基于机器学习的磁共振帕金森指数的自动测量方法研究

辨别和诊断帕金森症和帕金森叠加综合症现在是医学界面临的一项重
要挑战。 及时辨别帕金森病和帕金森叠加综合征需要准确测量磁共振帕金
森指数（ MRPI）。长久以来， MRPI 由于人工测量步骤繁多而无法在临床
上得到大规模推广。已有的自动测量算法无法满足临床实践要求的精度，
并且仍然依赖测量操作员的辅助决策。在本研究中，我们基于已有的阈值
分割 MRPI 测量法给出了一种改进方案。我们的改进方案使用了最先进的
基于深度学习的图像分割网络 nnUNet 和关键点检测网络 HRNet，使得
MRPI 的自动测量精度大幅度提升。 由于 MRPI 测量流程复杂并且对误差极
其敏感， HRNet 直接作用于中脑脑桥边界点检测无法满足误差要求。我们
在 HRNet 引入了一种专门用来评价中脑脑桥划分效果的异或损失函数，将
关键点检测的误差对 MRPI 测量的负面影响降到了最低。
我们对比了传统 MRPI 测量方法和改进的 MRPI 测量方法，改进后的
方法在测量准确性和稳定性上有了很大提升。改进方法在 465 个 MRI 组成
的数据集上把测量的相对误差从 32.23%降低到了 18.17%，把误差的标准
差从 2.07 降低到了 1.30。我们邀请了两位临床医生手动测量 MRPI 作为对
照，并将两种方法和医生的测量值进行相关性分析，其中改进方法与手动
标注的相关性与医生之间的相关在同一水平上。改进方法的测量误差与手
动测量的误差在统计学上没有显著性差异（ p-value>0.05）。因此，本文给
出的 MRPI 测量方法可以代替人工测量给出可靠的 MRPI 自动测量值。完
全自动并且准确可靠的 MRPI 测量算法将会极大地方便帕金森病和帕金森
叠加综合征的诊断。
 

[1] JANKOVIC J. Parkinson’s disease: clinical features and diagnosis[J]. Journal of neurology, neurosurgery & psychiatry, 2008, 79(4): 368-376.
[2] WILLIAMS D R, LEES A J. Progressive supranuclear palsy: clinicopathological concepts and diagnostic challenges[J]. The Lancet Neurology, 2009, 8(3): 270-279.
[3] WENNING G K, STEFANOVA N, JELLINGER K A, et al. Multiple system atrophy: a primary oligodendrogliopathy[J]. Annals of Neurology: Official Journal of the American Neurological Association and the Child Neurology Society, 2008, 64(3): 239-246.
[4] SJOSTROM A C, HOLMBERG B, STRANG P. Parkinson-plus patients--an unknown group with severe symptoms[J]. Journal of Neuroscience Nursing, 2002, 34(6): 314.
[5] MENDOZA-SANTIESTEBAN C E, GABILONDO I, PALMA J A, et al. The retina in multiple system atrophy: systematic review and meta-analysis[J]. Frontiers in neurology, 2017, 8: 206.
[6] CRONIN-GOLOMB A, REYNOLDS G O, SALAZAR R D, et al. Parkinson’s disease and Parkinson-plus syndromes[M]//The Oxford handbook of adult cognitive disorders. 2019.
[7] AYGUN D, KOCABICAK E, YILDIZ M O, et al. Effect of age and disease duration on the levodopa response in patients with advanced Parkinson’s disease for deep brain stimulation of the subthalamic nucleus[J]. Frontiers in Neurology, 2016: 97.
[8] KOGA S, AOKI N, UITTI R J, et al. When DLB, PD, and PSP masquerade as MSA: an autopsy study of 134 patients[J]. Neurology, 2015, 85(5): 404-412.
[9] CONSTANTINIDES V C, PARASKEVAS G P, VELONAKIS G, et al. MRI planimetry and Magnetic Resonance Parkinsonism Index in the differential diagnosis of patients with parkinsonism[J]. American Journal of Neuroradiology, 2018, 39(6): 1047-1051.
[10] HORIMOTO Y, AIBA I, YASUDA T, et al. Longitudinal MRI study of multiple system atrophy–when do the findings appear, and what is the course?[J]. Journal of neurology, 2002, 249(7): 847-854.
[11] PAVIOUR D C, PRICE S L, JAHANSHAHI M, et al. Regional brain volumes distinguish PSP, MSA‐P, and PD: MRI‐based clinico‐radiological correlations[J]. Movement disorders: official journal of the Movement Disorder Society, 2006, 21(7): 989-996.
[12] OBA H, YAGISHITA A, TERADA H, et al. New and reliable MRI diagnosis for progressive supranuclear palsy[J]. Neurology, 2005, 64(12): 2050-2055.
[13] WARMUTH-METZ M, NAUMANN M, CSOTI I, et al. Measurement of the midbrain diameter on routine magnetic resonance imaging: a simple and accurate method of differentiating between Parkinson disease and progressive supranuclear palsy[J]. Archives of neurology, 2001, 58(7): 1076-1079.
[14] 于静, 王晓, 胡君, 等. MRI 形态学测量对帕金森病, 多系统萎缩及进行性核上性麻痹的鉴别诊断价值[J]. 临床神经病学杂志, 2019, 32(6): 406-409.
