New approaches to non-stationary analysis of neurological signals

利用目前的成像技术和仪器，可以从人体中无创地获得丰富的信息。但是，我们如何最好地解释这些数据，以便更好地了解身体的功能，以进行科学探索和改善人类健康？生物医学信号面临的一个挑战是，它们在本质上往往是非静止的。我们试图通过开发新的理论方法来解决这个问题，并应用这些方法以一种以前不可能的方式来理解和分类人类疾病状态。我们工作的一个领域是更好地表征语音等声学信号的时频行为。通过模仿人耳的功能，我们开发了新的信号处理技术，使我们能够量化患有帕金森病的患者的低频声音震颤。目前还没有已知的治疗帕金森病的方法，也没有可靠的方法（侵入性或非侵入性）来诊断这种疾病。通过研究语音包络中的低频振幅和频率调制，我们可以正确诊断95%以上的早期帕金森病。未来几年对这一领域研究的支持将使我们不仅能更好地了解这种疾病，而且能显著改善其早期诊断。我们也一直在开发其他技术来分析更一般的非平稳信号，如脑诱发电位。在傅里叶分析方面的经典方法不能提供这种神经信号的清晰表示。相反，我们开创了一种方法，将这些信号分解成称为小波的基本信号的总和。我们已经证明，这种技术可以帮助阐明精神分裂症患者的皮质抑制损伤。通过研究精神分裂症可能的潜在神经机制，我们的研究结果可能有助于在不久的将来提出一种新的治疗方法，可以恢复对健康大脑功能的必要抑制。

With current imaging technology and instrumentation, there is a wealth of information that can be obtained non-invasively from the human body. But how do we best interpret this data to obtain a better understanding of the functioning of the body both for scientific exploration and for the improvement of human health? A challenge with biomedical signals is that they are often non-stationary in nature. We attempt to tackle this problem by developing new theoretical approaches and applying these approaches to understand and classify human disease states in a manner not previously possible. One area of our work has been to better characterize the time-frequency behaviour of acoustic signals like speech. By mimicking the functioning of the human ear, we have developed new signal processing techniques which allow us quantify low-frequency vocal tremors of patients suffering from Parkinson's disease. Currently there is no known cure for Parkinsonism, nor is there a reliable method (invasive or non-invasive) to diagnose the disease. By studying low-frequency amplitude and frequency modulations in the speech envelope, we can correctly diagnose more than 95% of early case Parkinson's disease. Support for research in this area over the next few years will allow us not only to better understand the disease but to dramatically improve its early diagnosis. We have also been developing other techniques to analyze more general non-stationary signals like brain evoked potentials. Classical approaches in terms of Fourier analysis do not provide a clean representation of such neurological signals. Instead we have pioneered an approach to decompose these signals into a sum of elementary signals known as chirplets. We have shown that this technique can help elucidate the impairment of cortical inhibition in patients with schizophrenia. By investigating the possible underlying neuromechanism(s) underlying schizophrenia, our results may help suggest new types of treatment that can restore the necessary inhibition for healthy brain function in the near future.