脊柱脊髓术中皮层体感诱发电位实时监测方法研究

The cortical somatosensory evoked potential monitoring can reflect the integrity of somatosensory pathway, and it is the principal means to monitor the spinal function during the spinal cord and spinal surgery. However, since the signal-noise-ratio of the cortical somatosensory evoked potential is extremely low, the real-time monitoring isn't achievable in current clinical practices. To solve the problem, making good use of the quasi-periodicity and the sparsity of the cortical somatosensory evoked potential, this project is going to study systematically the extracting method based on the transform joint sparse model in the complex intraoperative situations, and eventually achieve the real-time monitoring during the spinal cord and spinal surgery. The main contents include: (1) We study the impact of the anesthesia and the operation to the cortical somatosensory evoked potential in the spinal cord and spinal surgery and the constructing methods of sparse dictionary; (2) We study the designing methods of the transform joint sparse model； (3) We study the calculating methods of the transformation matrix and the sparse coefficients; (4) We study the calculating methods of the variability indexes of the amplitudes and latencies of the N20 of upper limb and the P40 of lower limb; (5) We study the real-time performance of the method for extracting the cortical somatosensory evoked potential in spinal cord and spinal surgery. The completion of this project may significantly improve the quality of intraoperative monitoring during the spinal cord and spinal surgery under current surgery conditions, and has a positive effect on raising the operation safety and reducing the postoperative complications, and on enriching and developing the medical signal processing theories and methods.

皮层体感诱发电位监测可反映躯体感觉传导通路的完整性，是目前脊柱脊髓外科术中监测脊髓神经功能的主要手段。但是由于皮层体感诱发电位信噪比极低，目前临床上无法实现实时监测。针对该问题，本项目拟充分利用皮层体感诱发电位的准周期性和稀疏性，系统研究术中复杂情况下基于变换域联合稀疏模型的皮层体感诱发电位提取算法，实现脊柱脊髓术中的实时监测。主要研究内容包括：(1)研究脊柱脊髓术中麻醉和操作对皮层体感诱发电位的影响及其稀疏字典的构造方法；(2)研究变换域联合稀疏模型的设计方法；(3)研究变换矩阵和稀疏系数求解方法；(4)研究上肢N20和下肢P40成分的波幅和潜伏期变化指数的计算方法；(5)研究脊柱脊髓术中皮层体感诱发电位提取方法的实时性。本项目的完成，将在现行的手术条件下显著提升脊柱脊髓术中监测的技术水平，对于提高手术安全性和减少术后并发症，丰富和发展医学信号处理理论和方法具有积极的作用。

诱发电位监测可反映躯体感觉传导通路的完整性，是目前术中监测神经功能的最主要手段。但是由于诱发电位信噪比极低，目前临床上主要采用累加平均法，无法实现诱发电位的实时跟踪。针对该问题，本项目充分利用诱发电位的准周期性和稀疏性，克服自发脑电的非平稳和非高斯性，提高了提取算法的性能。我们在提高诱发电位的信噪比、潜伏期的变化跟踪等几个方面取得了较大的研究进展。具体研究内容和成果包括：.(1)改进联合稀疏表示算法，提出变换域联合稀疏模型。该模型首先通过变换矩阵对连续两次脑电测量信号进行滤波处理，抑制自发脑电的干扰，然后对滤波后的信号采用联合稀疏分解得到公共稀疏系数和各自稀疏系数，最后重构得到单次诱发电位信号。该算法充分利用了诱发电位的准周期性和稀疏性，提高了估计效果。.(2)为了提高外输入自回归模型（ARX）模型对诱发电位潜伏期变化的动态追踪能力，将稀疏表示和ARX相结合提出诱发电位估计算法。在ARX中，自发脑电被看作为一个由白噪声驱动的自回归模型，而诱发电位可由一个输入信号已知的移动平均滤波器建模。我们利用诱发电位的稀疏性，将诱发电位用稀疏字典表示，并利用自回归参数修正字典，通过稀疏系数和字典的重构得到诱发电位。.(3)为更好的跟踪潜伏期的变化，提出多输入单输出ARX（MISO-ARX）算法。该算法将参考信号作为模板信号和子模板信号，利用MISO-ARX方法对单次观测信号的各个成分分别进行参数化建模。MISO-ARX方法避免了ARX方法中各个成分间的相互干扰，能够更好的进行潜伏期变化的跟踪。.(4)提出了基于最小P范数稀疏编码的单次诱发电位提取算法。该算法应用分数低阶稳定分布噪声()代替高斯白噪声表示自发脑电信号，将稀疏编码中的2范数问题转化成P范数问题。该算法根据自发脑电的特性建立更加符合实际情况的数学模型，能更有效的模拟实际信号，有利于提高估计的准确性。.(5)我们在医生的协助下实行大鼠实验，模拟了脊髓手术过程，采集了大鼠脊柱神经受损前后的体感诱发电位数据，进行了仿真实验。