Prediction of Breast Cancer Diagnosis by Blood Biomarkers Using Artificial Neural Networks

Abstract

The use of artificial neural networks (ANNs) has been very helpful in carrying out prediction, classification and data optimization tasks. In this work, ANNs were used to predict the diagnosis for breast cancer in a women population with overweight or obese and possible diabetes, coupled with a pre or postmenopausal stage and compared against other machine learning techniques reported in literature. The algorithms used to train the ANNs models were Scaled Conjugate Gradient, Resilient Backpropagation and Conjugate Gradient Backpropagation with Powell Beale Restarts. The algorithms results were compared with the original dataset creator’s work, as well as other authors using the same dataset for classification task, a better classification was accomplished using this work ANNs. With four predictors the next values were obtained, AUC = 0.96, sensitivity = 0.96, specificity = 0.96 and Youden index = 0.92. With nine predictors the next values were obtained, AUC = 0.96, sensitivity = 0.95, specificity = 0.97 and Youden index = 0.92. Different strategies are suggested to improve the results, exploring more hidden layers and a different validation method.