由於復健治療師的人數有限，家中復健經常作為中風病患的重要療程。在家中復健期間，中風患者可能會因為無法確定自己的運動功能是否有所改善而對家中復健療程感到沒有動力。針對這個問題，我們提出了基於動態伸臂運動練習的分析框架，該練習是針對輕度中風者並旨在訓練中風患者的上肢運動功能。首先，本研究的分析框架會提取患者的平滑度特徵，並根據這些特徵使用基於深度學習的多分類器來分類中風患者是左側偏癱還是右側偏癱。接下來，使用高斯混合模型作為量化患者運動數據的性能指標。最後，使用評分函數來將高斯混合模型的對數似然，也就是患者運動數據的性能指標映射到0到1的區間，作為可解釋的質量分數。本研究使用了臺北榮民總醫院的病人數據來驗證分析框架。;Due to the limited number of therapists, home-based rehabilitation exercises often be employed as an important treatment for stroke patients. During home rehabilitation, patients might feel unmotivated since they would not be sure about their motor functions have improved or not. To solve this issue, this paper proposes an analysis framework based on dynamic reaching exercise, a rehabilitation exercise for patients with mild stroke aims to train upper-limb motor function. First of all, our analysis framework extracts patient’s smoothness-based features and classify patient’s affected region according to these features with deep learning based multi-classifier. Next, we use Gaussian mixture model as the performance metric for quantitate patient’s movement data. Finally, we use scoring function to map the log-likelihood of Gaussian mixture model, the performance metric of patient’s movement data into the interval from 0 to 1, as an interpretable quality score. This presented framework is verified by using stroke patients’ movement data in dynamic reaching exercise at Taipei Veterans General Hospital.
