Abstract

--- - "Background: In physical therapy, postural tests are frequently used to diagnose neuropathies, particularly in diabetic individuals. This study aims to develop a method based on the analysis of time series that allows discriminating between healthy and diabetic subjects with or without a neuropathic condition." - "Research question: Do features obtained from time series corresponding to postural tests allow us to reliably discriminate between healthy, diabetic and neuropathic patients?" - "Methods: In this study, 32 people participated in the healthy, diabetic, and neuropathic categories (11, 9, and 12, respectively). The data was collected by positioning each participant on a Wii Balanced Board platform, under 8 different conditions. The analyzed time series are sensed by devices that capture variations in the subject's center of pressure when subjected to a test on different conditions over a short period of time. The method proposed considers statistical techniques used for characterizing the time series combined with machine learning techniques to classify the individual's profile into one of the three categories mentioned. The classification is supported by an underlying probabilistic model, based on the characteristics of the time series, generating average curves for each class, which are then used by the classification methods." - "Results: The empirical results include classification models for each class, obtaining a performance (F-score) over 98%. In addition, other models considering the particular conditions to which the subject is exposed during the test are developed, revealing that the conditions of eyes open and eyes closed show the highest levels of discrimination to classify participants into one of the three class categories." - "Significance: These results suggest a test protocol simplification and, at the same time, that the proposed method based on the analysis of the time series associated with the test used is highly predictive and may reliably complement or substitute a questionnaire-based diagnosis."