A scalable hierarchical federated learning approach using dense cross-fusion bridge network for glaucoma detection

2026-08-11

PRE

Saida Sarra Boudouh *

Tahar Bendouma

Maroua Cheknane

DOI：10.1007/s10586-026-06446-6

Original

Original request for help

Save

Abstract

En 中文

Glaucoma is one of the leading causes of irreversible blindness worldwide. Recent advances in deep learning have significantly improved the accuracy and efficiency of automated glaucoma detection systems. However, developing robust models typically requires large and diverse datasets, which are often distributed across multiple medical institutions and restricted by privacy regulations. Although traditional federated learning enables collaborative training without sharing raw data, it still faces challenges related to data heterogeneity and scalability. Hierarchical learning strategies have shown promise in addressing these limitations in distributed environments, yet their application to glaucoma detection remains limited. To address these challenges, we propose the Dense Cross-Fusion Bridge Network for Glaucoma Detection (DenseX-BridgeNet) within a hierarchical horizontal federated learning framework. The proposed architecture introduces an intermediate regional aggregation layer that improves scalability, reduces communication overhead, and minimizes direct interactions with the cloud server. In addition, a structured feature fusion strategy is employed to selectively integrate features from complementary backbone networks, reducing redundancy while maintaining discriminative information. The framework is evaluated using multiple glaucoma datasets representing diverse demographic and regional variations. Experimental results demonstrate consistent improvements in detection accuracy across participating clients, with performance increasing from 91.25% to 96.91%, 68.03% to 85.58%, and 66.09% to 75.29%, respectively.

Keywords:

Glaucoma detection

Hierarchical federated learning

Feature extraction

DenseX-BridgeNet

Transfer learning

A scalable hierarchical federated learning approach using dense cross-fusion bridge network for glaucoma detection

Abstract

Journal

Organization

Cited Papers

Citing Papers