FedQuantum-SkinX: Federated Learning with Quantum-Inspired Attention for Skin Cancer Diagnosis

Early and accurate detection of skin cancer remains a significant challenge in clinical practice due to the heterogeneity of lesion morphology, the scarcity of labeled datasets, and privacy concerns in distributed medical environments. The limited availability of labeled image datasets hinders the proper training of diagnostic models, and strict privacy regulations further restrict the sharing of patient data across institutions. Many existing machine learning and deep learning solutions exhibit limited generalization to new data, require substantial computational resources, and rely on centralizing all data, which is impractical for modern systems that combine edge devices and cloud resources. To address these issues, this study introduces a novel framework, termed FedQuantum-SkinX, that integrates three key components: quantum-inspired attention mechanisms, federated learning, and multimodal data fusion. This design enables FedQuantum-SkinX to deliver accurate skin cancer diagnosis while preserving patient privacy, scaling across numerous devices, and operating efficiently in distributed environments. The framework was evaluated on three public datasets (SLICE-3D, 3D-TBP, HAM10000) covering dermoscopic and non-dermoscopic modalities. When evaluated on the benchmark datasets, the framework achieved an average accuracy of approximately 99.0%, around 3% higher than several baseline models on the same datasets. In addition, it reduced training time by approximately 35%, indicating improved computational efficiency.