基于面孔识别的脑机接口技术在意识障碍中的应用

发布时间：2018-05-14 17:44

本文选题：BCI技术 + 面孔识别　；参考：《广州中医药大学》2017年硕士论文

【摘要】：目的:1.通过运用BCI检测意识障碍(Disorders of consciousness,DOC)患者意识水平,将所得在线准确率与随访获得的格拉斯哥结局量表(Glasgow outcome scale,GOS)评分进行相关性分析,判断本研究所运用的基于面孔识别的脑机接口(Brain-computer interface,BCI)技术对DOC患者预后评估的价值;2.通过分别做基于面孔识别的BCI在线准确率结果及CRS-R评分结果与GOS评分等级的相关性分析,探讨BCI在DOC患者预后评估中是否较CRS-R评分更为准确、客观,能否作为行为学量表补充,提高诊断的准确性。方法:对23名DOC患者进行基于面孔识别的BCI检测,每个患者每周检测两次,共进行5次检测;每次检测包括两个部分,即训练部分和测试部分,每部分包含10次小测试,每一次的小测试中相片以随机的方式进行闪烁,将5次测试所得在线准确率进行平均得出最终在线准确率,检测结束3个月后或是发病6个月后对患者进行电话随访,行GOS评分。第一部分以GOS评分等级为结局指标,做二分类的Logistic回归分析。第二部分在测试前或测试后24h内选取患者状态最佳的时刻予以行CRS-R评分。将BCI在线准确率结果及检测前后的CRS-R评分与随访获得的GOS评分分别进行spearman相关性分析,P0.05时认为与GOS评分显著相关。再将BCI及CRS-R评分共同纳入建立Logistic回归模型,判断二者联合是否能更准确的评估预后。结果:第一部分做BCI在线准确率与GOS评分等级的Logistic回归分析,得出相关系数为 8.97,对应的P=0.0130.05,Hosmer-Lemeshow 检验结果 χ2为 0.591,对应的P=0.988。BCI对良好结局的预测准确率为75%,对较差结局的预测准确率为93.3%,对DOC患者总预后的预测准确率为84%。做出BCI在线准确率对预后结局预测概率的ROC曲线,曲线下面积为0.925,95%置信区间为[0.806,0.966],提示BCI在线准确率可以较好地评估DOC患者的预后。第二部分将入院时及BCI检测前后24h内的CRS-R评分分别与GOS评分做Spearman相关性分析,入院时的CRS-R与GOS相关系数r值为0.354,对应的P=0.030.05;检测时的CRS-R评分与GOS评分相关系数r为0.505,对应的P=0.0140.05。CRS-R评分与GOS评分显著相关。BCI在线准确率与GOS评分做Spearman相关性分析,相关系数r为0.638,P=0.0010.05,二者显著相关。通过比较相关系数的大小,可得出BCI在线准确率比CRS-R评分与GOS评分相关性更强。再做BCI在线准确率及CRS-R与GOS评分等级的Logistic回归分析,BCI在线准确率对应的回归系数为6.301,P= 0.030.05,CRS-R评分的回归系数为 1.788,P= 0.0410.05。Hosmer-Lemeshow 检验结果 χ2 为 3.146,P= 0.925。二者结合的模型对良好预后的预测准确率为93%,对较差预后的预测准确率为80.7%,对DOC患者预后结局总预测准确率为87%。ROC曲线下的面积为0.943。与第一部分得出BCI单独预测DOC患者预后结局的准确率相比较,二者联合准确率更高。结论:1.基于面孔识别的BCI能够从脑反应的角度更为客观地检测DOC患者意识水平,并且能够很好地评估患者预后,在线准确率能够作为反映DOC患者脑功能及预后的一项客观指标,准确率达0.64以上的患者意识恢复的可能性较大;2.在DOC患者意识水平诊断及预后评估方面基于面孔识别的多模态脑机接口技术比目前临床常用的CRS-R量表准确性更高,二者结合能够更好地评估DOC患者的预后,BCI技术能够作为临床行为学量表的补充。
[Abstract]:Objective: 1. through the use of BCI to detect the consciousness level of Disorders of consciousness (DOC), the correlation analysis between the online accuracy rate and the Glasgow Outcome Scale (Glasgow outcome scale, GOS) obtained from the follow-up was carried out to determine the face recognition based brain machine interface (Brain-computer interface) used in this study (Brain-computer interface). The value of BCI) technology for evaluating the prognosis of DOC patients; 2. through the correlation analysis of the BCI online accuracy results based on face recognition and the correlation between the CRS-R score and the GOS grade, it is discussed whether BCI is more accurate than the CRS-R score in the prognosis assessment of DOC patients. It is objective, whether it can be supplemented by the behavioral scale, and improve the accuracy of the diagnosis. Methods: 23 DOC patients were detected by face recognition based BCI test. Each patient was detected two times a week, with a total of 5 tests. Each test included two parts, the training part and the test section, each part contained 10 small tests. Each of the small tests was flickered with the machine. The online accuracy rate of the 5 tests was obtained. On average, the final online accuracy was obtained. 