Applications of Transductive Spectral Clustering Methods in a Military Medical Concussion Database

Abstract

Traumatic brain injury (TBI) is one of the most common forms of neurotrauma that has affected more than 250,000 military service members over the last decade alone. While in battle, service members who experience TBI are at significant risk for the development of normal TBI symptoms, as well as risk for the development of psychological disorders such as Post-Traumatic Stress Disorder (PTSD). As such, these service members often require intense bouts of medication and therapy in order to resume full return-to-duty status. The primary aim of this study is to identify the relationship between the administration of specific medications and reductions in symptomology such as headaches, dizziness, or light-headedness. Service members diagnosed with mTBI and seen at the Concussion Restoration Care Center (CRCC) in Afghanistan were analyzed according to prescribed medications and symptomology. Here, we demonstrate that in such situations with sparse labels and small feature sets, classic analytic techniques such as logistic regression, support vector machines, naïve Bayes, random forest, decision trees, and k-nearest neighbor are not well suited for the prediction of outcomes. We attribute our findings to several issues inherent to this problem setting and discuss several advantages of spectral graph methods.

A decision tree framework for understanding blast-induced mild Traumatic Brain Injury in a military medical database

“…7,8 For example, we recently developed a graph theoretical model to predict an increase/decrease in symptomology based on medications prescribed prior to injury and neuropsychological testing conducted upon a patient’s visit to a clinic after injury. 9 In this work, we showed that such a model can reliably outperform other parametrically and/or machine-learning-based models in terms of predicting patient outcomes.…”

“…Using CSP, we previously showed how a graph-based representation can be more effective in terms of classification accuracy for a small (sparsely labeled) dataset. 9 However, this work was limited in that the graph-based structure failed to identify how information propagates from one region of the graph (i.e., cluster of nodes) to another separate region. Therefore, in this paper we sought to extend our previous work by demonstrating the manner in which CSP classifies symptomology.…”

Military Applications of Machine Learning: A Bibliometric Perspective

“…The models used in military training are based on data processing and are becoming increasingly common in military training [68]. In relation to body negative pressure, we found evidence from algorithm-supported studies on military data that focus on military medical studies [67].…”

“…In this category, we include the themes body negative pressure and training. In relation to body negative pressure, we found evidence from algorithm-supported studies on military data that focus on military medical studies [67].…”