By enabling real-time diagnostics, protecting patient privacy, and improving resource utilization in healthcare facilities, integrating federated learning (FL), machine learning, and artificial intelligence in edge devices can completely transform healthcare services. The early diagnosis of cervical anomalies in hospitals in three countries (Italy, Ethiopia, and Albania) was the main focus of this study, which investigated FL to evaluate colposcopy data. A ResNet50-based deep learning model was developed for feature extraction and classification using a data set of 385 colposcopy images annotated by five expert gynecologists in each region using the Computer Vision Annotation Tool (CVAT). In addition, the UNET deep learning technique was used to segment the regions of cervical cancer. To preserve patient privacy and ensure data sovereignty, the proposed approach utilizes federated learning to train the ResNet50 model collaboratively across multiple edge devices, using Flower 4.1 and Stratus Cloud tools for simulation. The proposed approach achieved a classification accuracy of 68.72% with a loss of 2.199% in a holdout test set, demonstrating slightly good generalization across devices with minimal performance degradation during federated aggregation. This study underscores the potential of advanced AI-driven methodologies in resource-constrained healthcare settings. In addition, the proposed approach minimizes the need for centralized data storage while protecting patient privacy using FL. The foundation for expanding this approach to larger datasets and a wider range of healthcare applications is laid by these encouraging outcomes, which will eventually increase access to trustworthy diagnostics in environments with limited resources.