Perumal, M.; Srinivas, M.

Medical data is not available for public access due to privacy concerns of the patients and the stakeholders trustworthiness. However, Artificial Intelligence, especially all deeplearning models, is data hungry and fails to produce clinically relevant results without much data. Moreover, augmentation strategies are deployed to overcome the less data hurdle. The promising future in this direction is generative AI augmented data. The chatGPT and DALLE2 have become commercial products leveraging the generative AI in Natural Language Processing and Computer Vision. The diffusion models have started giving many promising results in the generative AI in computer vision. And in medical imaging, they can potentially create synthetic data to augment the scarce dataset. Diffusion models coupled with federated learning can create synthetic data on a large scale without the need to violate data privacy. This synthetic dataset could be used for further training of deeplearning models without the issues of patients identity theft from reverse engineering data. A Federated Learning paradigm of diffusion models has been proposed to overcome this hurdle and its related challenges. Our work focuses on diffusion models in federated learning settings. We named our novel model FMedDiffusion or Federated Learning on Medical Image Diffusion. We trained our model under distributed federated settings imitating real world clinical settings. We have achieved impressive results over three medical image datasets, APTOS 2019 Blindness Detection, Retinal OCT Detection, and COVID CT Detection in the federated setting on par with the traditional training. Our model FMedDiffusion has achieved an FID score of 7.1821 on the APTOS 2019 Blindness Detection dataset, an FID score of 8.8154 on the Retinal OCT Detection dataset and an FID score of 7.4486 on the COVID CT Detection dataset.reverse engineering data. A Federated Learning paradigm of diffusion models has been proposed to overcome this hurdle and its related challenges. Our work focuses on diffusion models in federated learning settings. We named our novel model FMedDiffusion or Federated Learning on Medical Image Diffusion. We trained our model under distributed federated settings imitating real world clinical settings. We have achieved impressive results over three medical image datasets, APTOS 2019 Blindness Detection, Retinal OCT Detection, and COVID CT Detection in the federated setting on par with the traditional training. Our model FMedDiffusion has achieved an FID score of 7.1821 on the APTOS 2019 Blindness Detection dataset, an FID score of 8.8154 on the Retinal OCT Detection dataset, and an FID score of 7.4486 on the COVID CT Detection dataset.reverse engineering data. A Federated Learning paradigm of diffusion models has been proposed to overcome this hurdle and its related challenges. Our work focuses on diffusion models in federated learning settings. We named our novel model FMedDiffusion or Federated Learning on Medical Image Diffusion. We trained our model under distributed federated settings imitating real world clinical settings. We have achieved impressive results over three medical image datasets, APTOS 2019 Blindness Detection, Retinal OCT Detection, and COVID CT Detection in the federated setting on par with the traditional training. Our model FMedDiffusion has achieved an FID score of 7.1821 on the APTOS 2019 Blindness Detection dataset, an FID score of 8.8154 on the Retinal OCT Detection dataset, and an FID score of 7.4486 on the COVID CT Detection dataset.