Automatic prediction of cervical cancer from associated risk factors

CHELALICHE, ANIBAL S; KAUFMAN, CINTIA D; TAPIA ELIZABETH; SPETALE FLAVIO EZEQUIEL

Corrientes

Congreso; XII Argentine Congress of Bioinformatics and Computational Biology; 2022

Facultad de Ciencias Exactas y Naturales y Agrimensura (UNNE)

Background:Cervical cancer is the fourth most common neoplasm in women worldwide. For some time, a design has been pursued for a classification system that can be used in the diagnosis of this pathology using biochemical parameters, demographic data, environmental, genetic factors and images of the cervix.Results:In this work we present an Artificial Neural Network (ANN) to predict the risk factors of CC. In addition, the Support Vector Machine Recursive Feature Elimination (SVM-RFE) method was used to find the most important attributes for cervical cancer prediction. The dataset employed here contains missing values and is highly imbalanced. Therefore, the Tomek Links technique was employed to remove this data from the majority class (negative examples). The feature-selection method showed that the frequency of tobacco consumption, HVP and HIV infections, as well as the number of diagnosed sexually transmitted infections (STDs), could be useful for predicting cervical cancer. The ANN with the selected features from SVM-RFE and Tomek Links showed the best results with an accuracy of 78 %.Conclusions:The associations among certain oncogenic strains of HPV, others STDs and tobacco consumption, and the disease are well established. The development of this model applied to the clinic can become a useful tool for the first stages of patient screening, which helps health professionals to choose subsequent methodologies for disease control and treatment.