A Spatial-Frequency Feature Ensemble for Detecting Cervical Dysplasia from Pap Smear Images

Abstract

Among women, cervical cancer is the commonest and the most treatable and preventable type of cancer. In most cases, cervical cancer begins as precancerous changes which gradually develop into cancer. Pap smear is widely used for cervical cancer diagnosis. Cell analysis is a time-consuming and cumbersome job; thus, an automatic detecting framework is proposed. Wavelet transforms offer the associated coefficients as the input image data representation, used as feature vectors. Artificial Neural Networks (ANNs) have outstanding attributes such as enhanced input-to-output mapping, non-linearity, fault tolerance, adaptively, and self-learning. Classification of cervical cancers employs neural network systems that have a huge role in most applications related to image processing. For application in diverse fields such as bioinformatics and pattern recognition, most researchers choose ensemble classifiers. A spatial-frequency feature ensemble has been proposed in this work to identify cervical dysplasia from images of Pap smears.