利用99mTc-TRODAT-1 SPECT全腦影像特徵進行帕金森疾病分類之研究

The Study of Classified PD by Characteristics of Whole Brain 99mTc-TRODAT-1 SPECT Image

陳昭宇(Chao-Yu Chen)

指導教授 ： 陳泰賓

共同指導教授 ： 陳清江(Ching-Jiang Chen)

義守大學/電機資訊學院/資訊工程學系/碩士(2018年)

摘要

99mTc-TRODAT-1單光子電腦斷層(Single Photon Emission Computed Tomography, SPECT)造影技術，為臨床普遍用於評估腦部多巴胺傳導物質系統相關疾病之檢查。臨床上帕金森氏症(Parkinson's Disease, PD)的診斷大多數使用SPECT進行檢查，而帕金森疾病分類仍具有困難性及挑戰性。
本研究採用回顧性實驗設計，收集2003-2017年期間99mTc-TRODAT-1多巴胺掃描案例影像及診斷報告共計212例，其中對照組為正常者12例、實驗組為確診帕金森氏症200例並記錄年齡及性別。首先對99mTc-TRODAT-1 SPECT全腦影像建立位置尺度模型(Location-Scale Model)估算變異數之比值(F-Ratio)；接著萃取影像特徵係根據中位數將影像分成二區；接著估算大於等於暨小於中位數之影像總強度、影像平均強度、影像強度標準差及年齡，共八個特徵值進行帕金森疾病分類。統計分析包括描述性統計、Mann-Whitney U檢定、分類模型使用羅吉斯迴歸以及支持向量機，評估合理單光子電腦斷層影像之帕金森氏症分類模型。
結果顯示八個特徵值建立羅吉斯迴歸之分類模型，分辨正常腦部及帕金森氏症之準確率、靈敏度、特異性、陽性預測率、陰性預測率、Kappa分別為85.4%、85.5%、83.3%、98.8%、25.6%、33.5%。AUC值為0.932。代表此模型具有非常好的判別能力。
位置-比例模型為一種新穎且有用的重要特徵，利用全3D影像進行分類。 羅吉斯迴歸為可行的分類模型用來預測帕金森氏症。未來，為了減少偽陽性率，應於所提出之模型中增加更多正常腦部案例。同時，可考慮以先進的分類模型進行分類，如卷積神經網路或多層前饋式神經網路。

關鍵字

帕金森氏症 ； 位置尺度模型 ； 羅吉斯迴歸 ； 支持向量機

並列摘要

The 99mTc-TRODAT-1 Single Photon Emission Computed Tomography (SPECT) was often used to diagnosis brain function. The study was focused to find image features to analyze the classification of Parkinson's disease (PD).
The retrospective study was designed to collect 212 subjects SPECT. The experimental and control groups were involved 200 and 12 SPECT respectively. The seven novel features were extracted from 3D SPECT including UpperSum, UpperMean, UpperSD, LowerSum, LowerMean, LowerSD, and Location-Scale Model with subject’s age. The Mann-Whitney U test was used to exam significant features. Meanwhile, the logistic regression (LR) and Support Vector Machine (SVM) were applied to classify the groups of PD.
The accuracy, sensitivity, specificity, positive predictive value, negative predictive value, Kappa, and AUC (area under ROC) provided by LR with eight features on validated set were 85.4%, 85.5%, 83.3%, 98.8%, 25.6%, 33.5%, and 0.932 with respectively. The AUC generated by LR was higher than the SVM.
Location-Scale Model was a novel and useful significant features which was extracted whole 3D image to classify groups. The LR was feasible classified model to predict the groups of PD. In the future, the more negative cases should be added into the presented model in order to reduce false positive rate. Meanwhile, the state-of-art classified models might consider into comparison, such as convolutional neural networks or multilayer feedforward neural network.

並列關鍵字

99mTc-TRODAT-1 ； Location-Scale Model ； Logistic Regression ； Support Vector Machine

參考文獻

[1] Sasannezhad P, Juibary AG, Sadri K, Sadeghi R, Sabour M, Kakhki VRD, Alizadeh H. (99m)Tc-TRODAT-1 SPECT Imaging in Early and Late Onset Parkinson's Disease. Asia Ocean J Nucl Med Biol. 2017;5(2):114-119.

[2] Shih MC, Franco de Andrade LA, Amaro E Jr, Felicio AC, Ferraz HB, Wagner J, Hoexter MQ, Lin LF, Fu YK, Mari JJ, Tufik S, Bressan RA. Higher nigrostriatal dopamine neuron loss in early than late onset Parkinson's disease?--a [99mTc]-TRODAT-1 SPECT study. Mov Disord. 2007;22(6):863-6.

[3] Li J, Zhang HY, Jiang Y, Li TQ. [Comparison of Parkinson's monkey models induced by unilateral and bilateral intracerebroventricular injections of MPP]. Sheng Li Xue Bao. 2017;69(6):743-750.

[4] Hsiao IT, Weng YH, Lin WY, Hsieh CJ, Wey SP, Yen TC, Kung MP, Lu CS, Lin KJ. Comparison of 99mTc-TRODAT-1 SPECT and 18 F-AV-133 PET imaging in healthy controls and Parkinson's disease patients. Nucl Med Biol. 2014;41(4):322-9.

[5] Bor-Seng-Shu E, Felicio AC, Braga-Neto P, Batista IR, Paiva WS, de Andrade DC, Teixeira MJ, de Andrade LA, Barsottini OG, Shih MC, Bressan RA, Ferraz HB. Dopamine transporter imaging using 99mTc-TRODAT-1 SPECT in Parkinson's disease. Med Sci Monit. 2014;20:1413-8.