Fault diagnostic of variance shifts in clinical monitoring using an Artificial Neural Network Input Gain Measurement Approximation (ANNIGMA)

© 2018, Springer International Publishing AG, part of Springer Nature. Condition of a patient in an intensive care unit is assessed by monitoring multiple correlated variables with individual observations. Individual monitoring of variables leads to misdiagnosis. Therefore, variability of the correlated variables needs to be monitored simultaneously by deploying a multivariate control chart. Once the shift from the accepted range is detected, it is vital to identify the variables that are responsible for the variance shift detected by the chart. This will aid the medical practitioners to take the appropriate medical intervention to adjust the condition of the patient. In this paper, Multivariate Exponentially Weighted Moving Variance chart has been used as the variance shift identifier. Once the shift is detected, authors for the first time have used ANNIGMA to identify the variables responsible for variance shifts in the condition of the patient and rank the responsible variables in terms of the percentage of their contribution to the variance shift. The performance of the proposed ANNIGMA has been measured by computing average classification accuracy. A case study based on real data collected from ICU unit shows that ANNIGMA not only improve the diagnosis but also speed up the variable identification for the purpose of appropriate medical diagnosis.