Continuous, automated, electronic patient vital signs data are important to physicians in evaluating traumatic brain injury (TBI) patients' physiological status and reaching timely decisions for therapeutic interventions. However, missing values in the medical data streams hinder applying many standard statistical or machine learning algorithms and result in losing some episodes of clinical importance. In this paper, we present a novel approach to filling missing values in streams of vital signs data. We construct sequences of Hankel matrices from vital signs data streams, find that these matrices exhibit low-rank, and utilize low-rank matrix completion methods from compressible sensing to fill in the missing data. We demonstrate that our approach always substantially outperforms other popular fill-in methods, like k-nearest-neighbors and expectation maximization. Further, we show that our approach recovers thousands of simulated missing data for intracranial pressure, a critical stream of measurements for guiding clinical interventions and monitoring traumatic brain injuries.
N-FINDR suffers from several issues in its practical implementation. One is the search region which is usually the entire data space. Another related issue is its excessive computation. A third issue is the use of random initial conditions which causes inconsistency in final results that can not be reproducible. This paper develops two ways to speed up the N-FINDR in computation. One is to narrow down the search region for the N-FINDR to a feasible range, called region of interest (ROI) where data sphering/thresholding and the well-known pixel purity index (PPI) are used as a preprocessing to find a desire ROI. The other is to simplify the simplex volume computation where three methods are proposed for this purpose to reduce computational complexity of matrix determinant. In addition, in order to further reduce computational complexity two sequential N-FINDR algorithms are developed which implement the N-FINDR by finding one endmember after another in sequence so that the information provided by previously found endmembers can be used to reduce computational complexity. The conducted experiments demonstrate that while the proposed fast algorithms can greatly reduce computational complexity, their performance remains as good as the N-FINDR is and is not compromised by reduction of the search region to an ROI and simplified matrix determinant.
This project explored the use of advanced machine learning techniques to noninvasively estimate real-time intracranial pressure (ICP) in traumatic brain injury patients from continuous electronic physiological monitoring data. We hypothesized that advanced machine learning techniques could process and analyze electronic vital signs data collected noninvasively and, when correlated with central nervous system-invasive ICP monitoring data, could provide a valid analytic platform for the noninvasive monitoring of ICP for defined periods and clinical needs. We developed a machine learning algorithm that, using noninvasive vital signs features alone, could estimate the current ICP of a patient with an accuracy of + or - 4.6 mmHg. We further developed algorithms that are capable of predicting patient future ICP with + or - 1.5 mmHg 5 minutes into the future and + or - 5 mmHg standard deviation 2 hours into the future using the continuous recordings of heart rate, systolic blood pressure, mean arterial pressure, and ICP. Realization of this analytic platform will support translation of this work into robust, field-ready clinical instrumentation that permits high-quality ICP monitoring in austere care environments not suitable for central nervous system-invasive ICP monitoring.