臺灣博碩士論文加值系統

背景: 磁振影像的電腦化分析已發展多年，以自動化、快速、精準及降低分析者間變異度等為目的，在科學研究可提供量化分析，而在臨床應用端亦可輔助醫護人員進行診斷與治療。電腦化分析演算法中也包含了近年愈來愈成熟的機器學習與深度學習方法，例如影像之區域分割、標籤預測等。偏頭痛是一個病因尚未明確，病理複雜多元的第一型頭痛，且在結構影像中沒有一個明顯的特徵以肉眼判斷偏頭痛患者，所以本篇研究以偏頭痛的結構及功能性磁振影像資料為例，做電腦計算分析，且主要目的有二:
(1) 用電腦計算分析結構性T1權重影像以及靜息態功能性磁振影像，對四種資料分組進行比較：偏頭痛無與有憂鬱表現、偏頭痛無或有預兆、控制組與偏頭痛每個月嚴重頭痛五天以上者，以及控制組與偏頭痛超過十年者。
(2) 利用上述分組中有差異的結果作為特徵，綜合過去文獻的背景知識，建立機器學習及深度學習的模型來對結構性與功能性差異做預兆、憂鬱、頭痛之二元分類多標籤預測。
材料與方法： 本研究收集了控制組之影像資料46筆，以及偏頭痛患者的影像及臨床資料251筆，影像包含T1權重影像及靜息態功能性磁振影像。T1權重影像分析利用體素形態學以及表面形態學兩種方法，體素形態學計算空間標準化後的灰質密度分布，而表面形態學計算數值化特徵包括皮質區域體積、厚度及表面積及白質、腦室體積。功能性影像計算低頻率振幅比例、區域同質性及功能區連結(相關係數與動態變異程度)。除功能區連結以置換檢驗作為統計方法外，其餘皆用雙樣本之雙尾T檢定。最後，利用上述分析得到的數值或圖譜特徵與不同的機器學習模型，對四種標籤進行二元分類預測。
結果： 從結構影像發現偏頭痛發作五年以上會造成右頂葉及右前額葉灰質體積減少，而頭痛每月十天以上也有右頂葉及顳葉減少情形。患者是否有預兆不會影響灰質體積，而憂鬱表現則會發現灰質在多區有減少。功能性影像的低頻率振幅比例與區域同質性中，頭痛、預兆與憂鬱皆有不同的區域活化與去活化位置，以控制組與頭痛五年以上及頭痛每月十天以上兩組有多處差異，且這兩組的差異有幾處相似的區域在枕葉中。在功能區連結分析中，有無預兆組有最多且最強的差異，無論相關程度或動態變異程度分析。有無憂鬱組的功能性連結差異主要在左右顳葉間，而兩種頭痛組的功能區連結皆較散亂。機器學習與深度學習對偏頭痛的分類與子分類普遍不高，最好的結果為利用功能性連結特徵分類偏頭痛中有無預兆，準確率為73.3%。
結論： 偏頭痛的憂鬱、預兆及頭痛會對大腦的結構及分區功能產生不同的影響，在結構及功能性影像中會有諸多個體差異，而個體差異則會影響機器學習表現。

Background
Computational Analysis of MRI has been developed for a long time. With the purpose of automation, rapidity, precision and low inter-subject variability, it offers feature quantification for scientific research, and clinically assists to diagnosis and medical treatment. The algorithms also include machine learning (ML) and deep learning, and they can apply on image segmentation and classification. Migraine is a primary headache that the etiology is still unclear, and its pathology is diverse also complex. Therefore, in this study, we aimed to analyze clinical migraine data, using different algorithms to make an integrated investigation to complete three missions:
1. Applying computational analysis to T1W image and rsfMRI, and compare binary sub-groups in four groups: migraine without depression (MwoD) versus with depression (MwD), migraine without aura (MwoA) versus with aura (MwA), control versus migraine with severe headache above 5 days per month, and control versus migraine occurred over 10 years
2. To perform ML and deep learning binary classification by using structural and functional features and the knowledge based on previous studies.
