背景:神经反馈是一种非侵入性脑训练技术,用于通过改变脑活动模式来增强和治疗多动症。尽管如此,神经反馈增强的程度因个体/患者而异,其中许多人对这种治疗技术没有反应。因此,已经进行了几项研究来预测神经反馈训练的有效性,包括在开始治疗之前特别强调慢皮质电位(SCP)的theta/beta方案,以及根据不同人群的年龄和性别标准检查SCP标准。虽然其中一些研究未能做出准确的预测,其他人的成功率很低。这项研究探讨了脑电图(EEG)信号的不同频带中各个脑叶之间的功能连接,并且相位锁定的值用于预测神经反馈治疗开始前的潜在有效性。
方法:本研究利用孟德尔数据库中的脑电图数据。在这个数据库中,在涉及60名7-14岁多动学生的神经反馈会议期间记录EEG信号,不管性别。这些学生分为可治疗和不可治疗。所提出的方法包括五步算法。最初,数据经过预处理,以减少噪声使用多级滤波过程。第二步涉及从预处理的EEG信号中提取α和β频带,特别强调从神经反馈治疗的第10到20个疗程记录的脑电图。第三步,该方法通过使用锁相值评估不同脑叶的功能关系来评估两组之间的脑信号差异,一个关键的数据特征。第四步的重点是缩小特征空间,并确定神经反馈治疗最有效和最佳的电极。两种方法,概率指数(p值)通过t检验和遗传算法,被雇用。这些方法表明,最佳电极位于额叶和中央大脑皮层,特别是通道C3,FZ,F4,CZ,C4和F3,因为它们在两组之间表现出显着差异。最后,第五步,机器学习分类器被应用,并将结果合并以生成每个数据集的可治疗和不可治疗标签。
结果:在分类器中,支持向量机和Boosting方法在组合时表现出最高的精度。因此,所提出的算法在短时间内和有限的数据下成功地预测了多动症个体的可治疗性,在神经反馈方法中达到90.6%的准确度。此外,它有效地识别了神经反馈治疗中的关键电极,他们的人数从32人减少到6人。
结论:本研究引入了一种算法,用于预测多动症的神经反馈治疗结果,准确率为90.6%。通过确定最佳电极并将其数量从32个减少到6个,显着提高治疗效率。所提出的方法能够预测患者对神经反馈治疗的反应性,而不需要大量的疗程。从而节约时间和财力。
BACKGROUND: Neurofeedback is a non-invasive brain training technique used to enhance and treat hyperactivity disorder by altering the patterns of brain activity. Nonetheless, the extent of enhancement by neurofeedback varies among individuals/patients and many of them are irresponsive to this treatment technique. Therefore, several studies have been conducted to predict the effectiveness of neurofeedback training including the theta/beta protocol with a specific emphasize on slow cortical potential (SCP) before initiating treatment, as well as examining SCP criteria according to age and sex criteria in diverse populations. While some of these studies failed to make accurate predictions, others have demonstrated low success rates. This study explores functional connections within various brain lobes across different frequency bands of electroencephalogram (EEG) signals and the value of phase locking is used to predict the potential effectiveness of neurofeedback treatment before its initiation.
METHODS: This study utilized EEG data from the Mendelian database. In this database, EEG signals were recorded during neurofeedback sessions involving 60 hyperactive students aged 7-14 years, irrespective of sex. These students were categorized into treatable and non-treatable. The proposed method includes a five-step algorithm. Initially, the data underwent preprocessing to reduce noise using a multi-stage filtering process. The second step involved extracting alpha and beta frequency bands from the preprocessed EEG signals, with a particular emphasis on the EEG recorded from sessions 10 to 20 of neurofeedback therapy. In the third step, the method assessed the disparity in brain signals between the two groups by evaluating functional relationships in different brain lobes using the phase lock value, a crucial data characteristic. The fourth step focused on reducing the feature space and identifying the most effective and optimal electrodes for neurofeedback treatment. Two methods, the probability index (p-value) via a t-test and the genetic algorithm, were employed. These methods showed that the optimal electrodes were in the frontal lobe and central cerebral cortex, notably channels C3, FZ, F4, CZ, C4, and F3, as they exhibited significant differences between the two groups. Finally, in the fifth step, machine learning classifiers were applied, and the results were combined to generate treatable and non-treatable labels for each dataset.
RESULTS: Among the classifiers, the support vector machine and the boosting method demonstrated the highest accuracy when combined. Consequently, the proposed algorithm successfully predicted the treatability of individuals with hyperactivity in a short time and with limited data, achieving an accuracy of 90.6% in the neurofeedback method. Additionally, it effectively identified key electrodes in neurofeedback treatment, reducing their number from 32 to 6.
CONCLUSIONS: This study introduces an algorithm with a 90.6% accuracy for predicting neurofeedback treatment outcomes in hyperactivity disorder, significantly enhancing treatment efficiency by identifying optimal electrodes and reducing their number from 32 to 6. The proposed method enables the prediction of patient responsiveness to neurofeedback therapy without the need for numerous sessions, thus conserving time and financial resources.