帕金森氏症是常見的神經退化性疾病，其症狀會影響語言功能與日常生活，且確診後預期平均餘命為7到15年，早期診斷將有助於療程規劃並延緩病程發展。在臨床上，常使用單光子放射電腦斷層掃描(single photon emission computed tomography, SPECT)與X光電腦斷層掃描(computed tomography, CT)混合影像進行診斷。在此傳統方法中，根據CT影像手動圈選紋狀體的位置，接著將SPECT與CT影像對位以計算結合率(binding ratio, BR)的過程將耗費大量人力與時間成本，因此先前研究提出使用深度學習演算法自動分割二維CT影像的紋狀體位置。但該方法仍需手動圈選枕葉區計算參考值，且可藉由直接輸入三維SPECT影像作為訓練模型之資料庫，以獲得三維結構之資訊定位紋狀體並提升計算結合率的效率。因此本研究使用三維SPECT影像訓練3D U-Net圈選紋狀體，並根據影像活動度估計參考值，以實現結合率的全自動計算。根據結果顯示，本研究提出之參考值自動估計方法相較傳統方法R2達到0.94，證明此一方法能用於降低在手動圈選所耗費之成本。而透過深度學習分割紋狀體之結果顯示利用3D U-Net模型自動分割目標區之效能相較U-Net模型提高19%的Jaccard index。而比較不同訓練資料的結果發現輸入經過前處理所得到之BR影像，相較輸入原始SPECT影像效果還要好2.1%的Jaccard index。本研究所提出之方法中，使用三維深度學習模型分割SPECT影像中的紋狀體不但能分割高活動度區，在低活動度的目標區也能被分割出來，使得到的結合率與實際值(ground truth)相比平均絕對誤差僅0.034且R2達到0.98。在混淆矩陣的分析中，精密度(precision)為71.2%，靈敏度(sensitivity)為81.3%，皆比傳統方法還高，且分割一個受試者的目標區平均只需要3.6秒。透過上述研究結果可證明提出之方法更為快速且穩定，另外此方法相較傳統方法和先前研究不需要CT影像的輔助，同時具有減少對位的不確定性、更高的BR且更接近實際值等優勢。

Parkinson's disease is a common neurodegenerative disease able to affect language function and daily life. After confirmation of Parkinson's disease, the remaining life expectancy is about 7 to 15 years and thus the early diagnosis will be helpful for treatment plan to slow down the disease progression. In clinical practice, the hybrid images comprised of single photon emission computed tomography (SPECT) and X-ray computed tomography (CT) are frequently used for diagnosis. In this conventional method, the process of manually selecting the position of the striatum based on the CT image, and then aligning the SPECT and CT images to estimate the binding ratio (BR) wastes a lot of time and labor resource. Therefore, previous study has proposed the utilize deep learning algorithms to automatically segment the striatum region in two-dimensional CT images. Nevertheless, manual delineation of occipital region was necessary to estimate the reference value in that method, and the three-dimensional (3D) SPECT images are able to be directly input as the training set of model for acquiring the additional 3D structural information of striatum and improving the performance of estimating BR. Therefore, 3D SPECT images were used for training 3D U-Net to segment the striatum and the reference value was estimated from the activity of images for realizing the fully automatic calculation of the BR in this study. The results showed the R2 reached to 0.94 compared the proposed method of automatic reference value estimation with the conventional method, which proved the method was able to reduce the cost of manual delineation. The results of segmenting the striatum by deep learning showed that the Jaccard index using 3D U-Net model for automatic target region segmentation was 19% higher than that using U-Net model. Comparing the results of different training data shows that the BR image obtained through pre-processing is 2.1% of Jaccard index better than the original SPECT image. The proposed method in this study was capable of segmenting the target regions not only with high-activity but also with low-activity utilizing 3D deep learning model for the striatum segmentation in SPECT images. So that the mean absolute error of binding ratio was only 0.034 and the R2 reaches to 0.98 compared to the ground truth. In the analysis of the confusion matrix, the precision is 71.2% and the sensitivity is 81.3%, both of which were higher than that of the conventional method, and the proposed method only took 3.6 seconds to segment the target region of a subject in average. The results demonstrated that the proposed method was faster and more stable. In comparison with conventional method and previous study, the proposed method did not require the assistance of CT images and had the advantages of reducing the uncertainty of the alignment, increasing binding ratio and closing to the ground truth.

[1] L. M. L. de Lau and M. M. B. Breteler, "Epidemiology of Parkinson's disease," Lancet Neurol., vol. 5, no. 6, pp. 525-535, Jun 2006.
[2] L. V. Kalia and A. E. Lang, "Parkinson's disease," The Lancet, vol. 386, no. 9996, pp. 896-912, Aug 2015.
[3] L. I. Golbe and C. E. Leyton, "Life expectancy in Parkinson disease," Neurology, vol. 91, no. 22, pp. 991-992, 2018.
[4] S. Sveinbjornsdottir, "The clinical symptoms of Parkinson's disease," Journal of Neurochemistry, vol. 139, no. S1, pp. 318-324, 2016.
