The research, led by Suresh Arumugam of Dayananda Sagar University in Bengaluru together with colleagues from three other Indian institutions, introduces FedXAI-Health, a federated learning framework designed for multi-label thoracic disease classification. In federated learning, the model travels to the data rather than the other way around: each participating hospital trains a shared neural network on its own machines, and only the resulting model updates, never the patient images themselves, are sent to a central server for aggregation. The framework was tested in a simulated network of five hospital clients, each holding a differently distributed slice of the publicly available NIH Chest X-ray 14 dataset.

That heterogeneity is the crux of the problem the researchers set out to solve. In real-world federated networks, no two hospitals see the same mix of diseases, scanner types, or patient demographics, a condition known in machine learning as non-IID data. When local datasets diverge this sharply, the averaged model produced by standard federated algorithms can drift away from what any individual site needs. To simulate this faithfully, the team partitioned the data using Dirichlet distributions with concentration parameters ranging from 0.1, which produces extremely skewed client datasets, to 5.0, which approaches a more balanced split, and validated robustness across the entire spectrum.

The headline result is genuinely counterintuitive. Under identical experimental conditions, using an EfficientNetB0 backbone, a batch size of 512, and the Adam optimiser with a weight decay of 10 to the power of minus 4, the classic federated averaging algorithm known as FedAvg achieved a Macro-AUC of 0.8060, plus or minus 0.0048. That figure not only held up under severe non-IID conditions but actually exceeded the centralised baseline, in which all data would have been pooled on one server, by 2.0 percent. The centralised EfficientNetB0 model managed only 0.7859, plus or minus 0.0142. In other words, the privacy-preserving approach did not merely match the gold standard it is usually measured against; it beat it.

The authors attribute part of this advantage to a phenomenon familiar from the regularisation literature: training across heterogeneous sites forces the model to encounter a wider effective variety of disease presentations during optimisation, which can act as an implicit form of data augmentation and reduce overfitting to any single distribution. The team also benchmarked FedProx, a popular federated variant that adds a proximal term to keep local models from straying too far from the global one. FedProx reached a Macro-AUC of 0.7879, plus or minus 0.0020, respectable but below FedAvg in this setting.

That gap prompted one of the study’s most practically useful findings, an ablation study across five values of FedProx’s regularisation coefficient, mu, spanning 0, 0.001, 0.01, 0.1, and 1.0. Performance degraded monotonically as the coefficient increased, meaning the stronger the algorithm tried to tether local models to the global average, the worse the multi-label classification became. The lesson for practitioners is that minimal regularisation is optimal for multi-label medical imaging under these conditions, a direct challenge to the intuition that heterogeneous data always demands stronger constraints. For clinical teams designing federated deployments, this kind of tuning guidance can save months of trial and error.

Accuracy alone, however, is not enough to earn the trust of radiologists, and this is where the explainability component of FedXAI-Health becomes central. The framework pairs its federated training pipeline with two complementary attribution techniques: SHAP, which assigns each input pixel a contribution score for a given prediction, and Grad-CAM, which produces class-specific heatmap localisations over the image. The researchers compared the maps generated by both methods and confirmed that they produced clinically coherent, disease-specific localisations, meaning the models appeared to base their diagnoses on the actual anatomical regions associated with each condition rather than on spurious correlations such as hospital-specific imaging artefacts.

This matters because multi-label thoracic classification is a genuinely hard task. The NIH Chest X-ray 14 dataset covers fourteen distinct conditions, including cardiomegaly, pleural effusion, pneumonia, and pneumothorax, and a single radiograph can carry several of them simultaneously. A model that flags a disease without showing where it saw the evidence is of limited clinical value, and a federated model whose explanations differ wildly across sites would be even harder to trust. By demonstrating that attribution maps remain coherent across a decentralised, heterogeneous network, the study addresses the trustworthiness question at the same time as the privacy one, rather than treating explainability as an afterthought bolted onto a finished model.

The experimental hygiene underlying these claims deserves attention as well. All results are reported as means with standard deviations across three independent runs, and the dataset was split at the patient level between training and test sets, ensuring zero data leakage, a persistent pitfall in medical imaging research where multiple images of the same patient can silently straddle both sets. The team also ran a supplementary experiment on CIFAR-10 to contextualise the findings beyond the medical domain. Crucially for reproducibility, the complete source code, trained model checkpoints, experiment logs, and result files have been released publicly on GitHub, and the NIH dataset itself is openly available.