3 months after the end of the test or 6 months after the onset of the disease, the patients were followed up by telephone, and the GOS score was performed. The first part took the GOS score grade as the outcome index, and did the two classification of Logistic regression analysis. The second part took the CRS-R evaluation before or after the test of the best state of the patient. The results of BCI online accuracy, the CRS-R score before and after detection and the GOS score of the follow-up were analyzed with Spearman correlation respectively. P0.05 was considered to be significantly related to the GOS score. Then BCI and CRS-R scores were incorporated into the Logistic regression model to determine whether the combination of the two would be more accurate to evaluate the prognosis. The first part was B. The Logistic regression analysis of CI online accuracy and GOS grade showed that the correlation coefficient was 8.97, the corresponding P=0.0130.05, the Hosmer-Lemeshow test result chi 2 was 0.591, the corresponding P=0.988.BCI for the good outcome was 75%, the prediction accuracy of the poor outcome was 93.3%, and the prediction accuracy for the total prognosis of DOC patients was 84%.. To make the ROC curve of the prediction probability of BCI online accuracy on prognosis, the area under the curve is 0.925,95% confidence interval [0.806,0.966], suggesting that the BCI online accuracy can better evaluate the prognosis of DOC patients. The second part will be admitted to the hospital and the CRS-R score in 24h before and after BCI detection and GOS score for Spearman correlation analysis, admission to hospital. The correlation coefficient r value of CRS-R and GOS was 0.354, corresponding P=0.030.05, and the correlation coefficient r of CRS-R score and GOS score was 0.505, and the corresponding P=0.0140.05.CRS-R score and GOS score were significantly related to Spearman correlation analysis of.BCI online accuracy and GOS score, and the correlation coefficient was 0.638, and the comparison was significant. Through comparison, the correlation coefficient was significantly correlated. The correlation coefficient can be found that the BCI online accuracy rate is more correlated with the GOS score than the CRS-R score. In the Logistic regression analysis of BCI online accuracy and CRS-R and GOS grade, the regression coefficient corresponding to the BCI online accuracy is 6.301, P= 0.030.05, and CRS-R score is 1.788. The predictive accuracy of the combined model of P= 0.925. two for good prognosis was 93%, the prediction accuracy for poor prognosis was 80.7%. The total prediction accuracy for the prognosis of DOC patients was 87%.ROC curve, and the area of the 87%.ROC curve was 0.943. compared with the accuracy rate of the first part of BCI to predict the prognosis of DOC patients by BCI alone. Conclusion: 1. BCI based on face recognition can detect the consciousness level of DOC patients more objectively from the angle of brain response, and can evaluate the prognosis of patients well. The online accuracy can be used as an objective indicator to reflect the brain function and prognosis of DOC patients, and the possibility of consciousness recovery is more than 0.64. 2. the multimodal brain machine interface technology based on face recognition is more accurate in DOC patients' consciousness level diagnosis and prognosis assessment than the current clinical CRS-R scale. The combination of the two can better evaluate the prognosis of DOC patients, and the BCI technique can be used as a supplement to the clinical behavioral scale.

【学位授予单位】：广州中医药大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R741.044

【相似文献】