ImageTwin: Scientific Image Integrity Guide

Imagetwin is an artificial intelligence-powered software platform designed to detect figure duplication, image manipulation, plagiarism, and AI-generated content in scientific manuscripts. Developed to safeguard research integrity across biomedical and physical sciences, the software compares scientific figures against internal panels within the manuscript and an indexed database of over 150 million published scientific figures. By flagging anomalous patterns in gel electrophoresis, Western blots, microscopy, and histology before publication, Imagetwin helps editorial offices, research integrity officers, and investigators identify errors and prevent research misconduct.

In biomedical research and academic publishing, image integrity is a key element of pre-publication quality control. Automated forensic screening platforms provide systematic oversight where manual visual inspection cannot scale across large submission volumes, complementing broader research workflows such as detecting image manipulation in figures and institutional accountability for AI-generated figures.

What it is

Imagetwin is a cloud-based research integrity platform engineered specifically for the analysis of scientific imagery. The platform accepts manuscript files in PDF format or discrete image files, extracts visual figures and sub-panels, and applies machine learning algorithms to verify the authenticity and provenance of visual data. Rather than functioning as a generic reverse-image search engine, Imagetwin operates neural networks trained on scientific figure typologies, with particular sensitivity to halftone biological imagery including Western blots, Southern blots, fluorescence microscopy, and cell migration assays.

The platform evaluates figures across four primary integrity domains:

Internal Duplication: Identifies panels, bands, or micrographs that appear multiple times within the same manuscript, including instances where images have been cropped, rotated, flipped horizontally or vertically, scaled, or adjusted for contrast.
External Duplication and Plagiarism: Matches submitted figure panels against an indexed repository of more than 150 million published figures sourced from open-access literature and publisher partnerships, identifying figures published previously without attribution.
Image Manipulation: Detects alterations within individual figure panels, such as splicing artifacts, copy-move duplications (cloning background or target elements), differential filtering, and unauthorized insertions.
AI-Generated Content Detection: Analyzes visual feature distributions and generative artifact signatures to identify synthetic biomedical figures produced by diffusion models, generative adversarial networks (GANs), or other synthetic imaging tools.

Who runs it and what it covers

Imagetwin is developed and operated by ImageTwin AI GmbH, a research-technology company headquartered in Vienna, Austria. The technology originated in 2020 as an academic research initiative at TU Wien (Vienna University of Technology), led by co-founder Dr. Markus Zlabinger. Following incubation at TU Wien and input from integrity experts, the company was officially incorporated in 2021 by Patrick Starke and Markus Zlabinger. The platform is used across scientific publishers, academic universities, and research institutes internationally.

The operational scope of Imagetwin focuses on scientific research publications, with figure coverage governed by technical image categories:

Halftone Biological and Medical Imagery: Core detection models evaluate microscopy (light, confocal, electron), immunoblotting and gel assays (Western, Northern, Southern), flow cytometry plots, immunohistochemistry tissue sections, and laboratory photography.
Excluded Figure Formats: The software does not evaluate non-halftone visual data, such as schematic line diagrams, vector charts, scatter plots, bar graphs, mathematical models, or textual tables.
Reference Corpus Scale: The comparative external database encompasses over 150 million indexed figure panels from published scientific literature, updated continuously as new research appears.

Major scientific societies have integrated Imagetwin into routine editorial operations. For example, the American Society for Microbiology (ASM) deployed Imagetwin across its portfolio of 17 peer-reviewed journals, screening accepted manuscripts prior to publication to resolve unintentional figure duplications with authors and maintain strict evidentiary standards in microbial and biomedical research, alongside standards in biomedical engineering and federally mandated NIH data management and sharing plans.

How to search or use it

Imagetwin provides a web-based dashboard and an Application Programming Interface (API) for editorial management systems and institutional audit pipelines. Operating the web application follows a structured forensic workflow:

Access and Authentication: Log into the secure Imagetwin web application using authorized user credentials.
Upload Manuscript or Image Assets: Submit a complete manuscript file in PDF format, or upload individual figure panels in standard raster formats (PNG, JPEG, TIFF). A single scan accommodates one PDF document or up to 25 individual image files, with the system deconstructing composite figures into sub-images.
Select Screening Parameters: Designate whether the material should be screened exclusively against itself (internal duplication), against the global 150M+ external database, or within a private repository for confidential unpublished work.
Execute AI-Powered Integrity Scan: Initiate the automated scan. The platform processes figure panels through its computer vision pipeline, extracting feature descriptors and evaluating similarity metrics.
Review Flagged Findings: Open the interactive report. Flagged items display matched image pairs, highlighted regions of interest, and assigned confidence scores that quantify the detected match.
Inspect Discrepancies in the Forensic Toolbox: Use built-in visual inspection tools—including matched keypoint overlays, side-by-side comparative views, false-color filters, and zoom controls—to verify whether flagged duplications represent legitimate experimental controls or improper manipulations.
Export and Share Audit Documentation: Generate and export an audit report documenting verified findings, keypoint correlations, and match metadata for editorial deliberation or institutional records.

