Skip to content
Log in

Similar papers

Open access Aug 2026

FedSCF: Sensitivity-aware Collaborative Fusion for Personalized Federated Learning in Medical Image Classification

In smart healthcare scenarios, patient data are distributed across different medical institutions. Due to privacy protection and data security constraints, such data cannot be directly shared, which consequently affects the training performance of medical image models. Federated Learning (FL) enables collaborative model training among multiple parties without exchanging raw data, allowing models to leverage multi-source data to improve overall performance and generalization capability. However, under non-independent and identically distributed (non-IID) conditions, significant discrepancies among client data can lead to unstable convergence and performance degradation during training. Personalized Federated Learning (PFL) alleviates this issue to some extent. Nevertheless, existing methods mostly rely on coarsegrained parameter partitioning, which fails to capture internal parameter heterogeneity and lacks effective modeling of inter-client similarity, thereby limiting the effectiveness of collaboration. We propose a personalized federated learning framework, FedSCF, which models client heterogeneity at the parameter level. Specifically, a relative perturbation-based sensitivity evaluation is designed to identify critical parameters for personalized modeling, while the remaining parameters participate in cross-client sharing. The overlap of critical parameter masks is utilized to characterize inter-client similarity, enabling the construction of a dynamic collaboration structure. An adaptive threshold is further introduced to select collaborative neighbors, allowing clients with similar data distributions to form stable collaboration relationships. During the aggregation stage, a dual-path strategy is developed to jointly learn globally shared knowledge and personalized features, and a mask-guided parameter-wise fusion mechanism is employed to complete model updates. Experimental results on the Kvasir and Dermoscopic datasets demonstrate that FedSCF outperforms current state-of-the-art methods under heterogeneous data settings.

Mingjun Wei, Rongyang Xu, Qian Zhang et al. · 0 citations
Preprint Jul 2026

FedProIn: Mitigating Client Drift for Learnable Prototypes in Federated Medical Imaging

Federated learning (FL) is severely hindered by statistical heterogeneity due to variations in scanners, acquisition protocols, and patient populations. Such non-IID data induces client drift during local optimization, leading to unstable convergence and suboptimal global models when parameter-based aggregation is applied. We propose a prototype-based, influence-aware federated learning framework (FedProIn) that uses multiple learnable class prototypes to capture shared semantic structures across heterogeneous clients. We introduce feature divergence loss and prototype contrastive loss to mitigate client drift by decomposing it into feature drift and prototype drift. In addition, we propose a normalized influence aggregation strategy that adaptively weights client prototypes according to their contribution to the global representation, reducing the impact of biased or low-quality updates. Experimental results on two publicly available medical datasets, HAM10000 and Matek-19, demonstrate that FedProIn achieves accuracies of (83.5% IID, 81.1% non-IID) on HAM10000 and (96.2% IID, 95.8% non-IID) on Matek-19, respectively, outperforming existing baselines in both conditions. Our code is available at https://github.com/harsh-kmr/FedProIn.

Harsh Kumar, T. Garg, V. Sundaresan · 0 citations
Aug 2026

STPFL: Stable teacher-guided personalized federated learning for heterogeneous multi-institution medical imaging

Multi-institutional medical imaging data are distributed across sites and cannot be pooled for centralized training due to privacy and governance constraints. Federated learning enables collaborative training without sharing raw data; however, under non-IID client distributions, it often faces unstable optimization and weak global-local knowledge transfer. Personalized federated learning mitigates this issue through client-specific adaptation, but many regularization-, meta-learning-, or distillation-based methods still degrade under highly skewed partitions. To address this, we propose STPFL, a stable teacher-guided personalized federated learning framework. STPFL builds an EMA-based aggregated teacher to accumulate historical global knowledge and provide consistent guidance. Each client optimizes a unified objective combining supervised learning with temperature-scaled KL distillation from the teacher model. On the server side, STPFL applies validation-driven performance weighting as a reliability signal and performs Adam-style aggregation to improve convergence under cross-silo/multi-institutional client distributions. We evaluate STPFL under controlled data heterogeneity (α = 0–1) and cross-silo client-count scalability (1–10 clients) on three benchmark datasets (PathMNIST, BloodMNIST, TissueMNIST) and three real-world medical imaging datasets (brain tumour MRI, blood cell cancer, and Kermany2018 OCT). Compared with FedAvg, FedProx, MetaFed, and FedLFP, STPFL achieves 4–23% and 6–29% improvements in global accuracy and F1-score, while personalized models improve by 4–25% and 4–32%, respectively. Under cross-silo client-count scalability, STPFL further improves global accuracy and F1-score by 5–30% and 5–40%, with corresponding personalized gains of up to 13–33% and 17–47%.

