Sistema electrónico de monitoreo de bioseñales para el diagnóstico médico de COVID-19 en personas mediante inteligencia artificial

Abstract

The present research proposes the implementation of an electronic system for the monitoring of vital signs and diagnosis of COVID-19 in patients by means of artificial intelligence. To this end, bibliographic research in medical databases has been used to collect information about the symptoms that occur in patients with COVID-19, this point is important because the artificial intelligence algorithm is more efficient when there is enough data to learn. The dataset is composed of labeled data which are "Ambulatory", "Medical review" and "Hospitalization", so it is chosen to use supervised learning classification algorithms. A data exploration is performed, and then with the vital signs that are acquired from the MAX30100 and MLX90614 sensors, to train and choose the most efficient algorithm. The algorithms used are Support Vector Machines, Decision Trees and Multinomial Naive Bayes. All this process is performed on a Raspberry Pi Zero taking advantage of its resources and dimension suitable for a bracelet, the data are stored in a local database and sent to the cloud via HTTP requests, in order that the treating physician is kept informed of the health status of their patients, so you can also receive alert emails indicating the level of risk of the same. Using the evaluation metrics for the artificial intelligence algorithms, it was determined that the Support Vector Machine algorithm is the one that best fits in the classification of the 3 categories, having as results for the "Ambulatory" class a "Recall" equal to 0.75 and a precision of 0.90, for the "Medical Review" class 0.92 and 0.83 and for the "Hospitalization" class 1.00 and 0.95, concluding that the algorithm efficiently handles the classification of the 3 classes with respect to the other algorithms.