PLOS One has retracted the article titled “Hybrid deep learning and feature selection approach for autism detection from rs‑fMRI data.” The retraction was issued by The PLOS One Editors on July 27, 2026, and cites multiple concerns that call into question the reliability of the reported results and the article’s compliance with journal policies.
After publication, the Editors received concerns in several areas: compliance with the journal’s authorship policy, substantial similarity with a previously published study that was not cited or discussed, and highly similar numerical results in tables that represented different brain atlases and algorithmic approaches despite originating from the same ABIDE‑1 dataset. The Editors also noted that three references cited during review (References 26, 28 and 50 in the article) had been retracted while the manuscript was under review.
The notice highlights that Tables 1 and 2 in the retracted article show results that are highly similar to those in a prior publication (Nafisah et al., Frontiers in Artificial Intelligence, 2025) even though the two articles reportedly represent different brain atlases derived from the ABIDE‑1 dataset and different algorithmic methods. The Editors considered this notable because the similarity persisted despite the articles' differing methodological descriptions.
When confronted with these similarities, the fourth author stated that the datasets in the retracted article were mislabeled and clarified which datasets corresponded to which tables: Tables 1–3 were said to present the AAL, EZ and CC datasets, respectively. The fourth author further argued that applying different algorithms to the same dataset and atlas can yield highly similar numerical outputs and claimed that the two articles actually differ only at the 6th–8th decimal places; those small differences would not be apparent because the published tables round values to four decimal places.
The fourth author acknowledged that the prior related study should have been cited and disclosed at submission as a similar publication. This admission addresses the Editors’ concern that the earlier paper was neither cited nor discussed in the retracted article. Beyond the fourth author’s acknowledgement and explanation about dataset labeling and numerical rounding, the retraction notice reports that the underlying data files provided to the journal were not sufficient to resolve the outstanding concerns.
PLOS One evaluated the explanations and supporting files provided by the authors but found the information inadequate to fully resolve questions about dataset labeling and the reliability of the reported results. Because the supplied data were insufficient to address the Editors’ concerns, the matter remained unresolved in the view of the journal.
The Editors therefore concluded there were unresolved issues both with the reliability of the results as published and with compliance with PLOS policies, including proper disclosure of related publications and adherence to authorship rules.
In light of the unresolved concerns, The PLOS One Editors retracted the article. The notice records differing positions among the authors: MAE, AAE, MGK, AD, and IAF did not agree with the retraction. NM, SMA, IN, and MAAB either did not respond directly to the journal or could not be reached.
The retraction notice cites the original article (now marked RETRACTED): Abd Elaziz M, Mahmoud N, Ewees AA, Khattap MG, Dahou A, Alghamdi SM, et al. RETRACTED: Hybrid deep learning and feature selection approach for autism detection from rs‑fMRI data. PLoS One. 2026;21(4):e0339921. It also cites a related 2025 article: Nafisah I, Mahmoud N, Ewees AA, Khattap MG, Dahou A, Alghamdi SM, et al. Deep learning‑based feature selection for detection of autism spectrum disorder. Front Artif Intell. 2025;8:1594372.
The notice makes clear that the Editors’ decision to retract is based on unresolved questions about result reliability and policy compliance, and that the data provided by the authors did not allow the journal to satisfactorily address those questions.
Note: All facts in this summary and rewrite are drawn directly from the PLOS One retraction notice. Details not reported in that notice (for example, the content of the underlying data files beyond their insufficiency, or further methodological specifics) were not provided by the source and therefore are not included here.