Nonlinear Filtering of Athetoid Movement

描述(申请人提供)：本提案的目标是开发手足徐动症运动的非线性模型，并在过滤算法中使用这些模型，以帮助手足徐动症脑瘫(CP)患者访问计算机。对于患有脑瘫和其他疾病的人来说，运动障碍是使用计算机的主要障碍(因此也是影响其总体独立性的主要障碍)。这是一种复杂的运动障碍，研究已经确定了非自主运动的多种成分，包括准周期(即震颤)和非周期成分。线性滤波的好处是有限的，因为手足徐动症既减少了有目的运动的带宽，又破坏了相同频率的非自愿运动的剩余带宽。在这项研究中，将开发一种非线性方法来估计手足徐动症患者的意向计算机输入。本研究将进行如下研究：1.患有手足徐动型脑瘫的受试者将在没有任何输入过滤的情况下使用操纵杆执行一系列计算机任务。受试者将使用一种非常适合健美操用户的新型等距操纵杆来完成一系列目标获取或“点击图标”任务。操纵杆读数和屏幕光标位置将被记录下来。2.在这些实验中，将使用神经网络来建立预期运动与实际运动之间的映射的非线性黑箱模型。一种通用的神经网络结构，级联相关学习结构，将被用于建模。3.所开发的模型将使用记录的数据进行验证。验证将使用一种随机相似性度量进行，该度量非常适合评估人类行为建模数据。4.经过验证的模型将作为输入过滤器在线实施，然后由运动受试者在现实的计算机任务中进行现场测试。受试者将在有过滤和无过滤的情况下完成目标获取试验，并对结果进行统计比较。

DESCRIPTION (provided by applicant): The goal of this proposal is to develop nonlinear models of athetoid movement and to use these models in filtering algorithms to assist computer access for persons with athetoid cerebral palsy (CP). Athetosis is a major barrier to computer usage (and therefore to general independence) for persons with cerebral palsy and other diseases. It is a complex movement disorder in which studies have identified multiple components of involuntary motion, including both quasi-periodic (i.e., tremor) and aperiodic components. Linear filtering is of limited benefit, since athetosis both reduces the bandwidth of purposeful movement and corrupts the remaining bandwidth with involuntary movement at the same frequencies. In this research, a nonlinear method for estimating intended computer input by persons with athetosis will be developed. The research will proceed as follows: 1. Human subjects with athetoid CP will use a joystick to execute a series of computer tasks without any input filtering. Subjects will complete a series of target acquisition or "icon-clicking" tasks using a novel isometric joystick that is well suited for athetoid users. The joystick readings and screen cursor positions will be recorded. 2. Neural networks will be used to develop nonlinear black-box models of the mapping between intended and actual movement during these experiments. A versatile neural network structure, the cascade-correlation learning architecture, will be used for modeling. 3. The models developed will be validated using recorded data. Validation will be performed using a stochastic similarity measure that is well suited to evaluation of human performance modeling data. 4. The validated models will be implemented online as input filters and will then be field- tested by the athetoid subjects in realistic computer tasks. Subjects will complete target acquisition trials both with and without filtering, and the performance results will be compared statistically.