Materials and methods
In this study, we collected 46 normal control MRI and 251 migraine MRI with clinical data, and MRI sequences contain structural T1W image and rsfMRI. T1W analysis were using both VBM and SBM method. VBM calculates the spatial standardized GM distribution, and SBM calculates the numeric features including volume, thickness and surface area in each GM parcellation, and volume of WM and ventricles. In rsfMRI, we calculated fALFF map, ReHo map and functional connectivity in each subject. We use permutation test for functional connectivity statistics (correlation coefficient and dynamic variance), and two sample T-test for others. Finally, we apply ML by using numeric or volumetric feature to perform binary classification between two sub-groups.
Results
We found from structural MRI that migraine occurring above 5 years would decrease GM volume in right parietal and frontal lobe, and migraine with headache above 10 days per month would have similar pattern in frontal and temporal lobe. Aura would not affect GM volume, while depression would decrease GM volume in several regions. In fALFF and ReHo results, headache, aura and depression had distinct activation and deactivation regions. In the group of control versus migraine occurring above 5 years, and control versus migraine with headache above 10 days per month, the differences appeared in many places and both had similar patterns in occipital lobe. Among these 4 groups, the group MwoA versus MwA was found with the strongest and highest amount of difference in functional connectivity correlation coefficient and variance. In depression group, the differences were mainly in bilateral temporal lobe, and in two headache groups, the differences were more fragmentary. Performances for classifying migraine and migraine subgroup were not ideal. The best result was classifying aura in migraine using ML and functional connectivity features, and the testing accuracy is 73.3%.
Conclusion
The depression, aura and headache symptoms would affect in brain microstructure and functions respectively. Therefore, the structural and functional MRI may find various individual differences, and the individual differences affects directly the ML performance.

Contents
致謝 i
中文摘要 ii
Abstract v
Contents viii
List of Figures x
List of Tables xii
Chapter 1. Introduction 1
1.1 Migraine 1
1.2 Structural, Functional MRI and Computer Analysis 3
1.3 Traditional Machine Learning for classification 4
1.4 Convolutional Neural Network (CNN) for Classification 5
1.5 The Aim of This Study 9
Chapter 2. Material and Methods 11
2.1 Data Acquisition and Grouping 11
2.2 General Image Preprocessing 13
2.3 Image Processing Algorithms 13
2.3.1 Skull Stripping and HD-BET 13
2.3.2 Tissue Segmentation 15
2.3.3 Spatial Registration 17
2.4 Voxel-based Morphometry (VBM) 19
2.5 Surface-based Morphometry (SBM) by FreeSurfer 19
2.6 rsfMRI Processing 20
2.7 Machine learning with Numeric Features 22
2.8 CNN Architecture for Preprocessed Images 24
Chapter 3. Results 26
3.1 Morphometry Results 28
3.1.1 Statistics by FreeSurfer 28
3.1.2 Statistics by VBM 29
3.2 Functional Results 33
3.2.1 Statistics of fALFF 33
3.2.2 Statistics of ReHo 38
3.2.3 Statistics of Functional Connectivity 42
3.2.4 Summary of rsfMRI Results 52
3.3 Machine Learning Results 54
3.3.1 Machine Learning of Numeric Features 54
3.3.2 Deep Learning Results 54
Chapter 4. Discussion 56
4.1 Depression Pattern in Migraine 56
4.2 Aura Pattern in Migraine 57
4.3 Headache Pattern in Migraine 58
4.4 Feature Selection and Classification Performance 59
Appendix 61
Reference 67

List of Figures
Figure 1-1: Saturated activation functions. 7
Figure 1-2: Non-saturated activation functions. 8
Figure 2-1: The overview of this study. 11
Figure 2-2: The CNN architecture of HD-BET. 15
Figure 2-3: Demonstration of GMM. 16
Figure 2-4: The 3D-CNN architecture for MRI binary classification. 25
Figure3-1: GM volume difference between migraine without and with depression. 31
