Post by Sandy Engelhardt

🔥 The latest #Nature #Biomedical #Engineering Publication (IF 26.6) from our lab: Patient data memorization in #unconditional #latent #diffusion #models. 💡 The study tackles a key question for #generative #AI in 2D and 3D #medical #imaging: How much do these models “remember” from their training data? How similar is the output to the input? 🔍 Our results highlight why #privacy #checks for synthetic medical data are absolutely essential and we make a key contribution in terms of assessment of #image #similarity. Full paper (view only:) https://lnkd.in/dC5x2jcG 🫶 Incredible team effort: Salman ul Hassan Dar, Marvin Seyfarth, PD Dr. med. Isabelle Ayx, Theano Papavassiliu, Stefan O. Schoenberg, Robert Malte Siepmann, Fabian Christopher Laqua, Jannik Kahmann, Norbert Frey, Prof. Dr. med. Bettina Baeßler, Sebastian Foersch, Daniel Truhn, Jakob Nikolas Kather, Sandy Engelhardt This was a tremendous effort, more than 60 pages of rebuttal. Thanks to the entire team, especially Salman ul Hassan Dar and Marvin Seyfarth for their incredible dedication to the project. 🦾 Lucky to call these institutions our contributors: Universitätsklinikum Heidelberg (UKHD) Universität Heidelberg Universitätsmedizin Mannheim Universitätsmedizin Mainz Uniklinik RWTH Aachen Universitätsklinikum Würzburg Technische Universität Dresden Else Kröner Fresenius Center for Digital Health 🙏 Funding: Carl-Zeiss-Stiftung Bundesministerium für Forschung, Technologie und Raumfahrt Health + Life Science Alliance 👉 Happy to discuss further directions in the field at this years DGM4MICCAI Workshop in South Korea. #Deep #Generative #Models