Cervical cytology image segmentation is a crucial component in the automated analysis of cervical cytology screening. This research investigates the efficacy of federated learning against traditional learning methods utilizing variants of the U-Net model family. The study leverages the Cx22 dataset, recognized as the most prominent dataset, obviating concerns related to additional data generation. In the absence of a dedicated dataset like Cx22, attributed to the confidentiality of organizations, this research investigates the potential of federated learning to deliver comparable efficiency to conventional training and learning approaches that can aid in maintaining the confidentiality of the data from different organizations. Results indicate comparable accuracy levels, with models trained using federated learning attaining a remarkable dice coefficient score of 95%. Consequently, this suggests a promising avenue for advancing deep learning models even in scenarios where data privacy remains a paramount concern.