Utilization of temporal information for intracranial pressure development trend forecasting in traumatic brain injury

• Published in: 2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society Vol. 2012; pp. 3930 - 3934
• Main Authors: Mengling Feng, Zhuo Zhang, Cuntai Guan, Hardoon, D. R., King, N. K. K., Boon Chuan Pang, Beng Ti Ang
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.01.2012
• Subjects: Algorithms; Brain Injuries - physiopathology; Forecasting; Humans; Intracranial Pressure - physiology; Iterative closest point algorithm; Machine learning algorithms; Monitoring; Monitoring, Physiologic; Predictive models; Time Factors; Vectors
• ISBN: 1424441196; 9781424441198
• ISSN: 1094-687X; 1557-170X; 1558-4615
• DOI: 10.1109/EMBC.2012.6346826

Objective. Our primary objective is to demonstrate and statistically justify that forecasting models that utilize temporal information of the historical readings of ICP and related parameters are superior, in terms of performance, compared with models that do not make use of temporal information. Material & Method. 82 traumatic brain injuries patients, who were admitted between 2002 to 2007 and were continuously monitored on ICP for more than 24 hours, are selected for the study. Together with ICP, MAP and PbtO 2 were also monitored, and PRx was calculated as a moving correlation between ICP and MAP. The development trends of ICP and the related parameters are measured by first segmenting the time-series data into multiple periodic windows. The development trend of each periodic window is then discretized into three classes - elevate, stay or reduce - based on the concept of "trend line". A systematic framework is developed to compare the forecast performance between the temporal and non-temporal models. Findings. Experimental results demonstrate that the utilization of temporal information directly leads to a considerable boost in trend forecasting performance (on average 20% relative performance gain was achieved). Moreover, the performance gain is confirmed to be statistically significant (p-value <; 0.0001) based on a paired t-test.