KINEMATIC APPROACH APPLIED TO ARTIFICIAL NEURAL NETWORKS FOR EARLY SEPSIS DETECTION
Early detection; Sepsis; Kinematics; Artificial neural networks; Vital signs.
Sepsis is a severe disease that affects millions of people around the world, and its early detection is fundamental to improve the treatment effectiveness. Recently, several models have been proposed to classify sepsis-positive patients in advance or to identify the probability of the disease occurrence in the future. In both cases, the data input is usually composed of time series of vital signs or other clinical variables. The current research shows an innovative approach for early detection of sepsis by representing a patient as a moving particle in an N-dimensional space, where N is the number of the adopted vital signs. A Sepsis Point is established, which corresponds to the position occupied by a patient if he became positive for the disease. The position, velocity, and acceleration vectors of the patients relative to the Sepsis Point are calculated. These vectors are used to generate the Kinematic Variables, which are imputed in artificial neural network models for early detection of sepsis. The accuracies achieved by the Kinematic Approach were compared to the accuracies achieved by the same models using traditional vital signs as input. It was discovered that the Kinematic Approach resulted in greater accuracy models, proving this research’s hypothesis. Thus, the Kinematic Approach is expected to open new approaches for developing more accurate early detection sepsis models.