The application of eye movement biometrics in the automated detection of mild traumatic brain injury

Abstract

This paper presents a pilot study for the automated detection of mild traumatic brain injury (mTBI) via the application of eye movement biometrics. Biometric feature vectors from multiple paradigms are evaluated for their ability to differentiate subjects diagnosed with mTBI from healthy subjects within a small subject pool. Supervised and unsupervised machine learning techniques were applied to the problem, with preliminary results indicating a potential 100% classification accuracy from a supervised learning technique and 89% classification accuracy from an unsupervised technique.According to reports from the Centers for Disease Control and Prevention [4], approximately 1.7 million people are diagnosed with TBI each year in the United States, of which nearly 75% (or 1.3 million) are incidences of mTBI. This does not account for the undiagnosed occurrences of mTBI that are thought to exceed 25% of the reported figure, or nearly 425,000 undiagnosed cases per year.Each year there are approximately 52,000 TBI-related deaths in the United States, accounting for roughly one-third (30.5%) of all injury-related deaths [4]. mTBI increases the risk of TBI [5], and can cause neurological disorders which persist years after injury [16], affecting thought, behavior, and emotion, producing physical symptoms such as fatigue, nausea, vertigo, headache, lethargy, and blurred vision [11].

Research Trends in Artificial Intelligence Applications in Human Factors Health Care: Mapping Review

“…Mathematical/video data To detect mild traumatic brain injury (mTBI) via the application of eye movement biometrics Komogortsev and Holland [39] tion scores ≤ -0.870 and ≥0.79 as having mTBI, respectively To measure the feasibility of artificial neural networks in analyzing nurses' burnout process Ladstatter et al [41] Artificial neural networks identified a strong personality as one of the leading causes of nursing burnout; it produced a 15% better result than traditional statistical instruments 462 nurses Survey study: Nursing Burnout Scale Short Form To assess whether artificial neural networks offer better predictive accuracy in identifying nursing burnouts than traditional statistical techniques Ladstatter et al [42] Machine learning-based itchtector algorithm detected scratch movement more accurately when patients wore it for a longer duration 40 patients and 2 dermatologists Interview study: user experience and acceptance of the device To determine how wearable devices can help people manage their itching conditions Lee et al [43] Naïve Bayes Kernel resulted in the highest classification accuracy; it identified a higher proportion of medication errors and a lower proportion of procedural error than manual screening To augment the relationship between physical therapists and their patients recovering from a knee injury, using a wearable sensing device Muñoz et al [49] The machine-learning model accurately identified 70% of suicidality when compared to the default accuracy (56%) of a classifier that predicts the most prevalent class 26 patients Survey study: evaluating psychology students' communication habits using electronic services To identify periods of suicidality Nobles et al [50] Naïve Bayes and support vector machine correctly identified handover and patient identification incidents with an accuracy of 86.29%-91.53% and 97.98%, respectively To develop an efficient patientemotional classification computational algorithm in interaction with nursing robots in medical care Swangnetr and Kaber [59] Nearly half of the total comments analyzed described positive care experiences. Most negative experiences concerned a lack of posttreatment care and insufficient information concerning self-management strategies or treatment side effects NA Survey study regarding treatment, disease status, physical activity, functional assessment of cancer therapy, and social difficulties inventory To analyze the patient experience of care and its effect on health-related quality of life Wagland et al [60] After pharmacist review, only 17% of algorithm-identified patients were considered potentially undertreated NA (data from 14 primary care clinics)…”

Biometric Recognition via Eye Movements

“…Finally, because eye movements are a product of the human visual system that engages a variety of brain regions, manifestation of pathologies occurring in those regions can be detected in the features extracted from eye movements. An example involves the use of eye movements for the detection of mild traumatic brain injuries [Cifu et al 2015; Komogortsev and Holland 2014]. In particular, Shadmehr et al [2010] has suggested that the speed with which the eyes move during a saccade (i.e., saccade vigor) may be a reflection of between-subject differences in patterns of decision-making in health and disease.…”

Neurological Disorder Identification by Eye Movement Biometric using Machine Learning Schemes

“…Similar process has been employed for saccades. Based on obtained fixations and saccades, the four quantitative features (fixation count, fixation duration, vectorial saccade amplitude are fixation quantitative score) are generated [4], [10]. Fixation count is the total number of fixations.…”

“…They established a standardized protocol to differentiate between m-TBI and normal subjects. In recent research [10], it has been strongly conveyed that eye movement biometric has great impact in automated detection of m-TBI. In order to identify the m-TBI, they used two algorithms; one unsupervised technique involved a heuristic method and other was supervised support vector regression with radial basis function.…”