[5] C. Ramaker, J. Marinus, A. M. Stiggelbout, and B. J. van Hilten, "Systematic evaluation of rating scales for impairment and disability in Parkinson's disease," Movement Disorders, vol. 17, no. 5, pp. 867-876, 2002.
[6] C. G. Goetz et al., "Movement Disorder Society-sponsored revision of the Unified Parkinson's Disease Rating Scale (MDS-UPDRS): Process, format, and clinimetric testing plan," Movement Disorders, vol. 22, no. 1, pp. 41-47, 2007.
[7] W. Poewe et al., "Parkinson disease," Nature Reviews Disease Primers, vol. 3, no. 1, p. 17013, Mar 2017.
[8] C. A. Davie, "A review of Parkinson's disease," British Medical Bulletin, vol. 86, no. 1, pp. 109-127, 2008.
[9] P. D. Mozley et al., "Binding of [99mTc]TRODAT-1 to dopamine transporters in patients with Parkinson's disease and in healthy volunteers," Journal of Nuclear Medicine Technology, vol. 41, no. 4, pp. 584-9, Apr 2000.
[10] L. Livieratos, "Basic Principles of SPECT and PET Imaging," in Radionuclide and Hybrid Bone Imaging, I. Fogelman, G. Gnanasegaran, and H. van der Wall Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2012, pp. 345-359.
[11] B. Hutton, "SPECT imaging: Basics and new trends," Handbook of Particle Detection and Imaging, p. 917, Jan 2012.
[12] L. W. Goldman, "Principles of CT and CT Technology," Journal of Nuclear Medicine Technology, vol. 35, no. 3, p. 115, 2007.
[13] M. Ljungberg and P. H. Pretorius, "SPECT/CT: an update on technological developments and clinical applications," British Journal of Radiology, vol. 91, no. 1081, p. 20160402, Jan 2018.
[14] H. Schöder, Y. E. Erdi, S. M. Larson, and H. W. Yeung, "PET/CT: a new imaging technology in nuclear medicine," European Journal of Nuclear Medicine and Molecular Imaging, vol. 30, no. 10, pp. 1419-37, Oct 2003.
[15] W. Koch, J. Hornung, C. Hamann, G. Pöpperl, and K. Tatsch, "Equipment-independent reference values for dopamine transporter imaging with 123I-FP-CIT," Nuklearmedizin, vol. 46, no. 3, pp. 107-11, 2007.
[16] J. Schmidhuber, "Deep learning in neural networks: An overview," Neural Networks, vol. 61, pp. 85-117, Jan 2015.
[17] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "ImageNet classification with deep convolutional neural networks," Commun. ACM, vol. 60, no. 6, pp. 84–90, 2017.
[18] P. E. Utgoff and D. J. Stracuzzi, "Many-Layered Learning," Neural Computation, vol. 14, no. 10, pp. 2497-2529, 2002.
[19] J. L. Elman, E. A. Bates, M. H. Johnson, A. Karmiloff-Smith, D. Parisi, and K. Plunkett, Rethinking Innateness: A Connectionist Perspective on Development. MIT Press, 1996.
[20] G. E. Hinton, "A Practical Guide to Training Restricted Boltzmann Machines," in Neural Networks: Tricks of the Trade: Second Edition, G. Montavon, G. B. Orr, and K.-R. Müller Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2012, pp. 599-619.
[21] Z. Boger and H. Guterman, "Knowledge extraction from artificial neural network models," in 1997 IEEE International Conference on Systems, Man, and Cybernetics., Oct 1997, vol. 4, pp. 3030-3035.
[22] C. Cao et al., "Deep Learning and Its Applications in Biomedicine," Genomics Proteomics Bioinformatics, vol. 16, no. 1, pp. 17-32, Feb 2018.
[23] H. R. Roth et al., "Anatomy-specific classification of medical images using deep convolutional nets," in 2015 IEEE 12th international symposium on biomedical imaging, 2015, pp. 101-104.
[24] Q. Dou et al., "Automatic detection of cerebral microbleeds from MR images via 3D convolutional neural networks," IEEE transactions on medical imaging, vol. 35, no. 5, pp. 1182-1195, 2016.
[25] J. M. Wolterink, T. Leiner, B. D. de Vos, R. W. van Hamersvelt, M. A. Viergever, and I. Išgum, "Automatic coronary artery calcium scoring in cardiac CT angiography using paired convolutional neural networks," Medical image analysis, vol. 34, pp. 123-136, 2016.
[26] R. Arunkumar and P. Karthigaikumar, "Multi-retinal disease classification by reduced deep learning features," Neural Computing and Applications, vol. 28, no. 2, pp. 329-334, 2017.
[27] H. Zeng, M. D. Edwards, G. Liu, and D. K. Gifford, "Convolutional neural network architectures for predicting DNA–protein binding," Bioinformatics, vol. 32, no. 12, pp. i121-i127, 2016.
