emde Boas, Warren Blume, Christian Elger, Pierre Genton, Phillip Lee, and Jerome Engel Jr. Epileptic seizures and epilepsy: Definitions proposed by the international league against epilepsy (ilae) and the international bureau for epilepsy (ibe). Epilepsia, 46(4):470–472, 2005

    work page 2005

  2. [2]

    Magnetoencephalography—theory, instrumentation, and applications to noninvasive studies of the working human brain

    Matti Hämäläinen, Riitta Hari, Risto J Ilmoniemi, Jukka Knuutila, and Olli V Lounasmaa. Magnetoencephalography—theory, instrumentation, and applications to noninvasive studies of the working human brain. Reviews of Modern Physics, 65(2):413–497, apr 1993

    work page 1993

  3. [3]

    Magnetoencephalography (MEG): Past, current and future perspectives for improved differentiation and treatment of epilepsies

    Hermann Stefan and Eugen Trinka. Magnetoencephalography (MEG): Past, current and future perspectives for improved differentiation and treatment of epilepsies. Seizure, 44:121–124, jan 2017

    work page 2017

  4. [4]

    Epileptogenic zone localization using magnetoencephalography predicts seizure freedom in epilepsy surgery

    Dario J Englot, Srikantan S Nagarajan, Brandon S Imber, Kunal P Raygor, Susanne M Honma, Danielle Mizuiri, Mary Mantle, Robert C Knowlton, Heidi E Kirsch, and Edward F Chang. Epileptogenic zone localization using magnetoencephalography predicts seizure freedom in epilepsy surgery. Epilepsia, 56(6):949–958, jun 2015

    work page 2015

  5. [5]

    Magnetoencephalography for brain electrophysiology and imaging

    Sylvain Baillet. Magnetoencephalography for brain electrophysiology and imaging. Nature Neuroscience, 20:327, feb 2017

    work page 2017

  6. [6]

    Abd El-Samie, Turky N

    Fathi E. Abd El-Samie, Turky N. Alotaiby, Muhammad Imran Khalid, Saleh A. Alshebeili, and Saeed A. Aldosari. A Review of EEG and MEG Epileptic Spike Detection Algorithms. IEEE Access, 6:60673–60688, 2018

    work page 2018

  7. [7]

    Automated interictal spike detection and source localization in magnetoencephalography using independent components analysis and spatio-temporal clustering

    A Ossadtchi, S Baillet, J C Mosher, D Thyerlei, W Sutherling, and R M Leahy. Automated interictal spike detection and source localization in magnetoencephalography using independent components analysis and spatio-temporal clustering. Clinical Neurophysiology, 115(3):508–522, 2004

    work page 2004

  8. [8]

    Epileptic MEG Spikes Detection Using Common Spatial Patterns and Linear Discriminant Analysis

    M I Khalid, T Alotaiby, S A Aldosari, S A Alshebeili, M H Al-Hameed, F S Y Almohammed, and T S Alotaibi. Epileptic MEG Spikes Detection Using Common Spatial Patterns and Linear Discriminant Analysis. IEEE Access, 4:4629–4634, 2016

    work page 2016

  9. [9]

    Epileptic MEG Spikes Detection Using Amplitude Thresholding and Dynamic Time Warping

    M I Khalid, T N Alotaiby, S A Aldosari, S A Alshebeili, M H Alhameed, and V Poghosyan. Epileptic MEG Spikes Detection Using Amplitude Thresholding and Dynamic Time Warping. IEEE Access, 5:11658–11667, 2017

    work page 2017

  10. [10]

    Alotaiby, Saleh A

    Abderrazak Chahid, Turky N. Alotaiby, Saleh A. Alshebeili, and Taous-Meriem Laleg-Kirati. Feature Generation and Dimensionality Reduction using the Discrete Spectrum of the Schrödinger Operator for Epileptic Spikes Detection. In 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2019

    work page 2019

  11. [11]

    Semi-classical signal analysis.Mathematics of Control, Signals, and Systems, 25(1):37–61, 2013

    Taous-Meriem Laleg-Kirati, Emmanuelle Crépeau, and Michel Sorine. Semi-classical signal analysis.Mathematics of Control, Signals, and Systems, 25(1):37–61, 2013

    work page 2013

  12. [12]

    Todd B. Bates. Human brain 3D model– stock image. Apr 2012

    work page 2012

  13. [13]

    Epileptic seizures and depression may share a common genetic cause, study suggests

    Alexmit. Epileptic seizures and depression may share a common genetic cause, study suggests. Jan 2018

    work page 2018

  14. [14]

    Spatiotemporal signal space separation method for rejecting nearby interference in MEG measurements

    S Taulu and J Simola. Spatiotemporal signal space separation method for rejecting nearby interference in MEG measurements. Physics in Medicine and Biology, 51(7):1759–1768, mar 2006

    work page 2006

  15. [15]

    Gray and David L

    Robert M. Gray and David L. Neuhoff. Quantization.IEEE Transactions on Information Theory, pages 2325–2383, 1998

    work page 1998

  16. [16]

    Quantization and the method of k-means.IEEE Transactions on Information theory, 28(2):199–205, 1982

    David Pollard. Quantization and the method of k-means.IEEE Transactions on Information theory, 28(2):199–205, 1982

    work page 1982

  17. [17]

    Three-sigma rule

    MS Nikulin. Three-sigma rule. Encyclopedia of Mathematics, available at: http://www. encyclopediaofmath. org/index. php, 2011

    work page 2011

  18. [18]

    The three sigma rule

    Friedrich Pukelsheim. The three sigma rule. The American Statistician, 48(2):88–91, 1994

    work page 1994

  19. [19]

    Use of the ‘perceptron’algorithm to distinguish translational initiation sites in e

    Gary D Stormo, Thomas D Schneider, Larry Gold, and Andrzej Ehrenfeucht. Use of the ‘perceptron’algorithm to distinguish translational initiation sites in e. coli. Nucleic acids research, 10(9):2997–3011, 1982

    work page 1982

  20. [20]

    Gerald Z Hertz and Gary D. Stormo. Identifying dna and protein patterns with statistically significant alignments of multiple sequences. Bioinformatics (Oxford, England), 15(7):563–577, 1999

    work page 1999

  21. [21]

    Computer methods to locate signals in nucleic acid sequences

    Rodger Staden. Computer methods to locate signals in nucleic acid sequences. 1984

    work page 1984

  22. [22]

    POLY AR, a new computer program for prediction of poly (A) sites in human sequences

    Malik Nadeem Akhtar, Syed Abbas Bukhari, Zeeshan Fazal, Raheel Qamar, and Ilham A Shahmuradov. POLY AR, a new computer program for prediction of poly (A) sites in human sequences. BMC genomics, 11(1):646, 2010

    work page 2010

  23. [23]

    Detection of polyadenylation signals in human DNA sequences

    Jack E Tabaska and Michael Q Zhang. Detection of polyadenylation signals in human DNA sequences. Gene, 231(1):77–86, 1999. 13 A PREPRINT - J ULY 8, 2019

    work page 1999

  24. [24]

    Alotaiby, and Laleg-Kirati Alshebeili, Saleh A

    Abderrazak Chahid, Turky N. Alotaiby, and Laleg-Kirati Alshebeili, Saleh A. Taous-Meriem. Position weight matrix, gibbs sampler, and the associated significance tests in motif characterization and prediction.Scientifica, 2012, 2012. 14

    work page 2012