Multiclass Classification of Risk Factors for Cervical Cancer Using Artificial Neural Networks

World Health Organization statistics show that cervical cancer is the fourth most frequent cancer in women with an estimated 530,000 new cases in 2012. Cervical cancer diagnosis typically involves liquid-based cytology (LBC) followed by a pathologist review. The accuracy of decision is therefore highly influenced by the expert’s skills and experience, resulting in relatively high false positive and/or false negative rates. Moreover, given the fact that the data being analyzed is highly dimensional, same reviewer’s decision is inherently affected by inconsistencies in interpreting the data. In this study, we use an Artificial Neural Network based model that aims to considerably reduce experts’ inconsistencies in predicting cervical cancer. We rely on standard machine learning techniques to train the neural network using six experts’ predictions for cervical cancer (based on analysis of more than sixty parameters/risk factors) and we produce a model where the unanimous decision is predicted with very good accuracy