Ambulatory Movement and Activity Analysis for Patients with Neurological Disorders Based on Wearable Inertial Sensors

人体运动的评估是神经系统疾病诊断和治疗的核心。了解患者特定的运动模式对于医生诊断疾病，评估其严重程度和评估药物治疗效果至关重要。然而，目前常见的纯视觉运动检测的精度有限。这在很大程度上取决于医生的经验以及患者和护理人员描述其观察结果的能力。发作性症状，如癫痫发作，往往发生在没有通知别人和病人本身往往是健忘的癫痫发作。基于摄像机的运动分析系统大大增强了可以从单次癫痫发作中获得的信息。然而，它们的设置通常需要固定的临床环境。因此，这些系统不允许在延长的时间段内分析患者在其日常、非卧床例程中的运动。因此，医生无法获得有价值的信息，疾病症状可能无法评估。本项目的目的是开发一种针对神经系统疾病的运动分析系统，该系统可以连续获取和评估患者在日常生活中的运动数据。一个关键的研究重点是识别患者特定的运动模式的基础上附着在身体上的可穿戴惯性传感器。将开发用于学习患者特定运动模型的方法，这些方法允许自动识别关键事件，例如癫痫发作，并且可以从传感器数据重建患者运动，以便医生稍后进行检查。与使用摄像机相比，由患者身体上的惯性传感器捕获的信息非常有限。因此，从惯性传感器获得的信息将与关于人体运动的学习的先验知识相结合。为此，将实施训练设置，其中使用惯性传感器和光学运动捕捉系统同时记录每个患者的高细节运动数据。运动捕捉数据将用于了解传感器测量值与患者3D运动之间的关系。先前的研究已经证明了使用惯性传感器识别人类活动和身体姿势估计的可行性，并辅之以特定于人的运动模型。我们将专注于三种神经系统疾病：癫痫，帕金森病（PD）和多发性硬化症（MS）。对于癫痫，将在既定的临床环境中重点记录运动性癫痫发作，以进行术前癫痫监测。在PD的情况下，该系统将用于学习和识别可能由次优药物引起的患者特定运动控制问题（例如震颤、运动过度）。对于MS患者，系统将在诊断运动序列期间学习患者运动，以便对患者能力进行长期评估。

The assessment of human motion is central in the diagnosis and treatment of neurological disorders. Understanding patient-specific motion patterns is vital for physicians to diagnose diseases, to assess their severity and to evaluate the effects of medication. However, pure visual motion inspection, currently common practice, is subject to limited precision. It depends heavily on the experience of the physician and the ability of patients and caregivers to describe their observations. Paroxysmal symptoms, like epileptic seizures, often happen without notice of others and the patients themselves are often amnestic for their seizures. Camera-based motion analysis systems have greatly enhanced the information that can be derived from a single seizure. However, their setup typically requires a stationary, clinical environment. Therefore, these systems do not permit to analyze patient movements in their everyday, ambulatory routine over extended periods of time. Valuable information is thus not accessible for physicians and disease symptoms may remain unevaluated.The aim of this project is to develop a motion analysis system for neurological diseases that allows continuously acquiring and evaluating motion data of patients during their everyday life. A key research focus is to recognize patient-specific movement patterns based on wearable inertial sensors attached to the body. Methods for learning patient-specific motion models will be developed that allow for an automatic recognition of crucial events, such as seizures, and that can reconstruct the patient movements from the sensor data for a later inspection by a physician. The information captured by inertial sensors on a patient's body is very limited, as compared to using video cameras. Therefore, the information derived from the inertial sensors will be combined with learned, prior knowledge on human motion. For this purpose, a training setup will be implemented where high-detail movement data for each individual patient is recorded simultaneously with the inertial sensors and an optical motion capture system. The motion capture data will be used to learn the relationship between the sensor measurements and the patient¿s movements in 3D. Previous research has demonstrated the feasibility of using inertial sensors for recognition of human activities and for body pose estimation, assisted by person-specific motion models. We will concentrate on three neurological diseases: Epilepsy, Parkinson¿s Disease (PD) and Multiple Sclerosis (MS). For epilepsy, motor seizures will be focused and recorded in an established clinical setting for pre-surgical epilepsy monitoring. In the case of PD, the system will be used to learn and recognize patient-specific motor control issues (e.g. tremor, hyperkinesia) that can be caused by suboptimal medication. For MS patients, the system will learn patient motions during diagnostic movement sequences for a long-term assessment of patient capabilities.