Mahdi Saeedi

Date of Award

January 2023

Document Type


Degree Name

Master of Science (MS)


Biomedical Engineering

First Advisor

Kouhyar Tavakolian


In the forefront of biomedical engineering, timely detection and treatment of diabetic foot ulcers (DFUs) are pivotal for enhancing patient outcomes. This study explores the advancements of machine learning, adopting a federated learning framework to augment the precision and privacy of medical image analysis for DFU identification. Federated learning, a decentralized machine learning approach, proves essential, enabling the collaborative training of convolutional neural networks (CNNs) across a network of healthcare providers while upholding the sanctity of patient data confidentiality. The study employs the U-Net architecture, acclaimed for its adeptness in biomedical image segmentation. It showcases that through federated learning, the model attains high Dice and Intersection over Union (IoU) scores, mirroring those from centralized training benchmarks, thereby maintaining model efficacy without compromising data privacy.A salient aspect of this research is the use of sophisticated data transformation methods, particularly leveraging Albumentations, to standardize images from diverse imaging devices. This ensures uniformity in the data, addressing the common challenge of image standardization due to varying camera specifications. By normalizing the images, the model's ability to generalize across different data sources is significantly enhanced, making the federated learning approach even more robust. In the realm of quantifiable outcomes, our federated learning model demonstrated exceptional performance. Specifically, the U-Net architecture, under the federated paradigm, achieved a Dice coefficient of 0.9 and an Intersection over Union (IoU) score of 0.8, benchmarks that not only meet but, in some cases, surpass centralized training frameworks. These results underscore the viability of federated learning as an effective strategy for medical image analysis in scenarios where data privacy is paramount. Our research uniquely focuses on conducting experiments with both uniform and unbalanced data distributions in the federated training context. These experiments provide valuable insights into the adaptability and efficiency of our federated learning model in diverse data scenarios. Lastly, to foster further research and innovation in this field, we are providing the research community with comprehensive, well-documented source code. This initiative aims to facilitate continued advancements and exploration in the realm of federated learning, particularly in applications where data privacy is of utmost importance.