Bridging the Interoperability Gap in Healthcare AI: Adaptive Federated Learning for Secure, Cross-Platform Data Harmonization

The proliferation of real-time applications has motivated extensive research on analyzing and optimizing data freshness in the context of age of information. However, classical frameworks of privacy (e.g., differential privacy (DP)) have overlooked the impact of data freshness on privacy guarantees, and hence may lead to unnecessary accuracy loss when trying to achieve meaningful privacy guarantees in time-varying databases. In this work, we introduce age-dependent DP, taking into account the underlying stochastic nature of a time-varying database. In this new framework, we establish a connection between classical DP and age-dependent DP, based on which we characterize the impact of data staleness and temporal correlation on privacy guarantees. Our characterization demonstrates that aging, i.e., using stale data inputs and/or postponing the release of outputs, can be a new strategy to protect data privacy in addition to noise injection in the traditional DP framework. Furthermore, to generalize our results to a multi-query scenario, we present a sequential composition result for age-dependent DP. We then characterize and achieve the optimal tradeoffs between privacy risk and utility. Finally, case studies show that, when achieving a target of an arbitrarily small privacy risk in a single-query case, the approach of combining aging and noise injection can achieve a bounded accuracy loss, whereas using noise injection only (as in the DP benchmark) will lead to an unbounded accuracy loss.

The healthcare industry benefits from privacy-preserving machine learning (PPML) because it provides a groundbreaking solution to protect patient privacy while enabling complex data analysis. The increasing adoption of artificial intelligence (AI) and machine learning (ML) in healthcare requires absolute protection of sensitive patient information. The model training process in PPML utilizes federated learning, combined with differential privacy, homomorphic encryption, and secure multi-party computation, to enable collaborative model training without exposing raw data. The methods keep data on local servers, which reduces the chance of breaches while supporting analytical performance. Homomorphic encryption enables secure computations on encrypted data to protect confidential medical information, and differential privacy adds noise to protect individual identities. The current PPML solutions present significant technical obstacles, including inefficient computation, restricted scalability, and integration complications with healthcare IT systems. The healthcare sector requires the ability to process extensive real-time medical information while supporting system interoperability. The healthcare industry is investigating blockchain technology to enhance data security through immutable records that provide verifiable evidence. The healthcare industry should adopt PPML as its essential framework for conducting secure and ethical data analysis, as it incorporates artificial intelligence. This chapter examines the evolving PPML healthcare environment by discussing its methods, along with their advantages and the essential obstacles that need to be resolved to support its clinical and research implementation.

Healthcare system: Moving forward with artificial intelligence

The integration of Artificial Intelligence (AI) in healthcare holds significant promise, yet remains constrained by data scarcity, privacy and ethical concerns, and the inherent complexity of medical tasks. This review explores the intersection of deep generative modeling and federated learning, two complementary paradigms that enable collaborative and privacy-preserving innovation. We survey recent trends in deep generative models applied to healthcare, with a particular focus on federated settings. The review is structured around model architectures, providing a taxonomy-driven analysis that emphasizes solutions tailored to the heterogeneity of healthcare data. Although this review initially intended to cover multimodal applications, the current literature is largely limited to single-modality models. Among deep generative approaches, Generative Adversarial Networks dominate the field, particularly in domains such as medical image synthesis and data augmentation. In contrast, alternative architectures, including variational autoencoders, diffusion models, and autoregressive models, remain barely explored in federated healthcare scenarios. Notably, the number of related publications has increased substantially in recent years, rising from 1 in 2020 to 16 in 2024, reflecting the growing interest and momentum in this area. This paper highlights key challenges, open research questions, and future directions for the development of trustworthy, distributed generative AI in healthcare.

The application of Machine Learning (ML) in healthcare has opened unprecedented avenues for predictive analytics, diagnostics, and personalized medicine. However, the sensitivity of healthcare data and the ethical dilemmas associated with automated decision-making necessitate a rigorous ethical framework. This review paper aims to provide a comprehensive overview of the existing ethical frameworks that guide ML in healthcare and evaluates their adequacy in ad-dressing ethical challenges. Specifically, this article offers an in-depth examination of prevailing ethical constructs that oversee healthcare ML, spotlighting pivotal concerns: data protection, in-formed assent, equity, and patient autonomy. Various analytical approaches including quantitative metrics, statistical methods for bias detection, and qualitative thematic analyses are applied to address these challenges. Insights are further enriched through case studies of Clinical Decision Support Systems, Remote Patient Monitoring, and Telemedicine Applications. Each case is evaluated against existing ethical frameworks to identify limitations and gaps. Based on our com-prehensive review and evaluation, we propose actionable recommendations for evolving ethical guidelines. The paper concludes by summarizing key findings and underscoring the urgent need for robust ethical frameworks to guide ML applications in sensitive healthcare environments. Future work should focus on the development and empirical validation of new ethical frameworks that can adapt to emerging technologies and ethical dilemmas in healthcare ML.

