以連續影像對帕金森氏症患者步態之量化與評估

Abstract

帕金森氏症是一種慢性且漸趨嚴重的中樞神經系統疾病。臨床上的症狀包含了僵硬、震顫、行動遲緩及受損的姿態反應功能，而步態即為帕金森氏症之主要的特徵。時至今日，步態偏差評估在臨床的老年疾病及神經醫學疾病仍為一重要的議題，而且也是一個極佳的方法用於呈現帕金森氏症患者疾病的進程以及對於治療的反應。過去有許多研究電腦輔助步態分析方法，然而僅有有限的方法用於評估帕金森氏症步態以及證明其方法和多年來奉行的觀察診斷方法具有關聯性。因此，我們希望能發展一套客觀且量化之電腦輔助步態分析方法用於評估帕金森氏症步態。 在本研究裡，我們發展了一快速且大量的影像式且無需建構模型之帕金森氏症步態分析系統用於輔助臨床上的評估。演算法結合了以內核為基礎之主成分分析(Kernel-based Principal Component Analysis)及離散傅利葉轉換(Discrete Fourier Transform)，可用於得到時間的步態信號並進一步提取步態之頻譜以及最終量化的步態之步頻。結果顯示了對於帕金森氏症患者於用藥前後其平均步頻具有統計上的差異(t-test)。得到之步頻資料更進一步地和統一帕金森氏症分級量表(Unified Parkinson Disease Rating Scale)做關聯性的探討。在axial score, limb akinesia和Part Ⅲ這三者特定的評分分數上，得到了顯著性的關連性係數：0.7841、0.7352和0.7896 (p<0.05)。利用本研究發展的系統和演算法得到之步態步頻，証明了和某些臨床症狀上具有關聯性，且也許可用於輔助帕金森氏症步態之評估。

Parkinson’s disease (PD) is a chronic and progressive disease of central nervous system. The clinical symptoms include rigidity, tremor, bradykinesia, and impaired posture reaction, and gait is the major hallmark of PD. Nowadays the assessment of specific gait disturbances is an important issue in clinical geriatrics and neurological diseases and is also an excellent method for demonstrating change from treatment or from disease progression in parkinsonian patients. Many computer-aided gait analysis methods have been investigated, but there are limited methods proving to have dependable relationship with the observational diagnosis which has been followed for years. Therefore, we want to develop an objective and quantitative computer-aided gait analysis method for assessing parkinsonian gait. In this study, we describe the development of image-based model-free gait analysis system for fast and massive assistant clinical assessment in parkinsonian gait. The algorithm combines Kernel-based Principal Component Analysis (KPCA) with Discrete Fourier Transform (DFT) is proposed to obtain temporal gait signal and extracts the gait frequency spectrum and the gait terminology called stride frequency. The results demonstrate that the statistical difference between mean stride frequency of PD patients before and after drug treatment is confirmed by t-test (p<0.05). Stride frequency is further utilized to correlate with the Unified Parkinson Disease Rating Scale (UPDRS). The UPDRS sub-score: axial score, limb akinesia and Part Ⅲ have high and significant correlation coefficients of 0.7841, 0.7352 and 0.7896, respectively (p<0.05). The developed system and algorithm are proved to be correlated with some clinical symptoms and may be dependable and feasible for assisting the assessment of parkinsonian gait.