Logo image
Back
Explainability Of Privacy Preservation in Healthcare Data Processing
Dissertation   Open access

Explainability Of Privacy Preservation in Healthcare Data Processing

A B M Kamrul Islam Riad
University of West Florida Libraries
Doctor of Philosophy (PHD), University of West Florida
2026

Metrics

1 Record Views

Abstract

This dissertation develops an integrated framework for explainable, privacy-preserving healthcare data processing across text and image modalities. The work combines federated learning, differential privacy, self-supervised learning, and reinforcement learning with clinician-oriented explainability methods such as SHAP and LIME. Using publicly available and synthetic datasets, the framework evaluates privacy--utility trade-offs, robustness against privacy attacks, and the stability of explanations under noise and distributed training. The results demonstrate that carefully tuned privacy controls and model architectures can preserve diagnostic performance while providing transparent, audit-ready explanations aligned with HIPAA requirements. Beyond predictive performance, the proposed framework emphasizes governance-relevant evidence generation by explicitly coupling privacy mechanisms with explanation artifacts and data lineage. The dissertation introduces methodological components that enable systematic tracking of privacy expenditure and characterization of how privacy configurations influence both model behavior and explanation faithfulness. This perspective supports compliance-oriented evaluation by framing model outputs as accountable decisions that can be interrogated with respect to minimum necessary use, access constraints, and auditability. Empirical studies across medical imaging and clinical text demonstrate that privacy-preserving training can be operationalized under realistic constraints typical of healthcare settings, including heterogeneous data distributions and restrictions on centralizing protected in the health information. Collectively, these contributions advance the state of the art toward deployable healthcare AI systems that balance utility, privacy protection, interpretability, and regulatory alignment.
pdf
Explainability Of Privacy Preservation in Healthcare Data Processing1.72 MBDownloadView
Preprint Open Access

Details

Logo image