Strengths and limits

Evaluating automated figure screening software requires an understanding of both its technical advantages and its operational boundaries within academic research integrity workflows.

Strengths:

Cross-Publication Discovery: Unlike manual inspection, which is generally confined to panels within a single article, Imagetwin queries a massive multi-publisher database to surface inter-manuscript image reuse and cross-journal figure recycling.
Robustness to Geometric Transformations: Detection algorithms recognize manipulated images even after alterations such as cropping, aspect-ratio stretching, rotational shifts, horizontal or vertical reflections, and brightness adjustments.
Workflow Efficiency: Screening an entire multi-figure manuscript takes minutes, reducing editorial turnaround time compared to manual visual spot-checking and catching issues before publication.
Private Repositories: Institutions and publishers can maintain segregated reference repositories to check unreleased manuscripts without exposing pre-publication data to third parties.

Limits:

Halftone Specificity: The system cannot detect manipulation, duplication, or fabrication in vector graphics, statistical charts, numerical spreadsheets, or narrative text.
Human Adjudication Requirement: Algorithmic matches generate confidence scores and visual overlays, but software cannot determine intent. Legitimate biological controls, re-probed blot membranes, or authorized re-use across related companion studies require expert human review.
Proprietary Corpus Boundaries: While the index contains over 150 million figures, it cannot encompass all paywalled literature, non-indexed historical archives, or figures in closed databases.

Alternatives

Research institutions, publishers, and authors utilize several tools to screen figures and investigate image integrity concerns. The following table outlines primary tools across automated and manual inspection workflows:

Tool / Platform	Primary Mechanism	Database Comparison	Target Image Types	Operational Model
Imagetwin	Automated AI detection of duplication, manipulation, and AI-generated content	Indexed database of 150M+ scientific figures; private repository option	Halftone scientific figures (microscopy, blots, gels, histology)	Commercial cloud web application and API
Proofig AI	Automated computer vision screening for figure duplication and manipulation	Internal manuscript comparison and cross-manuscript database	Scientific research figures, blots, microscopy panels	Commercial SaaS platform for publishers and institutions
ImageJ / Fiji	Manual image processing, channel subtraction, color inversion, and histogram analysis	None (single-file or paired-file manual comparison only)	Scientific raster and multi-channel images	Free and open-source desktop software (NIH-maintained)
Forensically	Manual digital image forensics (Error Level Analysis, clone detection, noise analysis)	None (local browser-based inspection only)	General photographic and scientific raster images	Free web-based forensic toolkit

FAQ

Does Imagetwin replace human peer reviewers and editors?

No. Imagetwin is a diagnostic screening tool that highlights visual similarities, potential manipulations, and confidence scores. Editorial decisions, author correspondence, and formal integrity assessments require qualitative evaluation by editors, peer reviewers, or research integrity officers to distinguish between valid experimental controls and improper alterations.

Can Imagetwin detect AI-generated scientific images?

Yes. Imagetwin incorporates neural networks trained to identify synthetic and AI-generated image artifacts in scientific figures, helping editorial offices detect synthetic micrographs and artificially fabricated biological data.

What types of scientific figures does Imagetwin screen?

Imagetwin screens halftone scientific imagery, specifically Western blots, Southern and Northern blots, gel electrophoresis, light and confocal microscopy, immunohistochemistry, and laboratory specimen photography. It does not screen line graphs, flowcharts, vector diagrams, or tabular data.

Can an individual researcher use Imagetwin before submitting a paper?

Yes. Imagetwin offers accounts for individual researchers and research groups to audit manuscript figures prior to submission, verifying that image panels have not been inadvertently duplicated or mismatched during figure assembly.

Sources

ImageTwin AI GmbH, Official Product Documentation, Platform Features, and Company Overview: https://imagetwin.ai/ and https://imagetwin.ai/about (accessed 2026-11-07).
ImageTwin AI GmbH, Terms of Service, Legal Scope, and Scan Specifications: https://imagetwin.ai/terms and https://imagetwin.ai/pricing (accessed 2026-11-07).
American Society for Microbiology (ASM), Case Study on ImageTwin Workflow Integration and Editorial Deployment: https://imagetwin.ai/ and mBio publication records (accessed 2026-11-07).
Investigative Ophthalmology & Visual Science (IOVS), Editorial Implementation of Imagetwin: DOI 10.1167/iovs.67.13.13 (accessed 2026-11-07).