Automated Cervical Cancer Detection Using Feature-Fused Deep CNNs and Ensemble Learning

Abstract

Cervical cancer remains a significant global health concern, ranking as the fourth most common cancer among women. Early detection through Pap smear screening is vital for improving treatment outcomes. Computer-aided detection systems can support clinical decision-making by providing accurate and timely diagnoses. This paper proposes a deep learning model for automated cervical cancer detection using Pap smear images. Pre-trained Convolutional Neural Networks (CNNs), InceptionV3, InceptionResNetV2, and MobileNetV2, are fine-tuned with additional layers to extract specialized features through transfer learning. The extracted feature vectors are concatenated to form a unified representation, which is then used as input to multiple classification algorithms. Among these, the Bagging classifier with Random Forest as the base estimator achieves the highest performance. The model attained 97.25% accuracy, 97.26% precision, 97.28% recall, and a 97.26% F1-score on the SIPaKMeD dataset. It also achieved 96.72% accuracy on the Herlev dataset and 99.47% on the Mendeley Liquid-Based Cytology dataset. The results show that the proposed approach consistently outperforms individual CNN baselines as well as several state-of-the-art methods reported in the literature.