Enhanced Cervical Cancer Detection and Classification Through Deep Learning: a Review of Methodologies and Challenges
Saranya Al, S. Ravil, T. Kalaichelvi · 2025
Cervical cancer is still a significant public health issue, especially in developing nations where screening and early detection are not adequately implemented. This paper presents an overview of existing methods for the detection of cervical cancer, with emphasis on the use of deep learning methods to automate and enhance the accuracy of cervical cancer screening from Pap smear images. Although promising results from models like Convolutional Neural Networks (CNNs), U-Net, and Mask RCNN, some limitations still exist, e.g., data imbalance, unstable image quality caused by variability in staining and imaging conditions, and poor interpretability of models. These limitations affect the reliable use of these models in real clinical environments. This paper suggests an enhanced detection pipeline that integrates sophisticated segmentation methods, transfer learning, and stable classification algorithms to overcome these issues and improve the detection process. The classification process becomes more transparent by applying explainable AI methods to describe how predictions are made. To provide reliability, the paper combines multi-modal data, uses data augmentation to handle class imbalance, and includes model scalability to be deployed in low-resource settings. Future research aims to enhance segmentation accuracy further, improve transfer learning methods, and investigate generative models to mitigate current data limitations. These advancements seek to make cervical cancer detection systems more reliable, robust, and accurate.