Analysis of cervical cancer early detection and classification based on machine learning techniques

Abstract

In many low-income countries, cervical cancer is the second most recurrent cancer among females, following breast cancer. The manual interpretation of Pap smear images requires a significant number of highly skilled personnel, who are costly, time-consuming, labor-intensive, prone to errors, and scarce in many regions. To minimize the risk of both death and morbidity, early recognition of cancer is necessary. This research proposes developing an automated system for the early detection and classification of cervical cancer through machine learning. The research aims to develop a three-phase model for cervical cancer screening and classification. This chapter presents a unique machine learning-driven technique for classifying and identifying abnormal and healthy cervical cells. Initially, unwanted distortions in the Pap smears from the Herlev database are removed using an adaptive median filter. Machine learning models perform best when trained on high-quality data, as the quality of the features in the data defines its effectiveness. To improve the training process, the Auto-Encoder (AE) is used to extract features and reduce their dimensionality. Consequently, the labeling of healthy and unhealthy cervical cells is done using Cascaded Multilayer Perceptrons (c-MLP). To achieve the highest classification accuracy, the Bayesian Regularization technique is used to train the c-MLP. newline