Related Experiment Video
Updated: Jul 31, 2026
15:53
Utilizing 3D Printing Technology to Merge MRI with Histology: A Protocol for Brain Sectioning
Published on: December 6, 2016
15.0K
Experts fail to reliably detect AI-generated histological data
Jan Hartung1,2,3,4, Stefanie Reuter5, Vera Anna Kulow6
1Institute for Physiology, Faculty of Medicine, University of Freiburg, 79108, Freiburg, Germany. jan.hartung@physiologie.uni-freiburg.de.
Scientific Reports
|November 20, 2024
Summary
Artificial intelligence can now generate realistic histological images, fooling even scientific experts. This breakthrough necessitates new methods and policies to detect AI-generated data in scientific publications.
Area of Science:
- Digital pathology
- Artificial intelligence
- Image forensics
Background:
- AI image generation is rapidly advancing, posing challenges to image authenticity.
- AI-generated images can mimic complex structures like histological samples, raising concerns about data fabrication.
Purpose of the Study:
- To assess human and expert ability to distinguish AI-generated histological images from real ones.
- To evaluate the effectiveness of Stable Diffusion in creating convincing artificial histological samples.
Main Methods:
- Utilized Stable Diffusion, a recent generative algorithm, to create artificial histological samples.
- Conducted a study with over 800 participants to test discrimination between real and AI-generated images.
- Analyzed participant performance based on training data quantity.
Main Results:
- Even experts could not reliably differentiate between genuine and AI-generated histological images.
- Participant performance improved with training but remained imperfect.
- Convincing artificial images can be generated with even small amounts of training data.
Conclusions:
- Current AI image generation techniques can produce highly deceptive histological images.
- Existing human perceptual capabilities are insufficient to reliably detect AI-fabricated scientific data.
- Urgent development of detection methods and policy changes is required to ensure data integrity in scientific publications.