The broader significance of the work lies in what it suggests about the future of clinical AI. Federated learning has already been deployed in real hospital networks, most famously in a Nature Medicine study predicting clinical outcomes in COVID-19 patients across twenty institutions, and surveys of the field consistently identify non-IID heterogeneity as its central unsolved challenge. By showing that a well-tuned, lightweight federated setup can outperform centralised training on a benchmark thoracic dataset while producing verifiable explanations, FedXAI-Health strengthens the case that hospitals need not choose between collaboration and confidentiality. The code is open, the data is public, and the recipe is documented, which means other research groups can stress-test the results on their own distributions. If the findings generalise beyond simulated clients to genuine multi-site deployments, the era in which privacy regulations condemned each hospital to train diagnostic AI in isolation may finally be drawing to a close, replaced by networks of institutions that learn together while their patients’ images never leave the building.

Subject of Research: Explainable federated learning for privacy-preserving multi-label thoracic disease classification from chest X-rays under non-IID data distributions

Article Title: Explainable Federated Learning for Trustworthy Thoracic Disease Detection Under Non-IID Data Distributions

Article References: Arumugam, S., Sindhu, A., Saroja, M. N., Kannan, S., & Esakkiammal, A. (2026). Explainable Federated Learning for Trustworthy Thoracic Disease Detection Under Non-IID Data Distributions. Machine Learning, 115(10), Article 228. https://doi.org/10.1007/s10994-026-07174-z

Image Credits: AI Generated

DOI: 10.1007/s10994-026-07174-z

Keywords: federated learning, non-IID data, chest X-ray, explainable AI, SHAP, Grad-CAM, FedAvg, FedProx, EfficientNetB0, privacy-preserving machine learning, thoracic disease detection, NIH Chest X-ray 14

The heterogeneity intrinsic to multi-center medical imaging compounds these challenges further. Distinct differences in scanner hardware, imaging parameters, and anatomical coverage introduce substantial variations in data distributions, hampering the generalizability and robustness of centralized models. Simple aggregation and averaging of data from diverse sources often prove inadequate, as these models struggle to reconcile the disparities inherent to cross-institutional imaging. Federated learning (FL) has rapidly emerged as a compelling alternative paradigm, enabling decentralized model training without necessitating raw data sharing. This approach safeguards patient privacy while leveraging distributed datasets. Yet, FL is not a panacea; when confronted with pronounced cross-institutional heterogeneity, traditional federated averaging schemes underperform, especially in accommodating varying imaging geometries and multi-task objectives within a unified framework.

Addressing these nuanced challenges, researchers led by Hao Wang at Southern Medical University have introduced FedM2CT, an innovative federated metadata-constrained reconstruction framework designed to unify multivendor CT image reconstruction. The method deftly integrates mutual learning with metadata-driven modeling to perform all-in-one CT reconstruction across heterogeneous imaging geometries and protocol variations. This novel architecture astutely circumvents the limitations of prior models by embedding task-specific adaptability directly into the federated learning process. The core of FedM2CT lies in a trifecta of modules: the task-specific iRadonMAP (TS-iRadonMAP), condition-prompted mutual learning (CPML), and federated metadata learning (FMDL). These modules collaboratively maintain privacy, enable cross-client knowledge transfer, and robustly handle data heterogeneity.

TS-iRadonMAP serves as the frontline module, executing local CT image reconstruction through private models tailored to each client’s architecture and data characteristics. It retains sensitive imaging tasks locally while facilitating parameter exchange with the central server for collaborative enhancement. In parallel, CPML orchestrates the exchange of insights by fostering mutual learning within the image-domain submodules, harnessing conditional prompting driven by client-specific metadata such as imaging geometry and scanning parameters. This metadata undergoes transformation via a shallow multilayer perceptron (MLP), producing adaptive feature modulation coefficients that tailor model behavior dynamically across diverse scanning conditions.

A critical innovation in FedM2CT is its utilization of federated metadata learning on the server side. Recognizing that simple parameter averaging is insufficient in heterogeneous contexts, the server aggregates high-quality metadata—paired low-dose and normal-dose images—from multiple sources to train a global metamodel. This meta-model encapsulates cross-domain priors and is judiciously aggregated with client-uploaded CPML parameters through weighted fusion strategies. This fusion integrates global and local knowledge representations, mitigating client-specific heterogeneity and enhancing cross-protocol generalization.

