Abstract:
Accurate, and objective diagnosis of brain injury remains challenging. This study evaluated useability and reliability of computerized eye-tracker assessments (CEAs) designed to assess oculomotor function, visual attention/processing, and selective attention in recent mild traumatic brain injury (mTBI), persistent post-concussion syndrome (PPCS), and controls. Tests included egocentric localisation, fixation-stability, smooth-pursuit, saccades, Stroop, and the vestibulo-ocular reflex (VOR). Thirty-five healthy adults performed the CEA battery twice to assess useability and test-retest reliability. In separate experiments, CEA data from 55 healthy, 20 mTBI, and 40 PPCS adults were used to train a machine learning model to categorize participants into control, mTBI, or PPCS classes. Intraclass correlation coefficients demonstrated moderate (ICC > .50) to excellent (ICC > .98) reliability (p < .05) and satisfactory CEA compliance. Machine learning modelling categorizing participants into groups of control, mTBI, and PPCS performed reasonably (balanced accuracy control: 0.83, mTBI: 0.66, and PPCS: 0.76, AUC-ROC: 0.82). Key outcomes were the VOR (gaze stability), fixation (vertical error), and pursuit (total error, vertical gain, and number of saccades). The CEA battery was reliable and able to differentiate healthy, mTBI, and PPCS patients reasonably well. While promising, the diagnostic model accuracy should be improved with a larger training dataset before use in clinical environments.
摘要:
准确,脑损伤的客观诊断仍然具有挑战性。这项研究评估了旨在评估动眼功能的计算机眼动仪评估(CEA)的可用性和可靠性,视觉注意/处理,以及最近轻度创伤性脑损伤(mTBI)的选择性注意,持续性脑震荡后综合征(PPCS),和控制。测试包括自我中心定位,固定稳定性,顺利的追求,扫视,Stroop,和前庭眼反射(VOR)。35名健康成年人对CEA电池进行了两次测试,以评估可用性和重测可靠性。在单独的实验中,来自55个健康的CEA数据,20mTBI,40名PPCS成年人被用来训练机器学习模型,将参与者分为控制区,mTBI,或PPCS类。组内相关系数显示出中等(ICC>.50)至出色(ICC>.98)的可靠性(p<.05)和令人满意的CEA合规性。机器学习建模将参与者分为控制组,mTBI,和PPCS执行合理(平衡精度控制:0.83,mTBI:0.66和PPCS:0.76,AUC-ROC:0.82)。关键结果是VOR(凝视稳定性),固定(垂直误差),和追求(总误差,垂直增益,和扫视次数)。CEA电池可靠,能够区分健康,mTBI,和PPCS患者相当好。虽然有希望,在用于临床环境之前,应通过更大的训练数据集来提高诊断模型的准确性.