A federated analytics framework for interpretable subgroup discovery and rule-based decision support from medical image data

Subgroup discovery is well suited for identifying targeted and interpretable patterns in medical data. In this work, we present a Federated Subgroup Discovery framework FedSD for analysing heterogeneous medical datasets in a multi hospital setting while preserving data locality. Clinically meaningful features extracted from fundus images are represented as a symbolic transactional representation, and subgroup discovery is performed locally at each hospital. Only subgroup summaries are shared for server side aggregation, enabling collaborative analysis without exchanging raw data or instance level information. We evaluate our approach on multiple publicly available glaucoma datasets that are widely used in prior studies. The results show that the proposed framework captures meaningful hospital specific variations while also revealing globally stable rules through federated aggregation. The analysis highlights a clear distinction between locally strong and globally stable rules, as well as a non trivial relationship between subgroup coverage and quality in imbalanced medical data. Additional results show that the discovered subgroups are stable across data partitions and that the federated approach can be run without excessive computational overhead. These experimental evaluations indicate that Federated Subgroup Discovery provides an interpretable and privacy preserving solution for cross institutional medical data analysis. • Develop a federated analytics method for subgroup discovery across hospitals. • Extract symbolic representations from medical image features for rule mining. • Identify local and global patterns to support clinical decision insights. • Preserve privacy by sharing subgroup summaries instead of raw data. • Demonstrate interpretable rule discovery in multi-hospital image datasets.