Um Modelo de Predição de Mortalidade em Unidades de Terapia Intensiva Baseado em Deep Learning

Resumo

The usage of Deep Learning techniques has become even more frequent in medical research due to the possibilities that it offers to improve the quality of Clinical Decision Support. Several conventional prognostic models have been used in Intensive Care Units (ICU) to evaluate the risk of death. However, those models still cannot accurately predict that risk. For this reason, the aim of this article is to offer a model based on Deep Learning to predict the risk of mortality, especially in the fields of intensive care medicine, in order to make healthcare more efficient. The model consists of a Convolutional Neural Network that is divided into five stages, which contain nine hidden layers. In the proposed model we use quantitative methods in its processes. An experimental approach is given when comparing the predictive power of the proposed model to one of the most used models for predictions in ICU, the APACHE II. The data used were extracted from medical records available in the Multiparameter Intelligent Monitoring in Intensive Care III (MIMIC III) database. In order to evaluate the performance of the models, measures of accuracy, sensitivity, specificity and Area Under the ROC Curve (AUC) were used. After comparing the performance of the proposed model to the APACHE II, the proposed model presented positive results as it reached an AUC of more than 0,80, whereas the APACHE II reached an AUC of 0,71.