Underpinning the architectural design is a dual-domain iRadonMAP pipeline consisting of sinogram-domain processing, a learnable back-projection module, and image-domain refinement networks. The physically consistent back-projection submodule is sensitively dependent on varying imaging geometries and sampling protocols, necessitating local adaptation to ensure fidelity. Consequently, TS-iRadonMAP limits federated sharing to the image-domain subnetwork, preserving local uniqueness and privacy while exchanging only the shared submodule parameters for global coordination.

The iterative training workflow entails clients performing task-specific reconstructions using TS-iRadonMAP, followed by local mutual learning and conditional prompting within CPML. The server subsequently capitalizes on collected metadata to train the metamodel, which is then integrated with uploaded client parameters. This cycle repeats, progressively improving model performance across a spectrum of scanning environments, thus enabling robust and scalable federated learning that adapts fluidly to operational variances.

Empirical validation of FedM2CT unequivocally demonstrates its superiority over conventional CT reconstruction techniques. Across diverse experimental setups, the method significantly elevates peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) metrics while substantially reducing root mean squared error (RMSE). Its modulation transfer function (MTF) profiles attest to enhanced spatial resolution. Furthermore, in hybrid supervision scenarios—where only subsets of clients possess paired annotated data—FedM2CT sustains its advantage by delivering lower Fréchet Inception Distance (FID) and Learned Perceptual Image Patch Similarity (LPIPS) scores, confirming improved perceptual quality and artifact mitigation relative to federated baselines. By contrast, frequency-domain methods reliant on paired training data, such as FedFDD, falter in balancing denoising with detail preservation on unsupervised clients.

Beyond objective metrics, blind evaluations by professional radiologists endorse FedM2CT’s reconstructions, which exhibit superior visual coherence, texture fidelity, and noise suppression when juxtaposed against alternative methods. These qualitative insights underscore the clinical viability of FedM2CT in delivering diagnostically relevant images compatible across a variety of institutional configurations and scanner types.

Notwithstanding its advance, FedM2CT faces several practical challenges before clinical deployment. Chief among these is the requirement for aggregating diverse CT metadata on a central server, which may conflict with privacy-preserving principles and impede real-world scaling. The current study’s reliance on retrospective simulations, while methodologically rigorous, necessitates future prospective validations involving clinical patient cohorts. Additionally, the framework imposes moderately increased computational demands at the client side relative to existing FL methods; however, emerging strategies including model compression and edge computing can potentially offset these limitations. Selecting optimal hyperparameters remains an open research endeavor, ripe for automated tuning approaches and domain adaptation methodologies to maximize efficacy.

Looking forward, the integration of advanced architectures such as meta-learning or the incorporation of large language models (LLMs) represents a promising direction to further personalize and refine CT reconstructions within FedM2CT. Hao Wang envisions leveraging LLMs to modulate intermediate network features, thus dynamically tailoring model predictions to individual imaging contexts—a compelling frontier that could empower next-generation federated medical imaging.

In summary, the FedM2CT framework embodies a sophisticated synthesis of federated learning, metadata-driven adaptation, and mutual model refinement, providing an exceptional pathway to scalable, privacy-conscious, and accurate all-in-one CT reconstruction. Its foundational contributions address some of the most pressing obstacles in multi-institutional medical imaging research and set the stage for transformative clinical translation.

The research team comprises Hao Wang, Xiaoyu Zhang, Hengtao Guo, Xuebin Ren, Shipeng Wang, Fenglei Fan, Jianhua Ma, and Dong Zeng. Financial support for this pioneering work was provided in part by the National Key R&D Program of China under grants 2024YFA1012000 and 2024YFC2417800, as well as the National Natural Science Foundation of China under grant U21A6005.

The full scientific study, entitled “Federated Metadata-Constrained iRadonMAP Framework with Mutual Learning for All-in-One Computed Tomography Imaging,” was published in the journal Cyborg and Bionic Systems on August 27, 2025, and is accessible via DOI: 10.34133/cbsystems.0376.

Subject of Research: Computed Tomography Reconstruction, Federated Learning, Deep Learning, Medical Imaging

Article Title: Federated Metadata-Constrained iRadonMAP Framework with Mutual Learning for All-in-One Computed Tomography Imaging

News Publication Date: August 27, 2025

Web References: DOI: 10.34133/cbsystems.0376

Image Credits: Hao Wang, Southern Medical University

Keywords: Life sciences, Research methods, Social sciences