[28] O. Ronneberger, P. Fischer, and T. Brox, "U-Net: Convolutional Networks for Biomedical Image Segmentation," in Medical Image Computing and Computer-Assisted Intervention, 2015, pp. 234-241.
[29] W. Wang, K. Yu, J. Hugonot, P. Fua, and M. Salzmann, "Recurrent U-Net for Resource-Constrained Segmentation," in 2019 IEEE/CVF International Conference on Computer Vision, Oct 2019, pp. 2142-2151.
[30] M. D. Zeiler and R. Fergus, "Visualizing and Understanding Convolutional Networks," in European Conference on Computer Vision, 2014, pp. 818-833.
[31] Z. Ma, D. Chang, and X. Li, "Channel max pooling layer for fine-grained vehicle classification," arXiv preprint arXiv:1902.11107, 2019.
[32] P. Zhang, Y. Ke, Z. Zhang, M. Wang, P. Li, and S. Zhang, "Urban Land Use and Land Cover Classification Using Novel Deep Learning Models Based on High Spatial Resolution Satellite Imagery," Sensors, vol. 18, no. 11, p. 3717, 2018.
[33] H. Li, Z. Xu, G. Taylor, C. Studer, and T. Goldstein, "Visualizing the loss landscape of neural nets," presented at the Proceedings of the 32nd International Conference on Neural Information Processing Systems, 2018.
[34] C. Makowski et al., "Evaluating accuracy of striatal, pallidal, and thalamic segmentation methods: Comparing automated approaches to manual delineation," Neuroimage, vol. 170, pp. 182-198, Apr 2018.
[35] Y.-L. Chao, H.-C. Lin, N.-T. Chiu, and S.-H. Wang, "Quantitative Evaluation of Parkinson’s Disease Using Striatal Binding Ratio of 99mTC-TRODAT-1 SPECT/CT Images Based on Deep Learning Approaches," M.A. thesis, National Cheng Kung University, Taiwan, 2020.
[36] Z. Zhu, C. Liu, D. Yang, A. Yuille, and D. Xu, "V-NAS: Neural Architecture Search for Volumetric Medical Image Segmentation," in 2019 International Conference on 3D Vision, Sept 2019, pp. 240-248.
[37] Ö. Çiçek, A. Abdulkadir, S. S. Lienkamp, T. Brox, and O. Ronneberger, "3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation," in Medical Image Computing and Computer-Assisted Intervention, 2016, pp. 424-432.
[38] W. Chen, B. Liu, S. Peng, J. Sun, and X. Qiao, "S3D-UNet: separable 3D U-Net for brain tumor segmentation," in International MICCAI Brainlesion Workshop, 2018: Springer, pp. 358-368.
[39] A. Faro, D. Giordano, C. Spampinato, S. Ullo, and A. D. Stefano, "Basal Ganglia Activity Measurement by Automatic 3-D Striatum Segmentation in SPECT Images," IEEE Transactions on Instrumentation and Measurement, vol. 60, no. 10, pp. 3269-3280, 2011.
[40] B.-H. Yang et al., "Automatic Striatal ROI Delineation for Semi-quantitative Analysis of 99mTc-TRODAT-1 Brain SPECT Imaging," Journal of Medical and Biological Engineering, vol. 40, May 2020.
[41] N. Tzourio-Mazoyer et al., "Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain," Neuroimage, vol. 15, no. 1, pp. 273-89, Jan 2002.
[42] D. Kupitz et al., "Global scaling for semi-quantitative analysis in FP-CIT SPECT," Nuklearmedizin. Nuclear medicine, vol. 53 6, pp. 234-41, 2014.
[43] J. Protasiewicz. "Python AI: Why Is Python So Good for Machine Learning?" https://www.netguru.com/blog/python-machine-learning (accessed Aug, 2021).
[44] L. Huang, J. Qin, Y. Zhou, F. Zhu, L. Liu, and L. Shao, "Normalization techniques in training dnns: Methodology, analysis and application," arXiv preprint arXiv:2009.12836, 2020.
[45] U. Ruby and V. Yendapalli, "Binary cross entropy with deep learning technique for image classification," International Journal of Advanced Trends in Computer Science and Engineering, vol. 9, no. 10, 2020.
[46] S. Jadon, "A survey of loss functions for semantic segmentation," in 2020 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology, 2020, pp. 1-7.
[47] S. Kosub, "A note on the triangle inequality for the Jaccard distance," Pattern Recognition Letters, vol. 120, pp. 36-38, Apr 2019.
[48] G. Song, Y. Nie, J. Zhang, and G. Chen, "Research on the fusion method of 2D and 3D UNet in pulmonary nodules segmentation task," in 2020 International Conference on Computer Science and Management Technology, Nov 2020, pp. 44-47.
[49] Z. H. Cho, 7.0 Tesla MRI brain white matter atlas, Second edition ed. Berlin, Heidelberg: Springer, 2015, p. 457.
[50] E. Beauxis-Aussalet and L. Hardman, "Simplifying the visualization of confusion matrix," in 26th Benelux Conference on Artificial Intelligence, 2014.