| 摘要: | 本論文在探討使用穿戴式裝具來量化中風受試者的神經恢復等級與日常生活失能的程度，目的在使用穿戴式裝具以達成降低醫療專業人員評估病患的負擔。論文分為兩大部分，一是使用整合表面肌電圖與慣性感測器來預測神經的分級，另一部分則是使用慣性感測器來推論中風受試者神經恢復分級與生活功能的等級。
第一部份先使用不同機器學習演算法，結合表面肌電圖與慣性感測器，分類神經恢復的等級；其次基於此研究結果，嘗試使用一顆放在足背的慣性感測器，作為預測神經恢復分級的方式；第二部分則是使用4顆慣性感測器來擷取運動學資料，並進行臨床常用的10公尺步行測試，推論運動學與神經恢復等級以及日常生活能力障礙相關性，同時也提出新的步態對稱指標，使用受試者身高來修正對稱指數，用來推論受試者的失能與恢復狀況。
第一部分的試驗結果顯示，結合表面肌電圖與慣性感測器的演算法中，無論是支援向量機（Support Vector Machine, SVM）、k-近鄰演算法（k-Nearest Neighbor, k-NN）以及類神經網路（Artificial Neural Network, ANN）均可達不錯的分級能力，正確率範圍在82.86%到100%；其中以支援向量機的正確性最佳，高達92.5%到100%；最後使用單一顆慣性感測器結合運動生理學的演算法，正確率從86.8%到100%。
第二部分，結果顯示平均步行測試與傅格-梅爾量表下肢部分及伯格平衡量表均存在中度的正相關，相關係數分別為0.62與0.68；使用修正後步態對稱指數，與傅格-梅爾量表下肢部分和伯格平衡量表則呈現中度的負相關，相關係數傅格-梅爾量表下肢部分為-0.51，而伯格平衡量表為-0.52，顯示修正後步態對稱指數，具有可推論神經學與平衡恢復的能力。
綜上所述，從結合兩種感測器到最後使用4顆慣性感測器的研究，可以看出使用慣性感測器結合中風生理學的恢復特性，有機會用來推論受試者的神經恢復與日常失能的等級。修正後步態對稱指數則值得進一步探討與中風病患日常生活功能缺陷的關係。;In this thesis, two serial studies attempted to quantify neurological and functional defects by correlating clinical assessment scales and kinematics obtained from wearable devices, including surface electromyography and inertial sensors. This thesis contains two domains. One includes classifying neurological recovery status by integrating surface electromyography and inertial sensors to classify clinical Brunnstrom stages. Then, we managed to classify the neurological level through a single inertial sensor.
The results explore that the Support Vector Machine has the best accuracy in classifying the Brunnstrom stage than k-Nearest Neighbor and the Artificial Neural Network. The overall accuracy ranges from 82.86% to 100%, while the Support Vector Machine owns an accuracy ranging from 92.5% to 100%. Additionally, after adopting the threshold values, we get a good accuracy, ranging from 86.78% to 100% among different stages, to classify the Brunnstrom stages, by using a single inertial sensor on the subjects’ feet. This system has been proven feasible for predicting the lower limb Brunnstrom stage for hemiparetic and hemiplegic patients.
The other domain correlates clinical assessment scales and gait parameters from a 10-m walk test. In addition to gait parameters, we also set up a modified gait symmetric index to infer functional recovery results, the Fugl-Meyer assessment scale, and the Berg balance scale. These results explore only the average walking speeds correlate moderately with these two scales (γ = 0.62, 0.68, n = 23), while the modified gait symmetry index owns the moderate negative correlation(γ =-0.51, -0.52, n=23).
Conclusively, with appropriate algorithms and suitable locations to deploy inertial sensors, inertial-based wearable devices demonstrate promising applications in classifying neuromotor and functional recovery among patients after stroke. The modified gait symmetry index warrants further exploring the possibility of inferring functional recovery. |
