Internal Data Inconsistencies in Reported Performance Metrics for CSAM-ResUNet (Anal. Chem., DOI: 10.1021/acs.analchem.5c04049)

Academic fraud report · Geng Detector

Summary

This report flags serious internal inconsistencies in the reported performance improvements of the proposed CSAM-ResUNet model in a 2026 Analytical Chemistry paper on Raman spectral reconstruction of mixed microplastics (Weixiang Huang et al., DOI: 10.1021/acs.analchem.5c04049). The abstract states 35% PSNR and 80% SSIM gains; the Results section reports 126%/35%/83%/9% gains in SNR*, PSNR, SSIM and MRF respectively; the Conclusion collapses both PSNR and SSIM to a single 35% figure. Recalculation from Table 2 yields approximately 35.8% PSNR and 83.06% SSIM improvements, consistent with the Results text but not with the abstract (SSIM) or Conclusion (both metrics). A secondary concern is the inappropriate use of "enhancement" to describe a 9.1% reduction in the error metric MRF (0.1411 → 0.1282). Verdict: Suspect. No image-based manipulation was assessable from the supplied PDF. The findings are limited to textual/numerical inconsistencies; final judgment of misconduct requires institutional investigation.

Verdict

Suspect (🟡). The paper contains material internal contradictions in its headline performance metrics across abstract, results, and conclusion, plus a misuse of "enhancement" for an error metric. Findings are confined to textual and numerical inconsistencies; no image-manipulation evidence was assessable.

Key findings

Conflicting performance figures across sections for the same CSAM-ResUNet model.
Abstract claim of "35% PSNR, 80% SSIM" does not match Results ("35% PSNR, 83% SSIM") nor Conclusion ("35% PSNR and 35% SSIM").
Conclusion's statement that the method "provides over 35% improvement in PSNR and SSIM" is arithmetically incorrect for SSIM based on the paper's own Table 2.
Use of "enhancement" to describe a 9.1% decrease in the error metric MRF (0.1411 → 0.1282) is semantically inappropriate.

Evidence highlights

Reported values (Table 2): WTD+AirPLS baseline PSNR = 24.2648, SSIM = 0.4372, MRF = 0.1411. CSAM-ResUNet PSNR = 32.9637, SSIM = 0.8004, MRF = 0.1282.
Recalculated improvements: PSNR ≈ +35.8% (matches abstract/Results/Conclusion 35%); SSIM ≈ +83.06% (matches Results' 83%, conflicts with abstract's 80% and Conclusion's 35%); MRF ≈ −9.1%.
Cross-section discrepancy table:
Abstract: PSNR 35%, SSIM 80%
Results: SNR* 126%, PSNR 35%, SSIM 83%, MRF 9%
Conclusion: PSNR 35%, SSIM 35%
DOI of source: 10.1021/acs.analchem.5c04049 (Anal. Chem., 2026, Weixiang Huang, Jiajin Chen, Hao Xiong, Ligang Shao, Guishi Wang, Kun Liu, Chilai Chen, Xiaoming Gao).

Notes

No figures or image data were available for pixel-level duplication or splicing analysis; the report does not assert any image manipulation.
The discrepancies are consistent with careless copy-paste or sloppy drafting rather than deliberate fabrication of underlying experimental data, but the inability to verify which figures were intended leaves ambiguity.
Confidence in the numerical inconsistency finding is high (reproducible from Table 2). Confidence in the SSIM value in the abstract vs. results being an intentional misrepresentation rather than a typo is low.
An erratum clarifying the abstract and conclusion figures, and replacing "enhancement" with "reduction" for MRF, would resolve the most serious concerns.
Final determination of academic misconduct requires an institutional or publisher investigation.