Geng Academic Integrity Report – ShiftMorph (ACM MM 2024 submission): No Misconduct Detected

Academic fraud report · Geng Detector

Summary

Verdict: No academic misconduct detected. This integrity review examined the anonymous double-blind submission 'ShiftMorph: A Fast and Robust Convolutional Neural Network for 3D Deformable Medical Image Registration,' submitted to ACM MM 2024, under placeholder DOI 10.1145/nnnnnnn.nnnnnnn. The assessment covered image reuse and splicing (Figures 1–7), data fabrication and statistical anomalies (Tables 1–4), and methodological, citation, and timeline irregularities (Section 4.1 and References). No duplicate images, rotations, flips, or splicing artefacts were observed; figures consisted of network diagrams, mathematical schematics, violin/line plots, and registration visualisations consistent with normal scientific practice. Numerical results on Dice, HD95, and Folds metrics across OASIS and IXI datasets showed natural last-digit variation and coherent performance, including a plausible sharp reduction in Folds (%) from 0.7786% to 0.0231% using the SS (ShiftMorph-diff) component. The methodology is internally consistent, formulas are coherent, and cited datasets (OASIS, IXI, Lung250M-4B) are standard public benchmarks. Hardware (Intel Xeon Silver 4314, RTX 3090) is plausible, and references include 2023–2024 works compatible with an ACM MM 2024 submission timeline. Confidence is moderate; limitations include lack of raw pixel-level images and anonymised authorship, which precluded deeper forensic checks. No follow-up action is recommended.

Verdict

No academic misconduct detected. The submission is considered clear of the integrity issues examined.

Key findings

Image reuse and splicing (Figures 1–7): No evidence of duplicate images, rotations, flips, or splicing artefacts. Figures consist of network architecture diagrams, mathematical schematics, violin/line plots, and medical image registration visualisations consistent with normal scientific practice.
Data fabrication and statistical anomalies (Tables 1–4): Numerical results across OASIS and IXI datasets (Dice, HD95, Folds) display natural last-digit variation, coherent performance trends, and no signs of artificial arithmetic progressions or uniform standard deviations. Example: Folds (%) for ShiftMorph decreases plausibly from 0.7786% to 0.0231% when combined with the SS (ShiftMorph-diff) component.
Methodology (Section 4.1): Network modules (Shifted Embedding, Group Merging, etc.) are described with internally consistent logic and error-free formula derivations.
Datasets and citations: Uses standard public benchmarks (OASIS, IXI, Lung250M-4B) with valid references; citations include 2023–2024 works compatible with an ACM MM 2024 submission timeline.
Hardware and reproducibility: Reported equipment (Intel Xeon Silver 4314, RTX 3090) is plausible for a typical research lab setup.
Anonymity constraints: Authorship and affiliations are anonymised due to double-blind review; only the placeholder DOI 10.1145/nnnnnnn.nnnnnnn is available.

Evidence highlights

Figure set (Figures 1–7) shows no duplicated regions, splicing seams, or pixel-level manipulation traces consistent with PS-style tampering.
Tables 1–4 exhibit realistic variance and absence of "textbook-perfect" fits; e.g., Table 1 reports 0.7786% → 0.0231% Folds reduction with the SS variant, which is a coherent and large but explainable gain.
Section 4.1 outlines an internally consistent experimental protocol; reference list includes works up to 2024 (e.g., reference [44]), consistent with a 2024 submission.
DOI: 10.1145/nnnnnnn.nnnnnnn (placeholder, no official DOI assigned).

Notes

Limitations: Raw pixel-level images were not available for deeper forensic inspection; anonymous authorship precluded affiliation and ORCID cross-checks.
Confidence: Moderate. Findings are negative (no misconduct detected) but cannot rule out issues beyond the scope of the available evidence.
No further action recommended.