臺灣博碩士論文加值系統

帕金森氏症(Parkinson’s disease, PD)，是一種影響中樞神經系統的慢性神經退化疾病，通常會隨著時間緩慢地出現各種症狀。臨床表現以靜止性顫抖、肢體僵硬、行動緩慢為主要特徵，病情嚴重者走路會有隱痀的狀況，並且可能伴隨自主神經功能失調或精神行為異常表現。PD初期表現症狀容易被誤認為老年退化的自然現象，因此常被忽略且難以診斷，故本研究希望建立一個基於深度學習的帕金森氏症病患辨識模型來輔助醫師判斷出PD。
本研究使用由澄清醫院中港院區神經內科的神經生理檢測中心提供受測者資料庫，著重基於二氧化碳對自律神經功能病變患者的腦血流調控相關生理訊號之間的交互作用進行分析。本研究目前已收錄961位受測者，其中包括45歲以下健康群組(45-)、45歲以上健康群組(45+)、姿勢性直立心搏過速症病患群組(POTS)、糖尿病病患群組(DM)和帕金森氏症病患群組(PD)，記錄其在傾斜床測試(TTT)實驗中休息平躺(REST)、過度換氣(HV)、傾斜(Tilt-up)情況下的腦血流速(CBFV)、動脈血壓(ABP)、心率(HR)、呼吸率(BR)和潮氣末二氧化碳分壓(PETCO2)之訊號。觀察訊號在時域和線性的變化及利用Claassen和Battisti-Charbonney函數模型擬合出的參數及反應曲線，接著利用Mann-Whitney U檢定對非線性擬合參數進行統計分析，找出具顯著性差異的參數，在比較兩個函數模型非線性擬合參數的統計分析結果之後，最終將具顯著性差異參數數量較多的Claassen非線性擬合曲線作為辨識模型輸入。
本研究之辨識模型使用捲積神經網路(Convolution Neural Networks, CNN)對於不同群組進行深度學習訓練，使用二元與多元分類法，並透過準確率(Accuracy)、精準率(Precision)、召回率(Recall)及F1分數(F1-score)來評估模型性能。驗證結果顯示PD vs {45-, 45+, DM, POTS}五群組二元分類有最佳的準確率(0.87)及精準率(1)；PD vs DM vs 45+三元分類有最佳召回率(0.88)；PD vs DM vs 45- vs 45+四元分類有最佳F1分數(0.75)。

Parkinson’s disease(PD), a chronic neurodegenerative disease affecting the central nervous system, usually with various symptoms slowly over time. The clinical manifestations are mainly characterized by static tremor, limb stiffness, and slowness of movement. In severe cases, there will be kyphosis when walking, and may be accompanied by autonomic dysfunction or psychiatric and behavior abnormalities. The initial symptoms of PD are easily mistaken as a natural phenomenon of aging, so they are often ignored and difficult to diagnose. Therefore, this study hopes to establish a deep learning-based Parkinson's disease patient identification model to assist physicians in judging PD.
This study used the subject database provided by the Neurophysiological Testing Center of the Department of Neurology, Chung Kang Branch of Cheng Ching Hospital, focusing on the analysis of the interaction between physiological signals related to the regulation of cerebral blood flow in patients with dysautonomia based on carbon dioxide.This study currently includes 961 subjects, including healthy group under 45 years old (45-), healthy group over 45 years old (45+), patients with postural orthostatic tachycardia (POTS), patients with diabetes mellitus (DM) and patients with Parkinson's disease (PD). The signals of cerebral blood flow velocity(CBFV), arterial blood pressure(ABP), heart rate(HR), breath rate(BR) and end tidal carbon dioxide partial pressure(PETCO2) was recorded during resting phase(REST), hyperventilation phase(HV) and tilt phase(TILT-UP). Observe the changes of the signal in time domain and linearity and the response curvilinear fitted by the Claassen and Battisti-Charbonney function models, and then use the Mann-Whitney U test to perform statistical analysis on nonlinear fitting parameter to find significant differences. After comparing the statistical analysis results of the nonlinear fitting parameters of the two functional models, the Claassen nonlinear fitting curve with more significantly different parameters is finally used as the input of the identification model.
The identification model of this study uses Convolution Neural Networks (CNN) for deep learning training for different groups, uses binary and multivariate classification methods, and uses Accuracy, Precision, Recall Rate (Recall) and F1-score (F1-score) to evaluate model performance. The validation results show that PD vs {45-, 45+, DM, POTS} five groups of binary classification have the best accuracy (0.87) and precision (1); PD vs DM vs 45+ ternary classification has the best Recall (0.88); PD vs DM vs 45- vs 45+ quaternary classification had the best F1 score (0.75).

