1. Predicting Motor and Cognitive Improvement Through Machine Learning Algorithm in Human Subject that Underwent a Rehabilitation Treatment in the Early Stage of Stroke.
- Author
-
Sale, Patrizio, Ferriero, Giorgio, Ciabattoni, Lucio, Cortese, Anna Maria, Ferracuti, Francesco, Romeo, Luca, Piccione, Francesco, and Masiero, Stefano
- Abstract
Abstract Background The objective of this study was to investigate, in subject with stroke, the exact role as prognostic factor of common inflammatory biomarkers and other markers in predicting motor and/or cognitive improvement after rehabilitation treatment from early stage of stroke. Methods In this longitudinal cohort study on stroke patients undergoing inpatient rehabilitation, data from 55 participants were analyzed. Functional and clinical data were collected after admission to the rehabilitation unit. Biochemical and hematological parameters were obtained from peripheral venous blood samples on all individuals who participated in the study within 24 hours from the admission at the rehabilitative treatment. Data regarding the health status were collected at the end of rehabilitative treatment. First, a feature selection has been performed to estimate the mutual dependence between input and output variables. More specifically, the so called Mutual Information criterion has been exploited. In the second stage of the analysis, the Support Vector Machines (SVMs), a non-probabilistic binary machine learning algorithm widely used for classification and regression, has been used to predict the output of the rehabilitation process. Performances of the linear SVM regression algorithm have been evaluated considering a different number of input features (ranging from 4 to 14). The performance evaluation of the model proposed has been investigated in terms of correlation, Root Mean Square Error (RMSE) and Mean Absolute Deviation Percentage (MADP). Results Results on the test samples show a good correlation between all the predicted and measured outputs (i.e. T1 Barthel Index (BI), T1 Motor Functional Independence Measure (FIM), T1 Cognitive FIM and T1 Total FIM) ranging from 0.75 to 0.81. While the MADP is high (i.e., 83.96%) for T1 BI, the other predicted responses (i.e., T1 Motor FIM, T1 Cognitive FIM, T1 Total FIM) disclose a smaller MADP of 30%. Accordingly, the RMSE ranges from 4.28 for T1 Cognitive FIM to 22.6 for T1 BI. Conclusions In conclusion, the authors developed a new predictive model using SVM regression starting from common inflammatory biomarkers and other ratio markers. The main efforts of our model have been accomplished in regard to the evidence that the type of stroke has not shown itself to be a critical input variable to predict the discharge data, furthermore, among the four selected indicators, Barthel at T1 is the less predictable (MADP > 80%), while it is possible to predict T1 Cognitive FIM with an MADP less than 18%. [ABSTRACT FROM AUTHOR]
- Published
- 2018
- Full Text
- View/download PDF