Investigating deep learning approaches for cervical cancer diagnosis: a focus on modern image-based models

Background: Cervical cancer is a leading health concern for women globally, necessitating accurate and timely diagnostic methods. While the Papanicolaou smear (Pap smear) test remains the gold standard for cervical cancer screening, it is time-consuming and prone to human error. This highlights the need for automated diagnostic tools to improve efficiency and accuracy. Methods: This study evaluated the performance of deep learning models for automating cervical cancer diagnosis using Pap smear images. A new dataset was constructed by merging the Mendeley Liquid-Based Cytology (LBC) dataset (963 images) and the Malhari dataset (318 images), resulting in 1,281 images. Twenty-seven cutting-edge deep learning models, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), were used for classification. Data augmentation and transfer learning techniques were applied to enhance model performance. Results: The majority of ViT-based models achieved a high classification accuracy of 99.48%. Among the 13 CNN-based models evaluated, EfficientNetV2-Small was the only model to achieve the same accuracy level. The results demonstrate the superiority of ViT-based models in achieving high diagnostic substantial potential in automating cervical cancer diagnosis. These models can enhance diagnostic accuracy, reduce human error, and provide timely results, thereby supporting more efficient and reliable cervical cancer screening practices.