19 Nov 2024
10:00 Doctoral defense Room 85 of IC2
Theme
Strengthening Scientific Integrity: Digital Forensics for Biomedical Research Images
Student
João Phillipe Cardenuto
Advisor / Teacher
Anderson de Rezende Rocha - Co-advisor: Daniel Henriques Moreira
Brief summary
"Science has reported an increase in cases of misconduct, especially in the area of biomedicine. The scenario became critical when social networks began to use fraudulent articles to promote disinformation content. Despite all previous efforts to detect textual fraud (such as plagiarism detection), there is a lack of studies that address scientific misconduct in images. In addition, there is a worrying trend of problematic images being published in scientific articles. Image misconduct is evolving from simple inappropriate edits and reuses to systematic manipulations involving potentially illegal organizations known as “paper mills”. Furthermore, with the advancement of generative artificial intelligence (AI), it is expected that, sooner or later, such organizations will begin to completely synthesize scientific images - and even the entire article -, polluting the literature with false data. In turn, the scientific community has reacted slowly, and few reliable and effective solutions have been proposed, even for the detection of unsophisticated image manipulation. Motivated by this intriguing scenario, this research investigates scientific integrity using a forensic framework. Due to the lack of previous forensic research on this topic, we organize and point out the main computational challenges related to scientific image analysis, which we work to mitigate. As a result, we propose methods for automated image analysis, document and image provenance analysis, and AI-generated image detectors focused on the biomedical field to assist integrity offices and journal editorials during the decision-making process. The proposed methods accelerate integrity analysis using an end-to-end workflow that receives a collection of PDF documents (e.g., a set of scientific articles) and highlights suspicious images and documents that deserve further attention. Following the recommendation of integrity offices, the proposed solutions use explainable steps with results that are easy for analysts to interpret. In addition to the forensic methods, datasets and benchmarks created -- all open source and freely available -- this work also discusses the standard of biomedical images as scientific evidence and points out its main problems, aiming to promote possible requirements for the publication of reliable scientific images. This thesis is the result of a collaboration with scientific integrity researchers and an international team of digital forensics researchers to mitigate current threats to science and foster effective computational methods to improve its integrity.'
Examination Board
Headlines:
| Anderson de Rezende Rocha | IC / UNICAMP |
| Renan Moritz Varnier Rodrigues de Almeida | COPPE/UFRJ |
| André Carlos Ponce de Leon Ferreira de Carvalho | ICMC / USP |
| Nina Sumiko Tomita Hirata | IME / USP |
| Sérgio Luiz Monteiro Salles Filho | IG/UNICAMP |
Substitutes:
| Hélio Pedrini | IC / UNICAMP |
| Sonia Maria Ramos de Vasconcelos | IBqM/UFRJ |
| João Paulo Papa | FC / UNESP |