醫學影像分析中，大腦皮質下的分割是其中一項重要的任務。許多神經、神經衰退疾病與皮質下的區域相關，像是阿茲海默症、自閉症、亨丁頓舞蹈症等，要是能快速並準確的分割，能夠在早期幫助專家進行診斷或評估。經由專家的手動分割被認為是一種標準，但是通常相當費工與耗時，難以應用在較大的資料集上，因此顯現了自動化皮質下分割的重要性。在這篇論文中，我們利用7039筆3D T1w MRI的搭配FreeSurfer產出的分割結果作為輔助標籤訓練3D U-Net網路，經由四個實驗對模型進行最佳化，透過不同面向去評估模型選出最好的模型，並嘗試將模型應用在臨床的研究上，經由預測ADNI資料集的大腦體積進一步分析阿茲海默症患者與正常受試者的海馬迴體積衰減差異，觀察分割結果與實際病理狀況是否相符合。最終，透過多面向的實驗得出了一個快速、穩定、準確度高的模型，其分割結果與手動分割的體積誤差較低，並且實際應用在病理影像上能成功觀察出相對應的趨勢，在臨床的研究上將有利於專家進行快速的分析。

Subcortical brain segmentation is considered an important task in neuroimaging analysis. Many neurological and neurodegenerative diseases are related to subcortical areas, such as Alzheimer's disease, Autistic spectrum disorder, Huntington’s disease, etc. If the subcortical region can be segmented quickly and accurately, it will help experts diagnose or assess the condition early. Manual segmentation by experts is considered the gold standard. Still, it is often labor-intensive, time-consuming, and difficult to apply to larger datasets, thus highlighting the importance of automated subcortical segmentation. In this study, we used 7039 3D T1-weighted MRIs with FreeSurfer's segmentation results as auxiliary labels to train 3D U-Net, optimized the model through four experiments, evaluated the model through different aspects to select the best model, and tried to apply the model to clinical research. By predicting the brain volume of the ADNI dataset, we further analyzed the difference in the volume attenuation of the hippocampus between Alzheimer's patients and normal subjects and observed whether the segmentation results were consistent with the actual pathological conditions. Finally, a fast, stable, and high-accuracy model were obtained through multi-faceted experiments. The volume error of the segmentation results and manual segmentation was lower than FreeSurfer. The actual application of the model can successfully observe the corresponding trend in pathological images, which can be helpful for experts to conduct rapid analysis in clinical research.

Reference
[1] A. Convit et al., "Specific hippocampal volume reductions in individuals at risk for Alzheimer’s disease," Neurobiology of aging, vol. 18, no. 2, pp. 131-138, 1997.
[2] D. Van Rooij et al., "Cortical and subcortical brain morphometry differences between patients with autism spectrum disorder and healthy individuals across the lifespan: results from the ENIGMA ASD Working Group," American Journal of Psychiatry, vol. 175, no. 4, pp. 359-369, 2018.
[3] L. Schmaal et al., "Subcortical brain alterations in major depressive disorder: findings from the ENIGMA Major Depressive Disorder working group," Molecular psychiatry, vol. 21, no. 6, pp. 806-812, 2016.
[4] N. Schuff et al., "MRI of hippocampal volume loss in early Alzheimer's disease in relation to ApoE genotype and biomarkers," Brain, vol. 132, no. 4, pp. 1067-1077, 2009.
[5] G. Sánchez-Benavides, B. Gómez-Ansón, A. Sainz, Y. Vives, M. Delfino, and J. Peña-Casanova, "Manual validation of FreeSurfer's automated hippocampal segmentation in normal aging, mild cognitive impairment, and Alzheimer Disease subjects," Psychiatry Research: Neuroimaging, vol. 181, no. 3, pp. 219-225, 2010.
[6] E. R. Mulder et al., "Hippocampal volume change measurement: quantitative assessment of the reproducibility of expert manual outlining and the automated methods FreeSurfer and FIRST," Neuroimage, vol. 92, pp. 169-181, 2014.
[7] M. Jenkinson, C. F. Beckmann, T. E. Behrens, M. W. Woolrich, and S. M. Smith, "Fsl," Neuroimage, vol. 62, no. 2, pp. 782-790, 2012.
[8] B. Fischl, "FreeSurfer," Neuroimage, vol. 62, no. 2, pp. 774-781, 2012.
[9] J. P. Guenette et al., "Automated versus manual segmentation of brain region volumes in former football players," Neuroimage: clinical, vol. 18, pp. 888-896, 2018.
[10] A. B. Waters, R. A. Mace, K. S. Sawyer, and D. A. Gansler, "Identifying errors in Freesurfer automated skull stripping and the incremental utility of manual intervention," Brain imaging and behavior, vol. 13, no. 5, pp. 1281-1291, 2019.
