FedCervical: Shedding Light on Cervical Cancer Detection with Federated Learning and Explainable AI

Abstract

Cervical cancer, posing as a global health challenge, demands precise detection methods. This research brings in an innovative approach to identify cervical cancer, utilizing advanced machine learning on pap smear data. This research utilizes Federated Learning (FL) and Explainable AI (XAI) to improve cervical cancer detection accuracy while ensuring patient’s privacy. After studying the SIPaKMeD dataset, we have developed a FL model capable of classifying cervical cell images into five distinct categories. The decentralized nature of FL ensures robust data privacy, allowing multiple institutions to collaboratively train a model without sharing sensitive patient information. The FedAvg Model implemented using the Flower framework achieves a 95.12% accuracy and paves a way for further usage of such technologies in the field of medical sciences.

Acknowledging the sensitivity of health information, Federated Learning (FL) emerges as a crucial tool, ensuring privacy in data handling. Pap smears, integral to screening, occasionally pose challenges in early detection. FL, through decentralized models, optimally manages extensive datasets, enhancing diagnostic accuracy. In an era of prioritizing data privacy, FL becomes a safeguard, allowing collaborative model training without compromising individual confidentiality. Beyond security, FL democratizes cervical cancer diagnostics, offering patients nuanced insights into their results. This research signifies a substantial step in advancing cervical cancer detection, marrying optimal model performance with stringent privacy protection, resonating with evolving healthcare paradigms and individual empowerment in the diagnostic process. Also, the amalgamation of Explainable AI (LIME) with FL leads to a solution which can be used by onco- pathologists around to world to successfully diagnose cancerous cells.