Federated and Centralized Machine Learning for Cell Segmentation: A Comparative Analysis DOI Open Access

Sara Bruschi,

Marco Esposito, Sara Raggiunto

et al.

Electronics, Journal Year: 2025, Volume and Issue: 14(7), P. 1254 - 1254

Published: March 22, 2025

The automatic segmentation of cell images plays a critical role in medicine and biology, as it enables faster more accurate analysis diagnosis. Traditional machine learning faces challenges since requires transferring sensitive data from laboratories to the cloud, with possible risks limitations due patients’ privacy, data-sharing regulations, or laboratory privacy guidelines. Federated addresses issues by introducing decentralized approach that removes need for laboratories’ sharing. task is divided among participating clients, each training global model situated on cloud its local dataset. This guarantees only transmitting updated weights cloud. In this study, centralized compared federated one, demonstrating they achieve similar performances. Stemming benchmarking available models, Cellpose, having shown better recall precision (F1=0.84) than U-Net (F1=0.50) StarDist (F1=0.12), was used baseline testbench implementation. results show both binary multi-class metrics remain high when employing solution (F1=0.86) (F12clients=0.86). These were also stable across an increasing number clients reduced samples (F14clients=0.87, F116clients=0.86), proving effectiveness central aggregation locally trained models.

Language: Английский

Federated and Centralized Machine Learning for Cell Segmentation: A Comparative Analysis DOI Open Access

Sara Bruschi,

Marco Esposito, Sara Raggiunto

et al.

Electronics, Journal Year: 2025, Volume and Issue: 14(7), P. 1254 - 1254

Published: March 22, 2025

The automatic segmentation of cell images plays a critical role in medicine and biology, as it enables faster more accurate analysis diagnosis. Traditional machine learning faces challenges since requires transferring sensitive data from laboratories to the cloud, with possible risks limitations due patients’ privacy, data-sharing regulations, or laboratory privacy guidelines. Federated addresses issues by introducing decentralized approach that removes need for laboratories’ sharing. task is divided among participating clients, each training global model situated on cloud its local dataset. This guarantees only transmitting updated weights cloud. In this study, centralized compared federated one, demonstrating they achieve similar performances. Stemming benchmarking available models, Cellpose, having shown better recall precision (F1=0.84) than U-Net (F1=0.50) StarDist (F1=0.12), was used baseline testbench implementation. results show both binary multi-class metrics remain high when employing solution (F1=0.86) (F12clients=0.86). These were also stable across an increasing number clients reduced samples (F14clients=0.87, F116clients=0.86), proving effectiveness central aggregation locally trained models.

Language: Английский

Citations

0