Targeted transcranial direct current stimulation combined with bimanual training for children with cerebral palsy

项目总结 父母拨款的总体目标是确定如何以最佳方式瞄准经颅直流电 刺激(TDC)提高单侧脑性瘫痪患儿上肢训练效果的研究 麻痹(UCP)。确定这种干预是否导致UE功能改善的一个关键工具是使用 临床评估。然而，即使是最好的评估也没有识别运动学的能力 儿童手部运动的特征。大多数评估都是用时间来完成任务，这是手工的必然结果 功能或对运动质量和功能的主观评级。通过不使用 运动学方面，可能有助于优化干预措施的信息严重丢失。UE的现有方法 运动学采集由于其大小、成本和操作复杂性而不容易获得。 在本补充方案中，我们的目标是使用深度学习(DL)姿势估计模型和3D 深度传感摄像头开发一款经济高效、易于使用和紧凑的基于深度学习的 无标记运动学数据采集(DL-KDA)系统，可应用于UCP儿童。为了 要实现这一总体目标，我们必须确定从 系统。我们将首先开发用于构建和测试DL-KDA系统的模块化软件框架 与非常精确的基于标记的动作捕捉黄金标准(VICON)形成对比。我们将研究3D的效果 摄像机和DL参数/结构对健康成人运动学数据准确性的影响 BBT期间的参与者。来自DL-KDA和黄金标准系统的运动学数据以及 各个DL-KDA参数数据将被转换成AI/ML就绪的HDF5文件。这本书将在 公共DL和NIH数据库，以鼓励进一步开发使用运动学数据的ML应用程序。 一旦确定了经过验证的最佳DL-KDA配置，我们将调查该系统是否适用于 适用于患有UCP的儿童。由于现有训练数据集用于大多数DL姿态估计模型 不包括患有运动障碍的儿童和/或成人，潜在的伦理和科学偏见可能 如果适用于代表人数不足的群体，则会出现。为了解决这个问题，我们将使用我们的UCP大型数据库 过去9年的评估视频，生成了约12,000张UCP儿童的姿势图像。图像 将被转换为ML/AI Ready HDF5数据集，并在公共DL和NIH存储库中发布。这些 数据集将可供其他研究人员在使用或构建DL姿势估计模型时考虑 在UCP临床研究中的应用。我们将使用此数据集执行迁移学习并重新训练 先前确定的最优DL模型。重新训练的DL模型的性能将是统计上的 与原始的DL模型进行比较，以验证是否确实存在偏差。最后，我们将应用最优DL- 在BBT期间，KDA使用再训练模型对20名UCP儿童进行了测试。验证了UCP的运动学 人口也将被上传到公共数字图书馆和国立卫生研究院的储存库，以供未来的UCP研究使用。

PROJECT SUMMARY The overall objective for the parent grant is to determine how to optimally target transcranial direct current stimulation (tDCS) to enhance the efficacy of upper extremity (UE) training in children with unilateral cerebral palsy (UCP). A key tool to determine whether this intervention leads to improves UE function is the use of clinical assessments. However, even the best assessments do not have the capacity to identify kinematic features of a child’s hand movement. Most assessments use time to complete a task as a corollary of hand function or subjective ratings of movement quality and function. By not capturing movement patterns using kinematics, there is a vital loss of information that could help optimize interventions. Existing methods of UE kinematics acquisition are not easily accessible because of their size, cost, and operational complexities. In this supplement proposal, we aim to use Deep Learning (DL) pose estimation models along with 3D depth sensing cameras to develop a cost effective, easy to use, and compact Deep Learning based markerless kinematic data acquisition (DL-KDA) system that can be applied to children with UCP. In order to achieve this overall goal, we must establish the accuracy and validity of the kinematic data obtained from the system. We will begin by developing a modular software framework for building and testing DL-KDA systems against a very precise marker-based motion capture gold standard (VICON). We will study the effects of 3D camera and DL parameters/architecture on the accuracy of the resulting kinematic data from healthy adult participants during BBT. Kinematic data from both, the DL-KDA and gold standard systems along with the respective DL-KDA parameter data will be transformed into an AI/ML ready HDF5 file. This will be published in public DL and NIH data repositories to encourage further development of ML applications using kinematic data. Once a validated and optimal DL-KDA configuration is identified, we will investigate this system’s suitability for applications to children with UCP. Since existing training datasets used for most DL pose estimation models are not inclusive of children and/or adults with movement disorders, potential ethical and scientific biases may arise if applied to an underrepresented group. To address this, we will use our large database of UCP assessment videos over the last 9 years to generate ~12,000 pose images of children with UCP. The Images will be transformed into ML/AI ready HDF5 datasets and published in public DL and NIH repositories. These datasets will be available for other researchers to consider when using or building DL pose estimation models for applications in UCP clinical research. We will use this dataset to perform transfer learning and retrain the optimal DL model previously identified. The performance of the retrained DL model will be statistically compared to the original DL model to verify if bias was indeed present. Finally, we will apply the optimal DL- KDA using the retrained model to ~20 children with UCP during the BBT. Validated kinematics for the UCP population, as well, will be uploaded to public DL and NIH repositories for use in future UCP research.