Private record linkage (PRL) is the problem of identifying pairs of records that are similar as per an input matching rule from databases held by two parties that do not trust one another. We identify three key desiderata that a PRL solution must ensure: (1) perfect precision and high recall of matching pairs, (2) a proof of end-to-end privacy, and (3) communication and computational costs that scale subquadratically in the number of input records. We show that all of the existing solutions for PRL? including secure 2-party computation (S2PC), and their variants that use non-private or differentially private (DP) blocking to ensure subquadratic cost -- violate at least one of the three desiderata. In particular, S2PC techniques guarantee end-to-end privacy but have either low recall or quadratic cost. In contrast, no end-to-end privacy guarantee has been formalized for solutions that achieve subquadratic cost. This is true even for solutions that compose DP and S2PC: DP does not permit the release of any exact information about the databases, while S2PC algorithms for PRL allow the release of matching records. In light of this deficiency, we propose a novel privacy model, called output constrained differential privacy, that shares the strong privacy protection of DP, but allows for the truthful release of the output of a certain function applied to the data. We apply this to PRL, and show that protocols satisfying this privacy model permit the disclosure of the true matching records, but their execution is insensitive to the presence or absence of a single non-matching record. We find that prior work that combine DP and S2PC techniques even fail to satisfy this end-to-end privacy model. Hence, we develop novel protocols that provably achieve this end-to-end privacy guarantee, together with the other two desiderata of PRL. Our empirical evaluation also shows that our protocols obtain high recall, scale near linearly in the size of the input databases and the output set of matching pairs, and have communication and computational costs that are at least 2 orders of magnitude smaller than S2PC baselines.

Synthetic patient data in health care: a widening legal loophole

The success of Machine Learning (ML) models in healthcare relies heavily on the quality of data used. High-quality data are crucial for improving the predictive capabilities and the overall performance of ML systems. Despite this, research on data quality in healthcare and ML remains limited, with varying definitions of issues and dimensions across contexts. This study introduces a structured, expert-driven framework for prioritizing data quality dimensions critical to ML performance in healthcare. In contrast to performance evaluation studies involving machine learning algorithms or classifiers, this research does not encompass the training or comparison of predictive models. It addresses a key gap in ML data control by integrating ISO/IEC 25012 with Priestley’s classification and using the Analytic Hierarchy Process (AHP) to evaluate 15 dimensions based on expert judgment. The findings identify Completeness (21.25%), Accuracy (15.53%), Consistency (14.31%), Currentness (14.82%), and Precision (13.82%) as the most influential dimensions for ML healthcare outcomes. A One-at-a-Time (OAT) sensitivity analysis with ±17.6% perturbation confirms the robustness of prioritization despite expert input variability. Key contributions include: (1) a tailored framework for ML healthcare data; (2) AHP-based dimension prioritization; (3) validation through sensitivity testing; (4) insights into data quality’s impact on ML fairness and transparency; and (5) practical guidance for data governance and resource allocation. Future work will apply this framework to clinical datasets to validate its effectiveness in enhancing ML model performance and generalizability.

While there is significant enthusiasm in the medical community about the use of artificial intelligence (AI) technologies in healthcare, few research studies have sought to assess patient perspectives on these technologies. We conducted 15 focus groups examining patient views of diverse applications of AI in healthcare. Our results indicate that patients have multiple concerns, including concerns related to the safety of AI, threats to patient choice, potential increases in healthcare costs, data-source bias, and data security. We also found that patient acceptance of AI is contingent on mitigating these possible harms. Our results highlight an array of patient concerns that may limit enthusiasm for applications of AI in healthcare. Proactively addressing these concerns is critical for the flourishing of ethical innovation and ensuring the long-term success of AI applications in healthcare.

Differential privacy is a precise mathematical constraint meant to ensure privacy of individual pieces of information in a database even while queries are being answered about the aggregate. Intuitively, one must come to terms with what differential privacy does and does not guarantee. For example, the definition prevents a strong adversary who knows all but one entry in the database from further inferring about the last one. This strong adversary assumption can be overlooked, resulting in misinterpretation of the privacy guarantee of differential privacy. Herein we give an equivalent definition of privacy using mutual information that makes plain some of the subtleties of differential privacy. The mutual-information differential privacy is in fact sandwiched between $\epsilon$-differential privacy and $(\epsilon,\delta)$-differential privacy in terms of its strength. In contrast to previous works using unconditional mutual information, differential privacy is fundamentally related to conditional mutual information, accompanied by a maximization over the database distribution. The conceptual advantage of using mutual information, aside from yielding a simpler and more intuitive definition of differential privacy, is that its properties are well understood. Several properties of differential privacy are easily verified for the mutual information alternative, such as composition theorems.

