Geng Academic Fraud Detection Report: Susceptibility MRI Helps Predict Mild Cognitive Impairment Onset and Cognitive Decline in Cognitively Unimpaired Older Adults (DOI: 10.1148/radiol.250513)

Academic fraud report · Geng Detector

Summary

Verdict: Highly suspicious. The report flags substantial inconsistencies between the numerical statistical outputs and the visual representations in a 2025 Radiology paper on susceptibility-weighted MRI for predicting MCI. The primary concern is in Figure 3 (Kaplan-Meier survival curves): the curves for the entorhinal cortex, putamen, and caudate nucleus across the full sample (N=158) and PET subgroup (N=110) appear near-identical in shape, step locations, and endpoint heights, despite the Cox hazard ratios in Table 2 differing substantially (e.g., entorhinal HR=2.00 overall vs. HR=3.59 in PET subgroup; caudate HR=2.11, putamen HR=2.54 in PET subgroup). Such large HR differences should produce visibly different slopes and step drops. A secondary concern is that the confidence intervals in Figure 4 appear unnaturally uniform and symmetric across regions. Timeline and software version checks revealed no anomalies. The confidence is moderate-to-high for the Figure 3 inconsistency, as it relies on visual analysis combined with the internal numerical contradiction; Figure 4 concerns are weaker due to limited evidence. Final determination requires raw data and plotting code from the authors.

Verdict

🟠 Highly suspicious. The strongest signal is an internal contradiction between Table 2 hazard ratios and the near-identical Kaplan-Meier curves in Figure 3. Secondary geometric concerns affect Figure 4. No timeline or methodological inconsistencies were detected.

Key findings

Figure 3 vs. Table 2 contradiction (severity: 🔴, confirmed): Survival curves for different brain regions (entorhinal cortex, putamen, caudate) and cohorts (total N=158; PET subgroup N=110) appear visually congruent in step placement, shape, and terminal heights, yet the corresponding Cox HRs vary widely (entorhinal HR 2.00 vs. 3.59; caudate HR 2.11; putamen HR 2.54 in the PET subgroup).
Figure 4 confidence-interval geometry (severity: 🟠, unconfirmed): The shaded confidence bands in the linear mixed-effects prediction plot show unusually regular, symmetric shapes across regions (ENT, PT, CN), with limited irregularity at the edges that would be expected from 7.7-year longitudinal cognitive data with attrition.
Timeline and sample-flow check (severity: ✅, no anomaly): Submission (2025-02-19), acceptance (2025-07-24), software versions (MATLAB 2023b, R 4.3.0), and the sample flow (187 → 158; PET subgroup 110) are internally consistent.

Evidence highlights

Entorhinal cortex HR: 2.00 (total sample) vs. 3.59 (PET subgroup) — roughly 80% relative increase, which should produce visibly steeper decline in high-susceptibility quantile.
PET subgroup HRs reported as 2.11 (caudate) and 2.54 (putamen); these are not negligible differences and would still manifest as different curve trajectories.
Visual analysis summary states that the Kaplan-Meier trajectories for the high-susceptibility groups across regions and cohorts show "extremely high overlap" in step locations and terminal heights.
Figure 4 confidence intervals described as "highly similar geometric features" across regions, contrary to typical heterogeneous longitudinal cognitive trajectories.
No cross-source contradiction found in Materials and Methods, Table 1 demographics, or software/sampling timeline.

Notes

The discrepancy between numerical HRs and visual curves may stem from (a) use of a common plot template, (b) data fabrication, (c) post-hoc figure manipulation, or (d) a plotting error that re-used a single underlying dataset. The current evidence cannot distinguish among these.
The Figure 4 concern is flagged as "依据不足" (insufficient basis) in the original report and should be treated as supportive rather than primary evidence.
Recommended author response: provide raw R/MATLAB code generating Figures 3 and 4, plus individual-level time-to-event data and per-subject longitudinal cognitive scores.
The DOI 10.1148/radiol.250513 should be cited in any PubPeer comment or editorial communication. All numeric values above are reproduced directly from the original report; no additional figures were inferred.