Figure 3-2: GM volume difference between control and migraine with severe headache above 5 days per month. 32
Figure 3-3: GM volume difference between control and migraine occurred over 10 years. 32
Figure 3-4: The fALFF difference between MwoD and MwD. 33
Figure 3-5: The fALFF difference between MwoA and MwA. 35
Figure 3-6: The fALFF difference between control and migraine with severe headache above 5 days per month. 36
Figure 3-7: The fALFF difference between control and migraine occurred over 10 years. 37
Figure 3-8: The ReHo difference between MwoD and MwD. 39
Figure 3-9: The ReHo difference between MwoA and MwA. 39
Table 3-12: Information of ReHo difference between MwoA and MwA. 39
Figure 3-10: The ReHo difference between control and migraine with severe headache above 5 days per month. 40
Figure 3-11: The ReHo difference difference between control and migraine occurred over 10 years. 41
Figure3-12: Illustration of functional connectivity differences between MwoD and MwD. 43
Figure 3-13: Illustration of functional connectivity differences between MwoA and MwA. 44
Figure 3-14: Illustration of functional connectivity differences between control and migraine with severe headache above 5 days per month. 46
Figure 3-15: Illustration of functional connectivity differences between control and migraine occurred over 10 years. 47
Figure 3-16: Illustration of correlation variance differences between MwoD and MwD. 48
Figure 3-17: Illustration of correlation variance differences between MwoA and MwA. 49
Figure 3-18: Illustration of correlation variance differences between control and migraine with severe headache above 5 days per month. 51
Figure 3-19: Illustration of correlation variance differences between control and migraine occurred over 10 years. 52

List of Tables
Table 2-1. The T1 sequence and parameters by MR750 scanner. 12
Table 2-2: The parameters of our CNN model. 25
Table 3-1(A): Demographic data of all participants in this study. 27
Table 3-1(B): Data group of MwoD and MwD.27
Table 3-1(C): Data group of MwoA and MwA.27
Table 3-1(D): Data group of migraine with headache below or larger than 10 days.27
Table 3-1(E): Data group of migraine occurred above 5 years. 27
Table 3-2: Volume and volume ratio difference between MwoD and MwD. 28
Table 3-3: Volume and volume ratio difference of control versus migraine with severe headache above 5 days per month, and control versus migraine occurred over 10 years. 29
Table 3-4: Information of VBM difference between MwoD and MwD. 31
Table 3-5: Information of VBM difference between control and migraine with severe headache above 5 days per month. 32
Table 3-6: Information of VBM difference between control and migraine occurred over 10 years. 32
Table 3-7: Information of fALFF difference between MwoD and MwD. 34
Table3-8: Information of fALFF difference between MwoA and MwA. 35
Table 3-9: Information of fALFF difference between control and migraine with severe headache above 5 days per month. 36
Table 3-10: Information of fALFF difference between control and migraine occurred over 10 years. 37
Table 3-11: Information of ReHo difference between MwoD and MwD. 39
Table 3-13: Information of ReHo difference between control and migraine with severe headache above 5 days per month. 40
Table 3-14: Information of ReHo difference between control and migraine occurred over 10 years. 41
Table 3-15 (A): Functional connectivity differences between MwoD and MwD, p<0.002.43
Table 3-15 (B): Degrees of regions with functional connectivity difference between MwoD and MwD.43
Table 3-16 (A): Functional connectivity differences between MwoA and MwA p<0.001. 43
Table 3-16 (B): Degrees of regions with functional connectivity difference between MwoA and MwA.44
Table 3-17 (A): Functional connectivity differences between control and migraine with severe headache above 5 days per month p<0.005. 45
