Attention Based Meta-Module to Integrate Cervigrams with Clinical Data for Cervical Cancer Identification

Cervical cancer remains a significant burden on public health, particularly in developing countries, where its malignancy and mortality rates are alarmingly high. Early diagnosis stands as a pivotal factor in effectively treating and potentially curing the cervical cancer. This study introduces a novel approach of meta module based on recurrent gate architecture designed to enhance the classification of cervix images efficiently. This innovative framework incorporates a meta module capable of dynamically selecting image modalities’ most pertinent attributes. Furthermore, it integrates clinical data with extracted image features and employs a range of EfficientNet architectures (B0-B5) for image classification. Our results indicate that the EfficientNet B5 architecture outperforms its counterparts, achieving an AUC (Area Under the Curve) score of 55.1 and an F1-Score of 75.1. Overall, this work represents a crucial step towards improving the early detection of cervical cancer, which in turn can lead to more effective treatment strategies and, ultimately, better outcomes for patients worldwide.