Figure plagiarism and manipulation, an under-recognised problem in academia.

Abstract

Academic plagiarism undermines the integrity of scientific research. While text-based plagiarism detection tools are widely used, the rise of artificial intelligence (AI) has introduced new challenges, particularly in text and image generation and manipulation. We briefly discuss the evolving landscape of plagiarism and the innovations that have come about with the proliferation of AI, focusing on the implications for text and image manipulation in academic writing and research. We discuss some of the current tools and practices used to detect AI-generated and manipulated text and images, including plagiarism detection software, computer vision algorithms, and manual reverse image searches. AI can enhance manuscript readability but also facilitates plagiarism and bias reinforcement due to the material it is trained on. Text-based detection tools are adapting to AI-generated content, yet image-based detection lags behind. Though tools to detect AI manipulation show promise, they are not perfect, particularly for manipulated images. Simple reverse image searches are a promising tool and can sometimes identify plagiarized figures that have undergone limited manipulation, but human oversight is often necessary. We believe that integrating image fabrication, manipulation and plagiarism detection into standard fraud detection packages is essential to uphold academic integrity in the new world of AI. Enhanced vigilance and technology are critical, particularly in fields like medical imaging, where image authenticity directly impacts research and thus clinical outcomes. KEY POINTS: We discuss the problems related to the rise of AI with regard to image manipulation in academic work, and how radiology is particularly at risk. We shed light on the rarely and little discussed topic of AI image manipulation and outright fraud. We hope to incite further discussion and adoption of image fraud prevention software. We discuss the use of some tools which are gradually becoming adopted and how some journals have begun to screen for image manipulation and fraud. We suggest an easy technique of using reverse image search that can sometimes be extremely useful despite its simplicity and can be easily adapted into researchers' practice.