Table 3-17 (B): Degrees of regions with functional connectivity difference between control and migraine with severe headache above 5 days per month. 45
Table 3-18 (A): Functional connectivity differences between control and migraine occurred over 10 years, p<0.002. 46
Table 3-18 (B): Degrees of regions with functional connectivity difference between control and migraine occurred over 10 years. 46
Table 3-19: Correlation variance differences between MwoD and MwD, p<0.005. 48
Table 3-20 (A): Correlation variance differences between MwoA and MwA, p<0.001. 49
Table 3-20 (B): Degrees of regions with correlation variance differences between MwoA and MwA. 49
Table 3-21 (A): Correlation variance differences between control and migraine with severe headache above 5 days per month, p<0.001. 50
Table 3-21 (B): Degrees of regions with correlation variance differences between control and migraine with severe headache above 5 days per month. 50
Table 3-22 (A): Correlation variance differences between control and migraine occurred over 10 years, p<0.002. 51
Table 3-22 (B): Degrees of regions with correlation variance differences between control and migraine occurred over 10 years. 51
Table 3-19: Binary machine learning results for different groups. PCA: A feature reduction method for dimension reduction. 54
Table 3-20: Binary classification deep learning results for different groups. 55
Table 4-1: Candidate Features for binary classification in four groups. 60

1. Olesen, J., et al., The international classification of headache disorders, (beta version). Cephalalgia, 2013. 33(9): p. 629-808.
2. May, A. and L.H. Schulte, Chronic migraine: risk factors, mechanisms and treatment. Nature Reviews Neurology, 2016. 12(8): p. 455.
3. Wang, S.J., et al., Prevalence of migraine in Taipei, Taiwan: a population‐based survey. Cephalalgia, 2000. 20(6): p. 566-572.
4. Charles, A., The evolution of a migraine attack–a review of recent evidence. Headache: The Journal of Head and Face Pain, 2013. 53(2): p. 413-419.
5. Giffin, N., et al., Premonitory symptoms in migraine: an electronic diary study. Neurology, 2003. 60(6): p. 935-940.
6. Goadsby, P.J., et al., Pathophysiology of migraine: a disorder of sensory processing. Physiological reviews, 2017. 97(2): p. 553-622.
7. Hadjikhani, N., et al., Mechanisms of migraine aura revealed by functional MRI in human visual cortex. Proceedings of the national academy of sciences, 2001. 98(8): p. 4687-4692.
8. Akerman, S., P.R. Holland, and P.J. Goadsby, Diencephalic and brainstem mechanisms in migraine. Nature Reviews Neuroscience, 2011. 12(10): p. 570.
9. Bernstein, C. and R. Burstein, Sensitization of the trigeminovascular pathway: perspective and implications to migraine pathophysiology. Journal of clinical neurology, 2012. 8(2): p. 89-99.
10. Maniyar, F.H., et al., Brain activations in the premonitory phase of nitroglycerin-triggered migraine attacks. Brain, 2013. 137(1): p. 232-241.
11. Ratcliffe, G.E., et al., The relationship between migraine and mental disorders in a population-based sample. General hospital psychiatry, 2009. 31(1): p. 14-19.
12. Chen, Y.-C., et al., Comorbidity profiles of chronic migraine sufferers in a national database in Taiwan. The journal of headache and pain, 2012. 13(4): p. 311.
13. Antonaci, F., et al., Migraine and psychiatric comorbidity: a review of clinical findings. The journal of headache and pain, 2011. 12(2): p. 115.
14. Breslau, N., et al., Headache and major depression: is the association specific to migraine? Neurology, 2000. 54(2): p. 308-308.
15. Wang, S.-J., et al., Migraine and suicidal ideation in adolescents aged 13 to 15 years. Neurology, 2009. 72(13): p. 1146-1152.
16. Ogawa, S., et al., Brain magnetic resonance imaging with contrast dependent on blood oxygenation. proceedings of the National Academy of Sciences, 1990. 87(24): p. 9868-9872.
17. Buckner, R.L., J.R. Andrews‐Hanna, and D.L. Schacter, The brain's default network. Annals of the New York Academy of Sciences, 2008. 1124(1): p. 1-38.
