Privacy-preserving collaborative AI for distributed deep learning with cross-sectional data

Recent progress in Deep Learning (DL) has shown potential in intelligent healthcare applications, enhancing patients' quality of life. However, improving DL precision requires a larger and diverse dataset, leading to privacy and confidentiality challenges when consolidating data at a centralized server. To address this, we propose a skin cancer detection method prioritizing patient information and privacy. "Skin-net," a novel Convolutional Neural Network (CNN) model, integrates progressively private Federated Learning (FL) for accurate classification of complex skin lesion images. FL ensures data confidentiality during model training. Skin-net achieves promising results, with 98.3%+/- accuracy, 98.8%+/- sensitivity, and 97.9%+/- specificity, while preserving data privacy. It offers an effective pathway for skin cancer analysis and image augmentation, mitigating privacy concerns in medical image analysis.