目錄
致謝 i
摘要 ii
Abstract iv
目錄 vi
圖目錄 x
表目錄 xiv
符號說明 1
第一章 緒論 3
1.1 前言 3
1.2 研究動機 3
1.3 研究目的 4
1.4 論文架構 4
第二章 背景 5
2.1 帕金森氏症的診療 5
2.2 資料庫 6
2.3 訊號擷取流程 7
2.4 腦血流調控對二氧化碳的非線性反應 9
2.4.1 腦血管舒縮反應性(Cerebral Vasomotor Reactivity, CVMR) 9
2.4.2 腦血管電導指數(Cerebrovascular Conductance Index, CVCi) 10
2.4.3 Claassen非線性分析模式 10
2.4.4 Battisti-Charbonney非線性分析模式 10
2.5 深度學習(Deep Learning) 11
2.6 文獻回顧 12
2.6.1 先前研究相關文獻 12
2.6.2 腦血流調控相關文獻 13
2.6.3 帕金森氏症與腦血流調控相關文獻 14
2.6.4 深度學習相關文獻 15
第三章 研究方法 16
3.1 訊號分析方法 16
3.1.1 訊號處理 16
3.1.2 時域分析(Time Domain Analysis) 19
3.1.3 線性分析(Linear Analysis) 19
3.1.4 非線性分析(Nonlinear Analysis) 20
3.1.5 統計分析 26
3.2 深度學習辨識模型方法與架構 27
3.2.1 捲積層(convolution) 27
3.2.2 激活函數 28
3.2.3 池化層(pooling) 28
3.2.4 全連接層(fully connected)與softmax層 29
3.2.5 殘差 29
3.2.6 性能指標 30
第四章 PD病患之腦血流調控訊號對二氧化碳反應分析 32
4.1 訓練集及驗證集基本資料與年份分布 34
4.2 PD群組與健康受測者群組之生理訊號平均 36
4.3 HV階段前30秒腦血流調控相關生理訊號之時域分析 37
4.3.1 HV階段前30秒M\mathrm{P}_{{\mathrm{ETCO}}_\mathrm{2}}時域分析 38
4.3.2 HV階段前30秒MCBFV時域分析 40
4.3.3 HV階段前30秒MBR時域分析 42
4.3.4 HV階段前30秒MSBP時域分析 45
4.3.5 HV階段前30秒MABP時域分析 48
4.3.6 HV階段前30秒MHR時域分析 50
4.4 HV階段前30秒腦血流調控相關生理訊號之線性分析 52
4.4.1 HV階段前30秒M\mathrm{P}_{{\mathrm{ETCO}}_\mathrm{2}}對MCBFV線性分析 53
4.4.2 HV階段前30秒M\mathrm{P}_{{\mathrm{ETCO}}_\mathrm{2}}對MABP線性分析 54
4.4.3 HV階段前30秒M\mathrm{P}_{{\mathrm{ETCO}}_\mathrm{2}}對MSBP線性分析 55
4.4.4 HV階段前30秒M\mathrm{P}_{{\mathrm{ETCO}}_\mathrm{2}}對MBR線性分析 56
4.4.5 HV階段前30秒M\mathrm{P}_{{\mathrm{ETCO}}_\mathrm{2}}對MHR線性分析 57
4.5 HV階段腦血流調控對二氧化碳反應之非線性分析 58
4.5.1 Claassen分析模式下各病患群組之CBFV對CO2反應 58
4.5.2 Claassen分析模式下各病患群組之CVCi對CO2反應 62
4.5.3 Battisti-Charbonney分析模式下各病患群組之CBFV對CO2反應 66
4.5.4 Battisti-Charbonney分析模式下各病患群組之CVCi對CO2反應 70
4.6 統計分析 74
4.6.1 Claassen(CBFV)非線性擬合參數顯著性分析 75
4.6.2 Claassen(CVCi)非線性擬合參數顯著性分析 77
4.6.3 Battisti-Charbonney(CBFV)非線性擬合參數顯著性分析 79
4.6.4 Battisti-Charbonney(CVCi)非線性擬合參數顯著性分析 81
4.6.5 函數模型之顯著性差異比較 83
第五章 基於深度學習之PD病患群組辨識模型驗證結果 85
5.1 辨識模型輸入樣本與年份分布 85
5.2 二元分類驗證結果 87
5.2.1 五群組二元分類結果 87
5.2.2 四群組二元分類結果 88
5.2.3 三群組二元分類結果 89
5.2.4 二元分類結果性能評估 90
5.3 多元分類驗證結果 92
5.3.1 五元分類結果 92
5.3.2 四元分類結果 93
5.3.3 三元分類結果 94
5.3.4 多元分類結果性能評估 95
第六章 結論與討論 97
6.1 訊號分析 97
6.2 辨識模型 100
6.3 未來工作 103
參考文獻 104
附錄A 107