18. Andrews-Hanna, J.R., The brain’s default network and its adaptive role in internal mentation. The Neuroscientist, 2012. 18(3): p. 251-270.
19. Cover, T.M. and P.E. Hart, Nearest neighbor pattern classification. IEEE transactions on information theory, 1967. 13(1): p. 21-27.
20. Cortes, C. and V. Vapnik, Support-vector networks. Machine learning, 1995. 20(3): p. 273-297.
21. Hartigan, J.A. and M.A. Wong, Algorithm AS 136: A k-means clustering algorithm. Journal of the Royal Statistical Society. Series C (Applied Statistics), 1979. 28(1): p. 100-108.
22. Huppertz, H.J., et al., Differentiation of neurodegenerative parkinsonian syndromes by volumetric magnetic resonance imaging analysis and support vector machine classification. Movement Disorders, 2016. 31(10): p. 1506-1517.
23. Machhale, K., et al. MRI brain cancer classification using hybrid classifier (SVM-KNN). in 2015 International Conference on Industrial Instrumentation and Control (ICIC). 2015. IEEE.
24. Fukushima, K., Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biological cybernetics, 1980. 36(4): p. 193-202.
25. LeCun, Y., et al., Backpropagation applied to handwritten zip code recognition. Neural computation, 1989. 1(4): p. 541-551.
26. LeCun, Y., et al., Gradient-based learning applied to document recognition. Proceedings of the IEEE, 1998. 86(11): p. 2278-2324.
27. Krizhevsky, A., I. Sutskever, and G.E. Hinton. Imagenet classification with deep convolutional neural networks. in Advances in neural information processing systems. 2012.
28. Ioffe, S. and C. Szegedy, Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167, 2015.
29. Simonyan, K. and A. Zisserman, Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014.
30. Szegedy, C., et al. Going deeper with convolutions. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2015.
31. He, K., et al. Deep residual learning for image recognition. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
32. Szabó, N., et al., White matter microstructural alterations in migraine: a diffusion-weighted MRI study. Pain, 2012. 153(3): p. 651-656.
33. Rocca, M.A., et al., Brain gray matter changes in migraine patients with T2-visible lesions: a 3-T MRI study. Stroke, 2006. 37(7): p. 1765-1770.
34. Valfrè, W., et al., Voxel‐based morphometry reveals gray matter abnormalities in migraine. Headache: The Journal of Head and Face Pain, 2008. 48(1): p. 109-117.
35. Schmidt‐Wilcke, T., et al., Subtle grey matter changes between migraine patients and healthy controls. Cephalalgia, 2008. 28(1): p. 1-4.
36. May, A., Morphing voxels: the hype around structural imaging of headache patients. Brain, 2009. 132(6): p. 1419-1425.
37. Hougaard, A., et al., Structural gray matter abnormalities in migraine relate to headache lateralization, but not aura. Cephalalgia, 2015. 35(1): p. 3-9.
38. Lee, M.J., et al., Increased connectivity of pain matrix in chronic migraine: a resting-state functional MRI study. The journal of headache and pain, 2019. 20(1): p. 29.
39. Gudmundsson, L.S., et al., Migraine, depression, and brain volume: the AGES-Reykjavik Study. Neurology, 2013. 80(23): p. 2138-2144.
40. Kalavathi, P. and V.S. Prasath, Methods on skull stripping of MRI head scan images—a review. Journal of digital imaging, 2016. 29(3): p. 365-379.
41. Tustison, N.J., et al., N4ITK: improved N3 bias correction. IEEE transactions on medical imaging, 2010. 29(6): p. 1310.
42. Wang, L., et al., Correction for variations in MRI scanner sensitivity in brain studies with histogram matching. Magnetic resonance in medicine, 1998. 39(2): p. 322-327.
