Neil S. Rothman

Classification of Traumatic Brain Injury Severity Using Informed Data Reduction in a Series of Binary Classifier Algorithms

Assessment of medical disorders is often aided by objective diagnostic tests which can lead to early intervention and appropriate treatment. In the case of brain dysfunction caused by head injury, there is an urgent need for quantitative evaluation methods to aid in acute triage of those subjects who have sustained traumatic brain injury (TBI). Current clinical tools to detect mild TBI (mTBI/concussion) are limited to subjective reports of symptoms and short neurocognitive batteries, offering little objective evidence for clinical decisions; or computed tomography (CT) scans, with radiation-risk, that are most often negative in mTBI. This paper describes a novel methodology for the development of algorithms to provide multi-class classification in a substantial population of brain injured subjects, across a broad age range and representative subpopulations. The method is based on age-regressed quantitative features (linear and nonlinear) extracted from brain electrical activity recorded from a limited montage of scalp electrodes. These features are used as input to a unique “informed data reduction” method, maximizing confidence of prospective validation and minimizing over-fitting. A training set for supervised learning was used, including: “normal control,” “concussed,” and “structural injury/CT positive (CT+).” The classifier function separating CT+ from the other groups demonstrated a sensitivity of 96% and specificity of 78%; the classifier separating “normal controls” from the other groups demonstrated a sensitivity of 81% and specificity of 74%, suggesting high utility of such classifiers in acute clinical settings. The use of a sequence of classifiers where the desired risk can be stratified further supports clinical utility.

Quantitative EEG in babies at risk for hypoxic ischemic encephalopathy after perinatal asphyxia

Objective:To evaluate an electroencephalography (EEG)-based index, the Cerebral Health Index in babies (CHI/b), for identification of neonates with high Sarnat scores and abnormal EEG as markers of hypoxic ischemic encephalopathy (HIE) after perinatal asphyxia.Study Design:This is a retrospective study using 30 min of EEG data collected from 20 term neonates with HIE and 20 neurologically normal neonates. The HIE diagnosis was made on clinical grounds based on history and examination findings. The maximum-modified clinical Sarnat score was used to grade HIE severity within 72 h of life. All neonates underwent 2-channel bedside EEG monitoring. A trained electroencephalographer blinded to clinical data visually classified each EEG as normal, mild or severely abnormal. The CHI/b was trained using data from Channel 1 and tested on Channel 2.Result:The CHI/b distinguished among HIE and controls (P<0.02) and among the three visually interpreted EEG categories (P<0.0002). It showed a sensitivity of 82.4% and specificity of 100% in detecting high grades of neonatal encephalopathy (Sarnat 2 and 3), with an area under the receiver operator characteristic (ROC) curve of 0.912. CHI/b also identified differences between normal vs mildly abnormal (P<0.005), mild vs severely abnormal (P<0.01) and normal vs severe (P<0.002) EEG groups. An ROC curve analysis showed that the optimal ability of CHI/b to discriminate poor outcome was 89.7% (sensitivity: 87.5%; specificity: 82.4%).Conclusion:The CHI/b identified neonates with high Sarnat scores and abnormal EEG. These results support its potential as an objective indicator of neurological injury in infants with HIE.