[1] K. A. Jellinger, “How valid the clinical diagnosis of Parkinson’s disease in the community?” J Neurol Neurosurg Psychiatry 2003, no. 74, pp. 1003-1007, 2003.

[2] Ailiang Xie, James B. Skatrud, Barbara Morgan, Bruno Chenuel, Rami Khayat, Kevin Reichmuth, Jenny Lin and Jerome A. Dempsey, “Influence of cerebrovascular function on the hypercapnic ventilatory response in healthy humans,” The Journal of Physiology, vol. 577, no. 1, pp. 319-329, 2006.

[3] C. K. Willie, B. D. Macleod, A. D. Shaw, K. J. Smith, Y. C. Tzeng, N. D. Eves, K. Ikeda, J. Graham, N. C. Lewis, T. A. Day and P. N. Ainslie, “Regional brain blood flow in man during acute changes in arterial blood gases,” The Journal of Physiology, vol. 590, no. 14, pp. 3261-3275, 2012.

[4] Philip N. Ainslie and James Duffin, “Integration of cerebrovascular CO2 reactivity and chemoreflex control of breathing: mechanisms of regulation, measurement, and interpretation,” American Journal of Physiology-Regulatory, Integrative and Comparative Physiology, vol. 296, no. 5, pp. R1473¬¬-R1495, May 2009.

[5] Tsubasa Tomoto, Jonathan Riley, Marcel Turner, Rong Zhang and Takashi Tarumi, “Cerebral vasomotor reactivity during hypo- and hypercapnia across the adult lifespan,” Journal of Cerebral Flow & Metabolism, vol. 40, no. 3, pp. 600-610, 2020.

[6] Keju Ju, Lingling Zhong, Xiaoyu Ni, Hua Cao, Guanliang Cheng, Lianshu Ding, “Cerebral vasomotor reactivity predicts the development of acute stroke in patient with internal carotid artery stenosis,” Polish Journal of Neurology and Neurosurgery, vol. 52, issue 3, pp. 374-378, 2018.

[7] Luzius A. Steiner, Marek Czosnyka, Stefan K. Piechnik, Piotr Smielewski, Doris Chatfield, David K. Menon, John D. Pickard, “Continuous monitoring of cerebrovascular pressure reactivity allows determination of optimal cerebral perfusion pressure in patients with traumatic brain injury,” Critical Care Medicine, vol. 30, no. 4, pp. 733-738, 2002.

