An Explainable Multi-Level Framework for Cervical Cancer Detection Using Traditional Computer Vision and Deep Learning

The primary objective of this research is to develop an accurate and explainable automated cervical cancer detection framework using Pap smear images from the SIPAKMED dataset. The study aims to design and evaluate three novel deep learning architectures, namely PSMF Net, CAD Fusion, and KCPD Net, for effective multi-class cervical cell classification. It also focuses on extracting multi-scale morphological and cellular features through advanced attention and feature fusion mechanisms to improve diagnostic performance. Another objective is to enhance model interpretability by integrating Explainable AI techniques such as Grad-CAM and SHAP, enabling visualization of important regions influencing predictions. Furthermore, the research aims to provide a reliable, efficient, and clinically supportive screening system that facilitates early cervical cancer detection and assists healthcare professionals in decision-making.