Deep Learning and Permutation Entropy in the Stratification of Patients with Chagas Disease

Abstract

Chagas disease is a life threatening illness that in the last decades was becoming a public health problem because of the change in the epidemiological pattern. It may be silent and asymptomatic in the chronic phase. Hence the necessity of the development of early markers. To achieve this, we propose a deep neural network architecture in order to classify 292 patients into three groups: The Control group with 83 volunteers, the CH1 group with 102 patients with positive serology and no cardiac involvement and the CH2 group with 107 patients with positive serology and incipient heart failure. The used data comes from 24-hour ECG, the RR intervals from each subject was divided in 288 frames of 5 minutes each. Then it was preprocessed using permutation entropy obtaining the circadian profile for each patient. And by applying PCA each patient ended up represented by a vector of 144 entries. This was in turn used for the training of the proposed NN architecture. The classification performed with 91% accuracy and an average of 92% precision, consisting in a great work of classification validated by the AUC in each ROC curve. As this results were obtained with a limited quantity of data, this study can be improved provided with more samples, making this model a tool for analyzing ECG in order to try to do an early evaluation and diagnosis of a cardiac compromise related to the generally silent chronic phase.