Machine learning (ML) is increasingly embedded in healthcare as an operational capability, not only as a set of clinical prediction models. This manuscript synthesizes evidence across three interdependent areas where ML reshapes value and risk: (1) security and privacy of patient data and critical healthcare infrastructure, (2) healthcare supply chains that determine the availability of drugs, devices, and vaccines, and (3) patient engagement technologies—especially conversational agents—that influence access, adherence, satisfaction, and trust. Building on healthcare analytics work such as OncoViz USA for population insight (Hasan et al., 2021) and recent studies on infrastructure protection, supply resilience, and chatbots (Hasan et al., 2022; Rasel et al., 2022; Khan et al., 2024), we develop an integrative Secure–Supply–Engage (SSE) framework. The review highlights recurring technical risks—membership inference, adversarial manipulation, and distribution shifts—and organizational risks—misaligned incentives and weak governance (Finlayson et al., 2019; Shokri et al., 2017). Across domains, effective practice converges on four principles: privacy-by-design (e.g., differential privacy and federated learning), integrity monitoring across model and data pipelines, resilience metrics for supply decisions, and patient-centered evaluation of digital engagement tools (Dwork, 2006; Kaissis et al., 2020; Rieke et al., 2020; Tudor Car et al., 2020). We translate these principles into actionable design requirements, a threat-to-control matrix, and reporting templates that connect clinical performance to operational reliability and equity. By treating ML as a sociotechnical system rather than a standalone algorithm, the SSE framework clarifies how security, supply continuity, and engagement must be co-designed to make ML clinically useful, operationally reliable, and socially legitimate.

In the data-driven healthcare sector, balancing privacy protection and model performance is critical. This paper enhances accuracy and reliability in survival analysis by integrating differential privacy, deep learning, and the Cox proportional hazards model within a federated learning framework. Traditionally, differential privacy’s noise injection often degrades model performance. To address this, we propose two adaptive privacy budget allocation strategies considering weight changes across neural network layers. The first, LS-ADP, utilizes layer sensitivity to assess the influence of individual layer weights on model performance and develops an adaptive differential privacy algorithm. The second, ROW-DP, comprehensively assesses weight variations and absolute values to propose a random one-layer weighted differential privacy algorithm. These algorithms provide differentiated privacy protection for various weights, mitigating privacy leakage while ensuring model performance. Experimental results on simulated and clinical datasets demonstrate improved predictive performance and robust privacy protection.

The explosion of artificial intelligence in Healthcare (HC) marks the beginning of a new era, a transformative approach that grips the possibility of transforming the domain of patient outcome prediction. KDD for medical data has been widely developed; however, the advancement of heterogeneous medical data sources consisting of both structured (e.g., electronic health records and laboratory outputs) and unstructured data (e.g., Medical records and imaging reports) is still limited. In response, this work presents an AI-based predictive framework that utilizes state-of-the-art data harmonization methods to integrate heterogeneous HC data and improve predictive performance and clinical utility. The proposed approach ensures more reliable and complete patient analysis by tackling the issues associated with traditional systems, such as missing data, inconsistencies, and fragmented data. The framework has shown an order of improvement in predictive outputs referred to with other methods, as measured by Area under the Curve, F-measure, and Mean Absolute Error. Broadly, these results demonstrate the promise that AI-integrated data could have to transform patient intake, providing a cost-effective and reproducible means of supporting outcomes prediction and care trajectories across diverse contexts.

Differential Privacy and Homomorphic Encryption in Healthcare Artificial Intelligence

A major roadblock in machine learning for healthcare is the inability of data to be shared broadly, due to privacy concerns. Privacy preserving synthetic data generation is increasingly being seen as a solution to this problem. However, since healthcare data often has significant site-specific biases, it has motivated the use of federated learning when the goal is to utilize data from multiple sites for machine learning model training. Here, we introduce FELICIA (FEderated LearnIng with a CentralIzed Adversary), a generative mechanism enabling collaborative learning. It is a generalized extension of the (local) PrivGAN mechanism allowing to take into account the diversity (non-IID) nature of the federated sites. In particular, we show how a site with limited and biased data could benefit from other sites while keeping data from all the sources private. FELICIA works for a large family of Generative Adversarial Networks (GAN) architectures including vanilla and conditional GANs as demonstrated in this work. We show that by using the FELICIA mechanism, a site with a limited amount of images can generate high-quality synthetic images with improved utility, while none of the sites need to provide access to their real data. The sharing happens solely through a central discriminator with access limited to synthetic data. We demonstrate these benefits on several realistic healthcare scenarios using benchmark image datasets (MNIST, CIFAR-10) as well as on medical images for the task of skin lesion classification. We show that the utility of synthetic images generated by FELICIA surpasses that of the data available locally and we demonstrate that it can correct the reduced utility of a biased subgroup within a class.