Machine learning techniques for performance prediction of medical devices: infant incubators

Abstract

This paper presents development of Expert System for prediction of performance of infant incubators based on real-time measured data. Temperature error, preventive maintenance intervals, number of additional parts and utilization coefficient were used as input information for the development of this system. Expert system is based on Artificial Neural Network (ANN) and Fuzzy logic (FL) classifier. Feed forward back-propagation artificial neural network with 12 neurons in hidden layer and sigmoid transfer function, using Bayesian regulation algorithm has shown best properties for prediction of the functionality of incubators based on performance output error. Fuzzy logic using Mamdani implication logic was developed as an extension of ANN and finally used for prediction of device performance. The developed expert system presented in this paper presents the first step in researching possibilities of usage such systems for upgrading medical device management strategies in healthcare institutions to answer challenges of increased sophistication of devices, but patient safety demands as well.