A Segmentation-Assisted Multilevel Ensemble of Convolutional Deep Learning and Statistical Regression Models (SAME-SRM) for Cervical Precancer Identification in Pap Smear Images

Abstract

Cervical cancer remains one of the leading causes of mortality among women worldwide, primarily due to late diagnosis and limited access to accurate screening methods. This study presents a deep learning-based framework for the effective classification and segmentation of cervical cancer from Pap smear images. The proposed model, named SAME-SRM (Segmentation-Assisted Multi-feature Extraction with Spatial Refinement Mechanism), integrates advanced image preprocessing, contrast enhancement, and segmentation techniques to improve diagnostic accuracy. Using an optimized EfficientNet backbone, the framework effectively captures subtle morphological variations between normal and cancerous cervical cells, ensuring high precision and robustness in feature representation. Experimental evaluation conducted on benchmark cervical cancer datasets demonstrates the superior performance of the proposed SAME-SRM model compared to existing approaches such as DenseNet-121, CYENET, and SVM. The model achieves an accuracy of 97.92%, a precision of 96.25%, a recall of 96.88%, and an F1-score of 99.48%, validating its efficiency in automated diagnostic applications. The results highlight the potential of the proposed framework to assist pathologists in early cancer detection, reduce manual diagnostic errors, and promote the development of AI-driven medical screening systems for cervical cancer prevention and treatment planning.