發展一個以統計分析為基礎之自動分類器用於巴金森氏病的診斷

Developing an automatic classifier for Parkinson's Disease Diagnosis based on Statistical Analysis of SPECT Data

巴金森氏病(PD)是老年常見的神經退化性疾病，其主要的症狀為動作遲緩、僵直、姿態不穩與顫抖。雖然經過數十年的研究，對於巴金森氏病的腦血流變化仍然未有定論，而目前的單光子電腦斷層攝影對於評估局部的腦血流與了解疾病的病理生理學是一項有用的工具。本研究之主要目的，在發展一個以統計分析為基礎之自動分類器用以輔助及評估巴金森氏病的診斷。我們利用獨立元件分析的方法開發出一套演算法來分析正常人與巴金森氏病人的單光子電腦斷層攝影影像資料，並比較正常人與巴金森氏病人其大腦中的局部腦血流差異的位置。我們的結果顯示，辨識出的PD相關位置與DeLong's PD model的預測相當吻合。我們運用Support Vector Machine (SVM)發展出巴金森氏症的輔助判別，並利用Receiver Operator Characteristics (ROCs) curves來估其效能。由獨立元件分析所得到的巴金森氏症大腦局腦血流異常區域當成特殊指標，運用SVM自動分類器依此指標來做資料的分類。研究結果顯示，運用此一完整的方式可以達到92％的敏感度與96％的特異度，其ROC curves下面積更達99％，顯見這方法可以有效的協助臨床醫師做診斷上的協助指標。

Parkinson's disease (PD) is a common neurodegenerative disease with four cardinal motor features: resting tremor, bradykinesia, cogwheel rigidity and postural instability. The pathophysiological mechanisms of PD remain largely unknown, but the primary neurotransmitter deficit appears mainly from the loss of dopaminergic (DA) nigrostriatal neurons in substantia nigra pars compacta, resulting in a loss of dopamine, mostly prominent in the striatum, According to the basal ganglia circuit model, the loss of dopamine results in abnormal activity in internal globus pallidus (GPi) both directly and indirectly, contributing to the observed movement abnormalities in PD. Although the correlations between motor features and their cerebral substrates are not yet completely understood, a common expectation is that the alteration in Parkinson's disease of functional activity in basal ganglia should be associated with changes in regional cerebral metabolism (rCMR) and regional cerebral blood flow (rCBF).