ECoG,EEG-fMRI多模态癫痫监测与病灶定位研究

Epilepsy has been the hot point for cerebral disease in recent years. It is a kind of cerebral neuron system disease, which have characters of high morbidity、burstiness and repeatability. All of these characters together puts tremendous threats on patients' life security. Effective detection and localization of epilepsy have important clinical significance. Single module for noninvasive epilepsy prediction cannot offer a good prediction on epilepsy due to its low accuracy. Cortex EEG(ECoG) analysis method belongs to main stream methods for epilepsy diagnosis, however, this method has limitations like invasive detection and cannot get signal from whole part of cerebrum, all these side effects bring agony to epilepsy patients. Our research will combine EEG and MR technique, offers a new method for noninvasive epilepsy detection and localization. Our research based on previous EEG research methods and results, use fractional Fourier transform method to select signal features and then use support vector machine method to set standards for classification of epilepsy signal. We use the classification result to realize automatic epilepsy detection; At the meantime, we analysis ECoG signal、scalp EEG(sEEG) signal and Functional Magnetic Resonance data to reveal the relationship between these signals, we will also set up a brain signal transportation model to symbolize the relation. This model will be used to simulate the process of inner EEG signal transport to outside, thus we can obtain the relationship of intensity and position aspects between inner EEG signal and outside EEG signal. We will combine the localization result predicted by High Frequency Oscillations, and the EEG-fMRI method to get the final localization result. Finally, we will build a model to detect and localize signal which will lay the foundation for further noninvasive epilepsy detection and localization.

癫痫是一种发病率高的大脑神经系统疾病，具有突发性和反复性，对患者的生命安全构成极大的威胁。有效监测癫痫和对病灶定位对于该病的预防和治疗具有重要意义。单一因素无创监测癫痫发作存在精确度太低的问题，而目前主流的皮层脑电癫痫灶定位方法，存在无法检测全脑脑电信号并且有创的问题。本研究拟利用现有的皮层脑电信号检测方法与结果为基础，采用分数阶傅里叶变换提取脑电特效特征，用支持向量机算法建立分类标准，实现癫痫发作自动监测；同时，分析研究头皮脑电（sEEG）、皮层脑电（cEEG）与fMRI多模态信息的相关性，建立多模态癫痫信息传递模型；模拟内层脑电信号向外传播过程，获得sEEG与cEEG信号间的位置、强度对应关系，研究利用基于高频振荡信号（HFOs）的希尔伯特变换方法对皮层脑电信号分析结果，确定头皮脑电信号结合fMRI致病灶定位新方法，探索出一种多模态无创监测癫痫发作和病灶定位的新思路和新方法。

项目组从癫痫患者的头皮脑电、皮层脑电、立体定向脑电图与MRI多模态下的医学信号出发，开展了癫痫脑电信号的处理及有效特征提取工作，并进一步将有效特征应用于多模态的癫痫信息的传递模型当中。本项目完成了癫痫患者多模态信息的特征提取、特征筛选与分析工作；建立了癫痫发作基于脑电信号的SVM、MLP、LSTM机器学习预测模型，实现了癫痫发作的高准确度和高敏感度的预测；构建了脑电数据的癫痫动态网络并完成了癫痫网络动力学分析；进行了iEEG-MRI多模态的癫痫病灶区定位研究，提出一种基于立体定向脑电图频段能量分布特征图谱的致痫灶定位方法，在临床应用中具有广阔前景。发明了一种基于微带线的平衡驱动式磁共振射频线圈，在北京大学CMR磁共振平台的测试结果表明，该核磁线圈提高了信噪比，没有伪影，适用于高场强、频率为123.2MHZ的医用平台。