R. Semwal, Imlimaong Aier, P. Varadwaj · 0 citations
Open access Jul 2026

FLMMIF: privacy-preserving federated multi-modal medical image fusion

As a pivotal technique in smart healthcare, medical image fusion integrates complementary functional and structural information to facilitate accurate diagnosis and enhance clinical decision-making reliability. However, existing centralized methods typically raise serious data privacy concerns, while standard distributed approaches often fail to balance global generalization with local node personalization due to data heterogeneity. To address this, we propose FLMMIF, a privacy-preserving framework integrating a federated learning paradigm and low-rank adaptation for personalized and secure medical image fusion. During the local training phase, we utilize a dual-branch encoder and single-branch decoder, adopting a two-stage iterative strategy: initially training low-rank parameters to secure local personalization, followed by training full-rank parts to guarantee global baseline performance. Subsequently, this iterative process ensures that the model dynamically coordinates specific local features with general global knowledge before parameter transmission. Finally, we establish a metric-based aggregation mechanism on the server, FedIF, which evaluates the performance of uploaded models to assign higher aggregation weights to superior nodes for optimized global updating. Experimental results demonstrate that FLMMIF generates high-quality fusion results that effectively protect data privacy while achieving precise node-specific personalization.

Lei Meng, Shuangsong Ren, Jing Wang et al. · 0 citations
Open access Aug 2026

PPFedKD: Privacy‐Preserving Federated Learning and Adaptive Contrastive Distillation for Medical Image Diagnosis

Federated learning (FL) in medical image analysis offers privacy‐preserving collaborative training but faces challenges such as the impact of Non‐IID data on model performance and communication overhead that limits system efficiency. Existing methods often struggle to balance data heterogeneity with privacy protection in high‐dimensional pathological image classification tasks. To overcome these challenges, we propose a Privacy‐Preserving Self‐Distillation Federated Framework (PPFedKD) that enhances both model performance and privacy security through optimized feature learning and communication aggregation. The proposed framework introduces a adaptive contrastive distillation strategy, leveraging spatial and semantic augmentations to produce consistent feature representations and mitigate imbalances in client data distributions. It also employs a semi‐asynchronous dynamic aggregation mechanism with a lag threshold, allowing the server to accept slightly outdated updates, thereby reducing communication redundancy and wait times. Furthermore, privacy‐enhanced gradient perturbation is applied on the client side using a noise constraint strategy to effectively mitigate privacy leakage while ensuring stable convergence. Experimental results on diabetic retinopathy and breast cancer pathology datasets demonstrate that PPFedKD outperforms baseline methods in classification accuracy, privacy protection, and communication efficiency, providing a secure and effective solution for medical image classification.

Lei Yuan, Yaohua Luo, Mei Feng · 0 citations
Jul 2026

Federated Learning with Global Model Hint for Medical Image Object Detection.

Building an ideal medical image object detection model often requires sufficient training data, which can be challenging to obtain in practical scenarios. Manual annotation is labor-intensive, and sharing datasets may raise data privacy concerns. Although federated learning can partially address these issues, we find an amplified feature drift problem when it is directly applied to medical image object detection. Motivated by the observation that the global model's parameters tend to align more closely with those of the oracle model than with those of the client models, we propose FedMHDet: Model Hint Federated Learning Detection Model, a novel federated learning detection framework. During the training phase, FedMHDet leverages multi-scale feature consistency as a global model hint to guide client models, thus mitigating the feature drift problem. Extensive experiments on pulmonary lesion and brain tumor detection tasks show that FedMHDet achieves favorable overall performance. Compared to the strongest baseline under each corresponding metric, it improves average AP by 1.05 and 0.19, and average sensitivity by 1.10 and 0.43 on the two tasks, respectively. We also provide in-depth analyses to support the practical use of our method. The code is available at https://github.com/bbamai/FedMHDet.

Zhenghua Xu, Gaoxi Zhou, Hexiang Zhang et al. · 0 citations