Medical Image Classification with Privacy: Centralized and Federated Learning Comparison

• Leonardo Gabriel Ferreira Rodrigues UFU http://orcid.org/0009-0003-7246-2175
• Gleidson V. Gomes Barbosa UFU https://orcid.org/0009-0000-2159-335X
• Rodrigo Moreira UFV https://orcid.org/0000-0002-9328-8618
• Larissa Ferreira Rodrigues Moreira UFV https://orcid.org/0000-0001-8947-9182
• André Ricardo Backes UFSCar https://orcid.org/0000-0002-7486-4253

Abstract

The use of different techniques in image-based medical diagnostic computing has positively impacted the healthcare industry. Diseases such as colon and lung cancer can be detected using image classification strategies. In this context, Federated Learning (FL) has emerged as an enabling approach to disease detection with security and privacy when handling sensitive patient data. This study aimed to compare the centralized and FL classifications of histopathological images of lung and colon cancer. We compared the performance of FL with that of centralized learning, and the results indicated that FL outperformed centralized approaches in scenarios with heterogeneous datasets. These findings highlight the potential of FL as an alternative to medical image classification, particularly in privacy-sensitive applications.

Keywords: Federated Learning, Centralized Learning, Image Classification, Cancer, Deep Learning