MORPHFED: Federated Learning for Cross-institutional Blood Morphology Analysis

TL;DR

This work addresses the privacy and generalization challenges in automated blood morphology analysis by proposing a federated learning framework for cross-institutional training across PBS/BMA images. It evaluates four aggregation methods and two architectures (ResNet-34 and DINOv2-Small) under realistic non-IID conditions, demonstrating that federated training substantially outperforms local training and approaches centralized performance while preserving data privacy. The results reveal a trade-off between robustness to data heterogeneity (FedMedian) and preservation of rare but diagnostically crucial morphologies (FedOpt), with external validation showing improved generalization to unseen institutions. Overall, federated learning emerges as a practical, privacy-preserving approach for equitable and scalable hematology imaging AI in resource-limited settings.

Abstract

Automated blood morphology analysis can support hematological diagnostics in low- and middle-income countries (LMICs) but remains sensitive to dataset shifts from staining variability, imaging differences, and rare morphologies. Building centralized datasets to capture this diversity is often infeasible due to privacy regulations and data-sharing restrictions. We introduce a federated learning framework for white blood cell morphology analysis that enables collaborative training across institutions without exchanging training data. Using blood films from multiple clinical sites, our federated models learn robust, domain-invariant representations while preserving complete data privacy. Evaluations across convolutional and transformer-based architectures show that federated training achieves strong cross-site performance and improved generalization to unseen institutions compared to centralized training. These findings highlight federated learning as a practical and privacy-preserving approach for developing equitable, scalable, and generalizable medical imaging AI in resource-limited healthcare environments.

Figures (2)

• Figure 1: Sample cell types present in the two training datasets. Staining variation can be observed between Client 1 (first row) and Client 2 (second row) datasets
• Figure 2: (A) Federated Learning framework demonstrates privacy-preserving collaborative training where Client 1 and Client 2 perform local model training with parameter aggregation at a central server (B) Centralized Training paradigm with full access to combined dataset using 4-fold cross-validation