Nonlinear methods of analysis for biological signals

我们将研究用于分析非线性和混沌系统的技术，并将其应用于生物信号。特别是，我们对开发一种能够以无创方式监测肌肉疲劳的强大系统感兴趣。目前的“黄金标准”检查肌电信号的频谱（与肌肉收缩有关的电信号），跟踪中位数频率。当肌肉疲劳时，频谱会向较低的频率移动，导致中位数频率降低；然而，当收缩力和/或关节角度改变时，中位数频率也会改变，这使该方法混淆并限制了其实际应用。我们正在开发一种基于广义随机标度分形模型的肌电信号频谱的新方法。初步结果已经表明，这种方法可以分离收缩力、关节角度和肌肉疲劳的不同影响。这将使我们能够建立一个系统，不仅可以监测肌肉的静态收缩，还可以监测肌肉的动态收缩。研究可靠的无创监测肌肉疲劳的方法在人体工程学、损伤预防、运动医学和人类表现等应用领域具有重要意义。这些分析方法在肌肉活动发作检测中也有实用价值。肌肉活动的开始时间是生物力学和运动控制研究中使用的基本特征；然而，肌电信号的分析通常是手动进行的，或者如果使用自动化系统，通常仍然需要手动校正。问题是，要么噪声被错误地标记为肌电数据，要么低水平的收缩被忽略。这项研究将利用与背景噪声相比，肌电信号表现出与其分形几何和持久性相关的不同参数这一事实。从噪声中识别肌电信号的能力将提高自动发作检测的准确性和鲁棒性。这将为研究人员和临床医生节省时间，同时也有助于开发其他实时肌电信号分析系统。

We will be researching techniques developed for the analysis of nonlinear and chaotic systems and applying them to biological signals. In particular, we are interested in developing a robust system that is capable of monitoring muscular fatigue in a noninvasive manner. The current "gold standard" examines the spectrum of the myoelectric signals (electrical signals associated with muscle contractions), tracking the median frequency. As a muscle fatigues, the spectrum will shift towards lower frequencies, resulting in a decreased median frequency; however, the median frequency will also change when the force of the contraction and/or the joint angle is altered, which confounds this method and limits its practical utility. We are developing a novel method based on a Generalized Random Scaling Fractal model of the myoelectric signal spectrum. Initial results already indicate that this methodology can separate the different effects of contractile force, joint angle, and muscular fatigue. This will enable to construction of a system that is capable of monitoring muscular fatigue, not only for static contractions, but also for dynamic contractions. Research into reliable noninvasive methods for monitoring muscle fatigue has implications in application areas such as ergonomics, injury prevention, sports medicine, and human performance. These analysis methods also have utility in muscle activity onset detection. The onset time of muscle activity is a fundamental characteristic used in biomechanics and motor control research; however, analysis of the myoelectric signal for onset detection is often conducted manually, or if an automated system is used, manual correction is often still required. The problem is that either noise is falsely labeled as myoelectric data, or low levels of contractions are missed. This research will be leveraging the fact that myoelectric signals exhibits different parameters associated with its fractal geometry and persistence, as compared to its background noise. The ability to discern myoelectric signals from the noise will improve the accuracy and robustness of automatic onset detection. This will save time for researchers and clinicians, but also assist in the development of other real-time myoelectric signal analysis systems.