A Survey of Methods and Architectures for Automatic Cervical Cancer Detection and Segmentation Using Sparsity-Aware Orthogonal Initialization in Deep Neural Network Classifiers

Abstract

Cervical cancer is a major global health concern, particularly in developing regions where early detection remains limited. Traditional diagnostic methods such as Pap smear and colposcopy rely heavily on manual interpretation, which can be time-consuming and prone to variability. In recent years, deep learning (DL) techniques have emerged as powerful tools for automating cervical cancer detection and segmentation from medical images including Pap smear slides, MRI, and CT scans.This study presents a comprehensive survey of methods and architectures for automatic cervical cancer detection, with a focus on convolutional neural networks (CNNs), segmentation models such as U-Net and nnU-Net, and hybrid frameworks. Additionally, the role of sparsity-aware orthogonal initialization (SAOI) in improving training efficiency and model scalability is examined.Recent advancements demonstrate that deep learning models achieve high classification accuracy (above 95%) and segmentation performance (Dice scores up to 0.90). Hybrid models combining segmentation and classification outperform standalone approaches by improving feature representation and decision-making. However, challenges such as limited datasets, lack of generalization, and computational complexity remain.

This survey highlights current trends, comparative performance, and research gaps, emphasizing the need for scalable, explainable, and clinically deployable AI-based systems for cervical cancer diagnosis.