A Compact Stance-Indexed Anterior-Posterior COP Representation for Parkinson's Disease Classification from Plantar VGRF
Md. Sifat, Sania Akter, Akif Islam, Md. Ekramul Hamid
cs.AI
Quick verdict
Claims stance-indexed AP-COP compactly encodes PD-discriminative gait info; nested CV supports it within one database, but external validation is absent.
Summary
This empirical study asks whether a compact, stance-indexed representation of anterior-posterior center of pressure (AP-COP) carries discriminative information for Parkinson's disease classification from plantar vertical ground reaction force. Using the public GaitPDB database (165 participants: 93 with PD, 72 controls), the authors derive ten features — AP-COP position and bilateral asymmetry at five fixed windows of normalized stance (AP-COP10) — and evaluate them in 20x5-fold repeated nested participant-level cross-validation with training-only elastic-net stability selection and a class-balanced L2 logistic classifier capped at six features. AP-COP10 reached AUC 0.894, versus 0.908 for the full 35-candidate bank and 0.856 for the 25 complementary descriptors; under the same pipeline it outperformed harmonized literature-derived COP sets (Alam4 0.809, Jin7 0.784, Tong18 0.746; all Holm p=0.0012). Leave-one-source-study-out pooled AUC was 0.907, and discriminative information concentrated in early stance. Key limitations: all data come from one database and one sensor platform, so the holdout analysis is not external validation; walking-speed confounding is analyzed predictively, not causally; code is available only on request; and the five stance windows were not optimized.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Stance-indexed AP-COP10 achieves participant-level AUC 0.894, outperforming harmonized Alam4 (0.809), Jin7 (0.784), and Tong18 (0.746) under an identical nested pipeline (all Holm p=0.0012).
- Removing AP-COP10 from the 35-candidate bank causes a statistically supported AUC loss (0.908 to 0.856, p=0.0028), while adding the 25 complementary descriptors to AP-COP10 yields only +0.013 (p=0.361).
- Discriminative information is concentrated in early and early-mid stance (Hedges' g=1.303 for position at 25-35%).
- AP-COP10 is robust within GaitPDB: pooled leave-one-source-study-out AUC 0.907, exceeding Full35 (0.872) in every held-out protocol.
Strengths
Fully nested, representation-controlled design
Identical 20x5-fold outer splits, training-only elastic-net stability selection (100 subsamples, 60% threshold, |rho|>=0.85 pruning, <=6 features), and one class-balanced L2 logistic classifier shared by every representation (Sec 3.3.1, Table S13) isolate representation effects from pipeline differences — exactly the confound the authors criticize in prior GaitPDB work (Sec 2.3).
Honest, well-scoped statistics
Holm correction applied across exactly the three primary literature contrasts (all pHolm=0.0012); secondary nested contrasts kept out of the multiplicity family and reported as null (Full35 vs AP-COP10 p=0.361) rather than equivalence, with the missing equivalence margin explicitly acknowledged (Sec 5.2).
Extensive sensitivity program
Disclosed harmonization decisions
Marginal-vs-conditional insight
Weaknesses
No external validation
Artifacts not deposited
Speed confound only predictively addressed
Window design justified only by uniformity
Comparators are subsetted reimplementations
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper asks a representation-level question: does preserving anterior-posterior center-of-pressure (AP-COP) at five fixed windows of normalized stance (5-15%, 25-35%, 45-55%, 65-75%, 85-95%) retain PD-discriminative information that whole-trajectory COP summaries discard? The proposed AP-COP10 (five bilateral position plus five bilateral asymmetry features, Sec 3.2.3) is evaluated three ways: (1) against three harmonized literature-derived COP sets (Alam4, Jin7, Tong18); (2) as a strict subset of a 35-candidate engineered bank (Full35); (3) against the 25 complementary non-AP-COP descriptors. Central claims: AP-COP10 attains AUC 0.894 and beats all three literature sets under an identical pipeline; removing AP-COP10 from Full35 causes a supported AUC loss (Non-AP-COP25 = 0.856 vs Full35 = 0.908, delta 0.051, p=0.0028), while adding the 25 descriptors to AP-COP10 yields only +0.013 (p=0.361). The claims are appropriately scoped — the authors repeatedly note that LOSO within GaitPDB is not external validation (Sec 5.5, 5.7) and that the Full35-AP-COP10 null contrast is not evidence of equivalence (Sec 5.2).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is AP-COP10 in this Parkinson's gait study?
AP-COP10 is a set of ten features summarizing where the anterior-posterior center of pressure sits under each foot (position) and how the two feet differ (bilateral asymmetry) at five fixed windows of normalized stance (5-15% through 85-95%), computed from insole vertical ground reaction force signals.
How accurate is this method for detecting Parkinson's disease from gait?
On the public GaitPDB database (165 participants) under strict nested cross-validation, AP-COP10 reached AUC 0.894 for PD versus control; a larger 35-feature engineered bank reached 0.908. These are research benchmarks on one dataset, not clinical diagnostic performance.
Was the model validated on external data?
No. All participants come from one database (PhysioNet GaitPDB) and one sensor platform. The leave-one-source-study-out analysis (pooled AUC 0.907) tests robustness to acquisition protocol only, and the authors explicitly state it is not external validation.
Is the code for this study available?
The underlying data are public via PhysioNet, but the analysis notebooks and participant-level predictions are stated to be available from the authors only 'upon reasonable request'; nothing is publicly deposited.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures (1-6, S1) are not visually analyzable from the extracted text; figure-derived values cited here come from captions and body text.
- Supplementary tables S1-S13 were embedded and appear complete, but any supplementary content beyond them could not be verified.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.