[15] QUATTRONE A, NICOLETTI G, MESSINA D, et al. MR imaging index for differentiation of progressive supranuclear palsy from Parkinson disease and the Parkinson variant of multiple system atrophy[J]. Radiology, 2008, 246(1): 214-221.
[16] MORELLI M, ARABIA G, SALSONE M, et al. Accuracy of magnetic resonance parkinsonism index for differentiation of progressive supranuclear palsy from probable or possible Parkinson disease[J]. Movement Disorders, 2011, 26(3): 527-533.
[17] WARMUTH-METZ M, NAUMANN M, CSOTI I, et al. Measurement of the midbrain diameter on routine magnetic resonance imaging: a simple and accurate method of differentiating between Parkinson disease and progressive supranuclear palsy[J]. Archives of neurology, 2001, 58(7): 1076-1079.
[18] NICOLETTI G, FERA F, CONDINO F, et al. MR imaging of middle cerebellar peduncle width: differentiation of multiple system atrophy from Parkinson disease[J]. Radiology, 2006, 239(3): 825-830.
[19] PAVIOUR D C, PRICE S L, STEVENS J M, et al. Quantitative MRI measurement of superior cerebellar peduncle in progressive supranuclear palsy[J]. Neurology, 2005, 64(4): 675-679.
[20] NIGRO S, CERASA A, ZITO G, et al. Fully automated segmentation of the pons and midbrain using human T1 MR brain images[J]. PloS one, 2014, 9(1): e85618.
[21] OTSU N. A threshold selection method from gray-level histograms[J]. IEEE transactions on systems, man, and cybernetics, 1979, 9(1): 62-66.
[22] IGLESIAS J E, VAN LEEMPUT K, BHATT P, et al. Bayesian segmentation of brainstem structures in MRI[J]. Neuroimage, 2015, 113: 184-195.
[23] NIGRO S, ARABIA G, ANTONINI A, et al. Magnetic Resonance Parkinsonism Index: diagnostic accuracy of a fully automated algorithm in comparison with the manual measurement in a large Italian multicentre study in patients with progressive supranuclear palsy[J]. European radiology, 2017, 27(6): 2665-2675.
[24] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 3431-3440.
[25] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015: 234-241.
[26] ISENSEE F, PETERSEN J, KLEIN A, et al. nnu-net: Self-adapting framework for u-net-based medical image segmentation[J]. arXiv preprint arXiv:1809.10486, 2018.
[27] OKTAY O, SCHLEMPER J, FOLGOC L L, et al. Attention u-net: Learning where to look for the pancreas[J]. arXiv preprint arXiv:1804.03999, 2018.
[28] ÇIÇEK Ö, ABDULKADIR A, LIENKAMP S S, et al. 3D U-Net: learning dense volumetric segmentation from sparse annotation[C]//International conference on medical image computing and computer-assisted intervention. Springer, Cham, 2016: 424-432.
[29] ZHU W, HUANG Y, ZENG L, et al. AnatomyNet: deep learning for fast and fully automated whole‐volume segmentation of head and neck anatomy[J]. Medical physics, 2019, 46(2): 576-589.
[30] DAMADIAN R. Tumor detection by nuclear magnetic resonance[J]. Science, 1971, 171(3976): 1151-1153.
[31] HAZLEWOOD C F, CHANG D C, MEDINA D, et al. Distinction between the preneoplastic and neoplastic state of murine mammary glands[J]. Proceedings of the National Academy of Sciences, 1972, 69(6): 1478-1480.
[32] LAUTERBUR P C. Image formation by induced local interactions: examples employing nuclear magnetic resonance[J]. nature, 1973, 242(5394): 190-191.
[33] HUBEL D H, WIESEL T N. Receptive fields, binocular interaction and functional architecture in the cat's visual cortex[J]. The Journal of physiology, 1962, 160(1): 106.
[34] FUKUSHIMA K, MIYAKE S. Neocognitron: A self-organizing neural network model for a mechanism of visual pattern recognition[M]//Competition and cooperation in neural nets. Springer, Berlin, Heidelberg, 1982: 267-285.
[35] ATLAS L, HOMMA T, MARKS R. An artificial neural network for spatio-temporal bipolar patterns: Application to phoneme classification[C]//Neural Information Processing Systems. 1987.
[36] LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
[37] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Advances in neural information processing systems, 2012, 25.
[38] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv:1409.1556, 2014.
[39] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 3431-3440.
[40] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015: 234-241.
[41] ISENSEE F, PETERSEN J, KLEIN A, et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation[J]. Nature methods, 2021, 18(2): 203-211.
[42] SUN K, XIAO B, LIU D, et al. Deep high-resolution representation learning for human pose estimation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 5693-5703.
[43] MAREK K, CHOWDHURY S, SIDEROWF A, et al. The Parkinson's progression markers initiative (PPMI)–establishing a PD biomarker cohort[J]. Annals of clinical and translational neurology, 2018, 5(12): 1460-1477.
[44] JENKINSON M, BANNISTER P, BRADY M, et al. Improved optimization for the robust and accurate linear registration and motion correction of brain images[J]. Neuroimage, 2002, 17(2): 825-841.