Machine Learning and Signal Processing for Advances in Neurotechnology

自从人类能够思考以来，我们就一直在试图了解我们自己的大脑。然而，正如爱默生·M·普尔所指出的，“如果大脑如此简单，我们可以理解它，我们也会如此简单，以至于我们不能理解它。”尽管存在这一内在的谜团，但近年来，针对人脑的理解开展了激烈的、多学科的活动。神经学家、生物化学家、心理学家和其他人已经在这个主题上投入了几十年的时间，取得了令人印象深刻的成果。然而，直到最近，数学家和工程师才意识到，他们在推动这一领域的进步方面也发挥着重要作用。 大脑研究中的许多活动可以概括为神经技术这个术语。神经技术涉及旨在改善和修复大脑功能的技术的发展，或促进人脑和计算机之间的直接交互，即脑机接口(BCI)。按照这个思路，这项建议涉及到数学和工程原理在开发新的神经技术设备和程序中的应用。 我们在神经技术领域提供的一个诱人的前景是机器学习(ML)。由于大脑太复杂，无法直接建模，因此在这种情况下，机器学习方法的运作方式是仅使用来自大脑的观察数据构建一个基本的数学模型。申请人的团队此前曾使用脑电(EEG)的ML分析来诊断精神疾病，也用于脑机接口应用。此外，申请人的团队是第一个开发新的ML方法来确定有效的治疗严重抑郁症的方法。 精神/神经健康领域充满了ML的应用机会。我们打算开发新的ML方法来预测昏迷的恢复。我们还打算扩展我们新开发的ML算法，以增强精神疾病的治疗。第三，我们打算使用一种新的交互训练方法来开发改进的脑-机接口设备，其中人和计算机相互适应。我们还将开发新的ML工具来帮助实现这些目标。第一个这样的工具是扩展我们新开发的ML算法的能力，第二个是从脑电中识别代表大脑网络的新的、高度显著的生物标记物。 ML方法带来了一个新的以工程为中心的新视角来解决我们提出的长期存在的神经技术难题。我们的机器学习方法在医疗保健创新和开发新的脑机接口设备方面显示出巨大的前景。此外，我们建议的ML工具不仅对我们实现目标至关重要，而且还将为ML社区提供新的算法和技术。因此，这项研究为加拿大提供了通过加强和改进大脑研究中ML方法的使用来确认其在神经技术领域的存在的潜力。

For as long as human beings have been able to think, we have been trying to understand our own brains. However, as Emerson M. Pugh points out, ``If the brain were so simple we could understand it, we would be so simple we couldn't.'' Despite this inherent enigma, there has been intense, multi-disciplinary activity directed towards the understanding of the human brain in recent times. Neuroscientists, biochemists, psychologists and others have already been engaged for many decades on this topic, with impressive results. However, only recently have mathematicians and engineers realized that they too have an important role to play in furthering progress in this area. Much of the activity in brain research may be encapsulated into the term neurotechnology. Neurotechnology has to do with the development of technologies that are designed to improve and repair brain function, or that facilitate the direct interaction between the human brain and a computer, i.e., brain-computer interfaces (BCIs). In this vein, this proposal deals with the application of mathematical and engineering principles in the development of new neurotech devices and procedures. An enticing prospect we offer in the neurotech field is machine learning (ML). Since the brain is too complex to model directly, the machine learning approach in this context operates by constructing a rudimentary mathematical model using only observed data from the brain. The applicant’s team has previously used ML analyses of the electroencephalogram (EEG) to diagnose mental illness, and also for BCI applications. Also the applicant’s team was the first to develop novel ML methods for identifying effective treatments for major depression. The mental/neuro health field is rife with opportunity for applications of ML. We intend to develop novel ML methods for predicting recovery from coma. We also intend to extend our newly developed ML algorithms to enhance treatments for mental illness. Thirdly, we intend to develop improved BCI devices using a novel interactive training approach where the human and the computer adapt to each other. We will also develop new ML tools to aid in accomplishing these goals. The first such tool is to extend the capabilities of our newly developed ML algorithms, and the second is to identify new, highly salient biomarkers from the EEG that represent networks in the brain. The ML approach brings a fresh new engineering-centred perspective to solving the long-standing difficult neurotech problems we have proposed. Our machine learning approaches show significant promise for innovation in health care and in developing new BCI devices. Furthermore, our proposed ML tools are not only crucial to us achieving our goals, but will also offer the ML community new algorithms and techniques. Thus this research offers Canada the potential to assert its presence in the neurotech field by enhancing and improving the use of ML methods in brain research.