[11] J. B. Bae et al., "Identification of Alzheimer's disease using a convolutional neural network model based on T1-weighted magnetic resonance imaging," Scientific Reports, vol. 10, no. 1, pp. 1-10, 2020.
[12] P. Rajpurkar et al., "Chexnet: Radiologist-level pneumonia detection on chest x-rays with deep learning," arXiv preprint arXiv:1711.05225, 2017.
[13] N. J. Tustison et al., "N4ITK: improved N3 bias correction," IEEE transactions on medical imaging, vol. 29, no. 6, pp. 1310-1320, 2010.
[14] G. Yang et al., "DAGAN: deep de-aliasing generative adversarial networks for fast compressed sensing MRI reconstruction," IEEE transactions on medical imaging, vol. 37, no. 6, pp. 1310-1321, 2017.
[15] J. Dolz, C. Desrosiers, and I. Ben Ayed, "3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study," Neuroimage, vol. 170, pp. 456-470, Apr 15 2018, doi: 10.1016/j.neuroimage.2017.04.039.
[16] M. Lee et al., "Split-attention U-Net: a fully convolutional network for robust multi-label segmentation from brain MRI," Brain Sciences, vol. 10, no. 12, p. 974, 2020.
[17] P. Coupé et al., "AssemblyNet: A large ensemble of CNNs for 3D whole brain MRI segmentation," NeuroImage, vol. 219, p. 117026, 2020.
[18] C. Dai, Y. Mo, E. Angelini, Y. Guo, and W. Bai, "Transfer learning from partial annotations for whole brain segmentation," in Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data: Springer, 2019, pp. 199-206.
[19] A. Kaku et al., "DARTS: DenseUnet-based automatic rapid tool for brain segmentation," arXiv preprint arXiv:1911.05567, 2019.
[20] H. Li, H. Zhang, H. Johnson, J. D. Long, J. S. Paulsen, and I. Oguz, "Mri subcortical segmentation in neurodegeneration with cascaded 3d cnns," in Medical Imaging 2021: Image Processing, 2021, vol. 11596: SPIE, pp. 236-243.
[21] 翁振軒, "以深度學習法進行大腦皮質下分割之多面向研究," 碩士, 電機工程系, 國立臺灣科技大學, 台北市, 2021. [Online]. Available: https://hdl.handle.net/11296/swu8yn
[22] A. Di Martino et al., "The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism," Molecular psychiatry, vol. 19, no. 6, pp. 659-667, 2014.
[23] D. N. Kennedy, C. Haselgrove, S. M. Hodge, P. S. Rane, N. Makris, and J. A. Frazier, "CANDIShare: a resource for pediatric neuroimaging data," ed: Springer, 2012.
[24] A. Klein and J. Tourville, "101 labeled brain images and a consistent human cortical labeling protocol," Frontiers in neuroscience, vol. 6, p. 171, 2012.
[25] D. S. Marcus, T. H. Wang, J. Parker, J. G. Csernansky, J. C. Morris, and R. L. Buckner, "Open Access Series of Imaging Studies (OASIS): cross-sectional MRI data in young, middle aged, nondemented, and demented older adults," Journal of cognitive neuroscience, vol. 19, no. 9, pp. 1498-1507, 2007.
[26] A. Klein, T. Dal Canton, S. S. Ghosh, B. Landman, J. Lee, and A. Worth, "Open labels: online feedback for a public resource of manually labeled brain images," in 16th Annual Meeting for the Organization of Human Brain Mapping, 2010, vol. 84358.
[27] S. Duchesne et al., "The Canadian dementia imaging protocol: harmonizing national cohorts," Journal of Magnetic Resonance Imaging, vol. 49, no. 2, pp. 456-465, 2019.
[28] R. C. Petersen et al., "Alzheimer's disease neuroimaging initiative (ADNI): clinical characterization," Neurology, vol. 74, no. 3, pp. 201-209, 2010.
[29] J. G. Sled, A. P. Zijdenbos, and A. C. Evans, "A nonparametric method for automatic correction of intensity nonuniformity in MRI data," IEEE transactions on medical imaging, vol. 17, no. 1, pp. 87-97, 1998.
[30] Ö. Çiçek, A. Abdulkadir, S. S. Lienkamp, T. Brox, and O. Ronneberger, "3D U-Net: learning dense volumetric segmentation from sparse annotation," in International conference on medical image computing and computer-assisted intervention, 2016: Springer, pp. 424-432.
[31] H. B. Mann and D. R. Whitney, "On a test of whether one of two random variables is stochastically larger than the other," The annals of mathematical statistics, pp. 50-60, 1947.