43. Cabezas, M., et al., A review of atlas-based segmentation for magnetic resonance brain images. Computer methods and programs in biomedicine, 2011. 104(3): p. e158-e177.
44. Ou, Y., et al., Comparative evaluation of registration algorithms in different brain databases with varying difficulty: results and insights. IEEE transactions on medical imaging, 2014. 33(10): p. 2039-2065.
45. Yan, C.-G., et al., DPABI: data processing & analysis for (resting-state) brain imaging. Neuroinformatics, 2016. 14(3): p. 339-351.
46. Destrieux, C., et al., Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature. Neuroimage, 2010. 53(1): p. 1-15.
47. Fischl, B., et al., Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain. Neuron, 2002. 33(3): p. 341-355.
48. Yu-Feng, Z., et al., Altered baseline brain activity in children with ADHD revealed by resting-state functional MRI. Brain and Development, 2007. 29(2): p. 83-91.
49. Zou, Q.-H., et al., An improved approach to detection of amplitude of low-frequency fluctuation (ALFF) for resting-state fMRI: fractional ALFF. Journal of neuroscience methods, 2008. 172(1): p. 137-141.
50. Zang, Y., et al., Regional homogeneity approach to fMRI data analysis. Neuroimage, 2004. 22(1): p. 394-400.
51. Koshiba, Y. and S. Abe. Comparison of L1 and L2 support vector machines. in Proceedings of the International Joint Conference on Neural Networks, 2003. 2003. IEEE.
52. Mukaka, M.M., A guide to appropriate use of correlation coefficient in medical research. Malawi Medical Journal, 2012. 24(3): p. 69-71.
53. Grieve, S.M., et al., Widespread reductions in gray matter volume in depression. NeuroImage: Clinical, 2013. 3: p. 332-339.
54. Liu, F., et al., Abnormal amplitude low-frequency oscillations in medication-naive, first-episode patients with major depressive disorder: a resting-state fMRI study. Journal of affective disorders, 2013. 146(3): p. 401-406.
55. Liu, C.-H., et al., Resting-state brain activity in major depressive disorder patients and their siblings. Journal of affective disorders, 2013. 149(1-3): p. 299-306.
56. Iwabuchi, S.J., et al., Localized connectivity in depression: a meta-analysis of resting state functional imaging studies. Neuroscience & Biobehavioral Reviews, 2015. 51: p. 77-86.
57. Yu, D., et al., Regional homogeneity abnormalities affected by depressive symptoms in migraine patients without aura: a resting state study. PLoS One, 2013. 8(10): p. e77933.
58. Takahashi, R., et al., Gender and age differences in normal adult human brain: Voxel‐based morphometric study. Human brain mapping, 2011. 32(7): p. 1050-1058.
59. Maleki, N., et al., Her versus his migraine: multiple sex differences in brain function and structure. Brain, 2012. 135(8): p. 2546-2559.
60. Faragó, P., et al., Interictal brain activity differs in migraine with and without aura: resting state fMRI study. The journal of headache and pain, 2017. 18(1): p. 8.
61. Niddam, D.M., et al., Reduced functional connectivity between salience and visual networks in migraine with aura. Cephalalgia, 2016. 36(1): p. 53-66.
62. Tedeschi, G., et al., Increased interictal visual network connectivity in patients with migraine with aura. Cephalalgia, 2016. 36(2): p. 139-147.
63. Buono, V.L., et al., Functional connectivity and cognitive impairment in migraine with and without aura. The journal of headache and pain, 2017. 18(1): p. 72.
64. Kravitz, D.J., et al., The ventral visual pathway: an expanded neural framework for the processing of object quality. Trends in cognitive sciences, 2013. 17(1): p. 26-49.
65. Kim, J., et al., Regional grey matter changes in patients with migraine: a voxel‐based morphometry study. Cephalalgia, 2008. 28(6): p. 598-604.
66. Khosla, M., et al., 3D Convolutional Neural Networks for Classification of Functional Connectomes, in Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. 2018, Springer. p. 137-145.