ABSTRACT
Fever is one of the major clinical symptoms of undifferentiated fever cases. Early diagnosis of undifferentiated fever is a challenging task for the physician. The aim of this study was to classify infectious and noninfectious diseases from 24-hour continuous tympanic temperature recordings of patients with undifferentiated fever using a machine learning algorithm (artificial neural network). This was an observational study conducted in 103 patients who presented with undifferentiated fever. Twenty-four-hour continuous tympanic temperature was recorded from each patient. Features were extracted from temperature signals and classified into infectious and noninfectious diseases using an artificial neural network (ANN). The ANN classifier provided the highest accuracy at 91.3% for differentiating infectious and noninfectious diseases from undifferentiated fever cases. Significant kappa agreement (κ = 0.777) was found between the final diagnosis as determined by the physician and the classification obtained using an ANN classifier. Based on our results, we conclude that the continuous 24-hour temperature monitoring and application of an ANN classifier provides a simple noninvasive and inexpensive supplementary diagnostic method to differentiate infectious and noninfectious diseases.
Subject(s)
Algorithms , Body Temperature , Communicable Diseases/classification , Fever/diagnosis , Monitoring, Physiologic/methods , Neural Networks, Computer , Noncommunicable Diseases/classification , Adult , Circadian Rhythm , Communicable Diseases/diagnosis , Diagnosis, Differential , Ear, Middle , Female , Health Records, Personal , Humans , Machine Learning , Male , Middle AgedABSTRACT
Body temperature is a continuous physiological variable. In normal healthy adults, oral temperature is estimated to vary between 36.1°C and 37.2°C. Fever is a complex host response to many external and internal agents and is a potential contributor to many clinical conditions. Despite being one of the foremost vital signs, temperature and its analysis and variations during many pathological conditions has yet to be examined in detail using mathematical techniques. Classical fever patterns based on recordings obtained every 8-12 h have been developed. However, such patterns do not provide meaningful information in diagnosing diseases. Because fever is a host response, it is likely that there could be a unique response to specific etiologies. Continuous long-term temperature monitoring and pattern analysis using specific analytical methods developed in engineering and physics could aid in revealing unique fever responses of hosts and in different clinical conditions. Furthermore, such analysis can potentially be used as a novel diagnostic tool and to study the effect of pharmaceutical agents and other therapeutic protocols. Thus, the goal of our article is to present a comprehensive review of the recent relevant literature and analyze the current state of research regarding temperature variations in the human body.