[8] Jurgen A. H. R. Claassen, Rong Zhang, Qi Fu, Sarah Witkowski and Benjamin D. Levine, “Transcranial Doppler estimation of cerebral blood flow and cerebrovascular conductance during modified rebreathing,” Journal of Applied Physiology, vol. 102, no. 3, pp. 870-877, 2007.

[9] A. Battisti-Charbonney, J. Fisher, and J. Duffin, “The cerebrovascular response to carbon dioxide in humans,” The Journal of Physiology, vol. 589, no. 12, pp. 3039-3048, 2011.

[10] Jun Gao, Qian Jiang, Bo Zhou and Daozheng Chen, “Convolutional neural networks for computer-aided detection or diagnosis in medical image analysis: An overview,” Mathematical Biosciences and Engineering, vol. 16, issue 6, pp. 6536-6361, 2019.

[11] Luis C.S. Afonso, Gustavo H. Rosa, Clayton R. Pereira, Silke A.T. Weber, Christian Hook, Victor Hugo C. Albuquerque, João P. Papa, “A recurrence plot-based approach for Parkinson’s disease
Identification,” Future Generation Computer Systems, vol. 94, pp. 282-292, 2019

[12] Ammarah Farooq, Syed Muhammad Anwar, Muhammad Awais and Saad Rehman, “A Deep CNN based Multi-class Classification of Alzheimer’s Disease using MRI,” IEEE International Conference on Imaging Systems and Techniques, 2017.

[13] 洪于凱，過度換氣下腦血流速對二氧化碳反應為基礎的自主神經失調病患
群組分析，逢甲大學，2016。

[14] 謝承蒲，基於機器學習與腦血流調控對二氧化碳反應非線性模式的姿勢性直立心搏過速症病患臨床決策支持系統，逢甲大學，2021。

[15] 陳威宇，基於機器學習的姿勢性直立心搏過速症病患臨床決策支持系統研發與人機介面設計，逢甲大學，2021。

[16] Kojiro Ide, Michael Eliasziw and Marc J. Poulin, “Relationship between middle cerebral artery blood velocity and end-tidal PCO2 in the hypocapnic-hypercapnic range in humans,” Journal of Applied Physiology, vol. 95, no. 1, pp. 129-137, Jul. 2003.

[17] Martha F. Hanby, Ronney B. Panerai, Thompson G. Robinson and Victoria J. Haunton, “Is cerebral vasomotor reactivity impaired in Parkinson disease?” Clinical Autonomic Research, vol. 27, pp. 107-111, 2017.

[18] Carlos Henrique Ferreira Camargo, Eduardo Antunes Martins, Marcos Christiano Lange, Henrique Alvaro Hoffmann, Jissa Jeanete Luciano, Marcelo Rezende Young Blood, Marcelo Derbli Schafranski, Marcelo Machado Ferro and Edmar Miyoshi, “Abnormal Cerebrovascular Reactivity in Patients with Parkinson’s Disease,” Parkinson’s Disease, vol. 2015, 5 page, 2015.

[19] W. Wang, J. Lee, F. Harrou and Y. Sun, “Early Detection of Parkinson's Disease Using Deep Learning and Machine Learning,” IEEE Access, vol. 8, pp. 147635-147646, 2020.

[20] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, “Deep Residual Learning for Image Recognition,” Preceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778, 2016.

[21] Boukaye Boubacar Traore, Bernard Kamsu-Foguem and Fana Tangara, “Deep convolution neural network for image recognition,” Ecological Informatics, vol. 48, pp. 257-268, 2018.

[22] 徐銓，基於多重感知集成式捲積神經網路及睡眠生理訊號時頻圖之自動睡眠階段判讀系統：使用多重數據集驗證，逢甲大學，2021。

[23] 羅忠滐，基於腦血流非線性反應與深度學習之糖尿病病患群組分類與辨識，逢甲大學，2022。