BioFors: A Large Biomedical Image Forensics Dataset

Sabir, Ekraam; Nandi, Soumyaroop; AbdAlmageed, Wael; Natarajan, Prem

doi:10.1109/iccv48922.2021.01078

Cited by 7 publications

(11 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The images represent one of the retraction reasons, which has been growing in recent years, the increased interest in the subject highlighting its unexpected magnitude but also the development of tools that facilitate the detection of image manipulation in scientific articles (Bik et al 2016; Koppers et al 2017; Bucci 2018; Bik et al 2018; Christopher 2018; Sabir et al 2021).…”

Section: Discussionmentioning

confidence: 99%

Correction of the scientific production: publisher performance evaluation using a dataset of 4844 PubMed retractions

Toma¹,

Padureanu

Toma

2022

Preprint

View full text Add to dashboard Cite

Background: Withdrawal of problematic scientific articles after publication is one of the mechanisms for correcting the literature available to publishers, especially in the conditions of the ever-increasing trend of publishing activity in the medical field. The market volume and the business model justify publishers' involvement in the post-publication quality control (QC) of scientific production. The limited information about this subject determined us to analyze retractions and the main retraction reasons for publishers with many withdrawn articles. We also propose a score to measure the evolution of their performance. The data set used for this article consists of 4844 PubMed retracted papers published between 1.01.2009 and 31.12.2020. Methods: We have analyzed the retraction notes and retraction reasons, grouping them by publisher. To evaluate performance, we formulated an SDTP score whose calculation formula includes several parameters: speed (article exposure time (ET)), detection rate (percentage of articles whose retraction is initiated by the editor/publisher/institution without the authors' participation), transparency (percentage of retracted articles available online and clarity of retraction notes), precision (mention of authors' responsibility and percentage of retractions for reasons other than editorial errors). Results: The 4844 withdrawn articles were published in 1767 journals by 366 publishers, the average number of withdrawn articles/journal being 2.74. Forty-five publishers have more than ten withdrawn articles, holding 88% of all papers and 79% of journals. Combining our data with data from another study shows that less than 7% of PubMed journals withdrew at least one article. Only 10.5% of the withdrawal notes included the individual responsibility of the authors. Nine of the top 11 publishers had the largest number of articles withdrawn in 2020, in the first 11 places finding, as expected, some big publishers. Retraction reasons analysis shows considerable differences between publishers concerning the articles ET: median values between 9 and 43 months (mistakes), 9 and 73 months (images), 10 and 42 months (plagiarism & overlap). The SDTP score shows, between 2018 and 2020, an improvement in QC of four publishers in the top 11 and a decrease in the gap between 1st and 11th place. The group of the other 355 publishers also has a positive evolution of the SDTP score. Conclusions: Publishers have to get involved actively and measurably in the post-publication evaluation of scientific products. The introduction of reporting standards for retraction notes and replicable indicators for quantifying publishing QC can help increase the overall quality of scientific literature.

show abstract

Section: Discussionmentioning

confidence: 99%

Correction of the scientific production: publisher performance evaluation using a dataset of 4844 PubMed retractions

Toma¹,

Padureanu

Toma

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Recently, a new biomedical image forensics dataset (Bio-Fors) [13] was released to promote the development of automated detection methods. The dataset comprises images belonging to four categories collected from biomedical research documents.…”

Section: Introductionmentioning

confidence: 99%

“…The paper also proposed three biomedical forensic tasks to overcome the lack of structured problem definitions in literature. There are three manipulation detection tasks described in [13] -external duplication detection (EDD), internal duplication detection (IDD) and cut/sharp transition detection (CSTD). These tasks together cover popular forms of semantic and digital manipulations found in biomedical literature.…”

Section: Introductionmentioning

confidence: 99%

“…the images could be spliced, cropped with an overlap from a larger image or simply reused. Figure 1 shows a manipulated sample under the EDD task from [13].…”

Section: Introductionmentioning

confidence: 99%

“…Related research areas of image matching and splicing detection are matured. However, as shown in [13], traditional computer vision methods trained on natural images are not suitable for biomedical forensics. Difficulties in detecting keypoints from biomedical images limit keypoint-descriptor based methods [14,15,16].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

MONet: Multi-scale Overlap Network for Duplication Detection in Biomedical Images

Sabir¹,

Nandi²,

AbdAlmageed³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Manipulation of biomedical images to misrepresent experimental results has plagued the biomedical community for a while. Recent interest in the problem led to the curation of a dataset and associated tasks to promote the development of biomedical forensic methods. Of these, the largest manipulation detection task focuses on the detection of duplicated regions between images. Traditional computer-vision based forensic models trained on natural images are not designed to overcome the challenges presented by biomedical images. We propose a multi-scale overlap detection model to detect duplicated image regions. Our model is structured to find duplication hierarchically, so as to reduce the number of patch operations. It achieves state-of-the-art performance overall and on multiple biomedical image categories.

show abstract