Vision Transformer with Automated Clinical Feature Extraction and Dual Attention Bias for Cervical Cancer Classification

Cervical cancer is the fourth leading cause of cancer-related deaths among women worldwide, with a particularly high burden in low-income countries. To address the need for cost-effective and accessible screening solutions, we propose a novel Vision Transformer (ViT)-based model for classifying cervical cancer from tele-cervicography images. To enhance classification performance, the model automatically extracts clinically meaningful anatomical features—such as the cervical region, external os location, and acetowhite lesion distribution—and integrates them into a dual attention bias mechanism that directs the model's focus to key diagnostic areas. This mechanism emphasizes both the global cervical area and the spatial relationship between acetowhite lesions and the external os, enabling the model to attend to clinically significant regions more effectively. Trained on a dataset of over 17,000 tele-cervicography images, the proposed approach achieved a 2.04% improvement in accuracy, a 4.89% increase in specificity, and a 4.07% improvement in precision compared to the baseline ViT, while maintaining comparable sensitivity. Furthermore, the automated preprocessing pipeline enables fully automated clinical workflows, offering a scalable diagnostic mechanism suitable for real-world medical applications.