基于被动听觉ERP范式的慢性意识障碍患者个体评估研究  

Individual-Level Assessment in Patients with Disorders of Consciousness under Passive Auditory ERP Paradigm

作　　者：王小宇 杨艺 李凡 陈雪玲 高寒冰 马兆楠 何江弘 丛丰裕 Wang Xiaoyu;Yang Yi;Li Fan;Chen Xueling;Gao Hanbing;Ma Zhaonan;He Jianghong;Cong Fengyu(School of Biomedical Engineering,Faculty of Electronic Information and Electrical Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China;Department of Neurosurgery,Beijing Tiantan Hospital Affiliated to Capital Medical University,Beijing100700,China)

机构地区：[1]大连理工大学电子信息与电气工程学部生物医学工程学院,辽宁大连116024 [2]首都医科大学附属北京天坛医院神经外科,北京100700

出　　处：《中国生物医学工程学报》2022年第2期129-139,共11页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金(91748105);国家自然科学基金青年科学基金项目(81600919);北京市科技新星计划(Z181100006218050)。

摘　　要：近年来,大脑功能检测技术逐渐应用于意识障碍(DoC)患者的意识水平评估,以期降低仅依靠临床主观行为量表评估的误诊率。其中,被动听觉事件相关电位(ERPs)范式下诱发的N1和失匹配负波(MMN)成分被认为是反应DoC患者意识水平的关键脑功能指标,但目前仍局限于组别层面的统计分析结果,其在DoC患者个体水平的评估效能尚待挖掘。根据N1和MMN成分的特性,提出了一种适用于被动听觉ERPs范式下DoC患者个体评估的深度学习算法。该方法引入了特征融合的数据扩增策略,即在单被试数据中随机抽取不同类型的试次进行空间级联形成新的融合样本,并结合EEGNET深度学习算法实现自动化特征提取和分类识别。在包含132名被试的单试次数据集上进行测试。对健康被试(38名),微意识状态(40名)和植物状态(54名)患者进行三分类评估。统计分析结果表明,该样本融合策略可显著提升模型分类表现,并在单被试融合样本数量扩增至1 000时模型取得样本水平75.14%的平均分类准确率,以及被试水平83.00%的平均分类准确率,83.79%的精确率和84.02%的召回率。所提出的深度学习框架可有效克服传统方法中存在的个体评估能力不强的问题,为DoC患者个体评估提供新的研究思路。In recent years, there have been increased efforts to assess the levels of consciousness in patients with disorders of consciousness(DoC) using neuroimaging techniques, aiming to improve diagnosis and identification beyond current subjective behavioral assessments that suffer from high misdiagnosis rates. Previous evidence suggests that N1 and mismatch negativity(MMN) components elicited by passive auditory event-related potentials(ERPs) paradigms are critical neurophysiological markers of DoC. However, as such evidence is limited to group-level analysis, the extent to which they enable residual consciousness detection at the individual-level is unclear. Considering the characteristics of N1 and MMN components, we proposed a deep learning algorithm for the individual assessment of patients with DoC under a passive auditory ERPs paradigm. The algorithm proposed a data augmentation strategy, which randomly fused single-trials elicited by different types of stimuli in the spatial domain to form fusion samples, and a deep learning classifier, known as EEGNET, to achieve automatic feature extraction and classification. The proposed method was evaluated in a three-class classification task(38 healthy controls, 40 minimally conscious state, and 54 vegetative state patients) using a single-trial dataset including 132 subjects. Statistical results showed that the proposed data augmentation method significantly improved the classification performance in the current task, and it achieved the highest 75.14% mean classification accuracies in sample level as well as 83.00% mean classification accuracies, 83.79% precision rate, and 84.02% recall rate in subject level when the number of single-subject samples was augmented to 1000. In conclusion, the proposed method could overcome the drawbacks of poor assessment performance in the conventional individual-level assessment methods, providing a new strategy for individual-level assessment in patients with DoC.

关 键 词：意识障碍 事件相关电位 失匹配负波 深度学习 数据扩增 

分 类 号：R318[医药卫生—生物医学工程]