Nội dung được dịch bởi AI, chỉ mang tính chất tham khảo
Phân đoạn tự động các động mạch lớn cho đánh giá huyết động học tính toán
Tóm tắt
Động lực học chất lỏng tính toán (CFD) ngày càng được sử dụng để đánh giá điều kiện lưu lượng máu ở bệnh nhân mắc bệnh tim bẩm sinh (CHD). Điều này yêu cầu hình thái học cụ thể của bệnh nhân, thường được thu thập từ hình ảnh cộng hưởng từ mạch máu 3D (CMR) đã được phân đoạn. Tuy nhiên, quy trình phân đoạn mất nhiều thời gian và cần sự tham gia của chuyên gia. Nghiên cứu này nhằm phát triển và xác nhận một phương pháp học máy (ML) cho việc phân đoạn động mạch chủ và động mạch phổi để ứng dụng CFD. 90 bệnh nhân CHD đã được chọn lựa theo cách hồi cứu cho nghiên cứu này. Hình ảnh CMR 3D được phân đoạn thủ công để thu được nền tảng đúng (GT), nhãn động mạch chủ và động mạch phổi. Những dữ liệu này đã được sử dụng để đào tạo và tối ưu hóa mô hình U-Net, sử dụng phân chia 70-10-10 cho đào tạo - xác thực - kiểm tra. Hiệu suất phân đoạn chủ yếu được đánh giá bằng điểm Dice. Các mô phỏng CFD đã được thiết lập từ phân đoạn GT và ML sử dụng một quy trình lưới và mô phỏng bán tự động. Áp lực trung bình và các trường vận tốc trên 99 mặt phẳng dọc theo đường tâm của mạch đã được trích xuất, và tỷ lệ sai số phần trăm trung bình (MAPE) được tính cho mỗi cặp mạch (ML so với GT). Một người quan sát thứ hai (SO) đã phân đoạn tập dữ liệu kiểm tra để đánh giá sự biến thiên giữa các quan sát viên. Các bài kiểm tra Friedman đã được sử dụng để so sánh các số liệu giữa ML so với GT, SO so với GT và ML so với SO, cũng như các sai số trong trường áp lực/vận tốc. Điểm Dice của mạng (ML so với GT) là 0.945 (phạm vi quý tứ: 0.929–0.955) cho động mạch chủ và 0.885 (0.851–0.899) cho động mạch phổi. Sự khác biệt với điểm Dice giữa các quan sát viên (SO so với GT) và điểm Dice ML so với SO không có ý nghĩa thống kê đối với cả động mạch chủ và động mạch phổi (p = 0.741, p = 0.061). Các MAPE ML so với GT cho áp suất và vận tốc ở động mạch chủ lần lượt là 10.1% (8.5–15.7%) và 4.1% (3.1–6.9%), trong khi đối với động mạch phổi là 14.6% (11.5–23.2%) và 6.3% (4.3–7.9%), tương ứng. Các MAPE giữa các quan sát viên (SO so với GT) và ML so với SO cho áp suất và vận tốc có cùng độ lớn với ML so với GT (p > 0.2). Phương pháp ML có thể phân đoạn thành công các mạch lớn cho CFD, với sai số tương tự như sự biến thiên giữa các quan sát viên. Phương pháp nhanh chóng, tự động này giảm thời gian và công sức cần thiết cho phân tích CFD, làm cho nó trở nên hấp dẫn hơn cho việc sử dụng lâm sàng thường quy.
Từ khóa
#động lực học chất lỏng #bệnh tim bẩm sinh #phân đoạn tự động #học máy #mô phỏng CFDTài liệu tham khảo
Biglino G, Capelli C, Bruse J, Bosi GM, Taylor AM, Schievano S. Computational modelling for congenital heart disease: how far are we from clinical translation? Heart. 2017;103:98–103. https://doi.org/10.1136/heartjnl-2016-310423.
Wang C, Frimmel H, Smedby Ö. Fast level-set based image segmentation using coherent propagation. Med Phys. 2014;41. https://doi.org/10.1118/1.4881315.
Weese J, Kaus M, Lorenz C, Lobregt S, Truyen R, Pekar V. Shape Constrained Deformable Models for 3D Medical Image Segmentation. Lect Notes Comput Sci (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). 2001;2082:380–7. https://doi.org/10.1007/3-540-45729-1_38.
Peters J, Ecabert O, Meyer C, Schramm H, Kneser R, Groth A, et al. Automatic Whole Heart Segmentation in Static Magnetic Resonance Image Volumes. Lect Notes Comput Sci (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 2007;4792 LNCS:402–10. https://doi.org/10.1007/978-3-540-75759-7_49.
Chen C, Qin C, Qiu H, Tarroni G, Duan J, Bai W, et al. Deep learning for cardiac image segmentation: a review. Front Cardiovasc Med. 2020:25. https://doi.org/10.3389/FCVM.2020.00025.
Berhane H, Scott M, Elbaz M, Jarvis K, McCarthy P, Carr J, et al. Fully automated 3D aortic segmentation of 4D flow MRI for hemodynamic analysis using deep learning. Magn Reson Med. 2020;84:2204–18. https://doi.org/10.1002/MRM.28257.
Zhuang X, Li L, Payer C, Štern D, Urschler M, Heinrich MP, et al. Evaluation of algorithms for multi-modality whole heart segmentation: an open-access grand challenge. Med Image Anal. 2019;58: 101537. https://doi.org/10.1016/J.MEDIA.2019.101537.
Bai W, Sinclair M, Tarroni G, Oktay O, Rajchl M, Vaillant G, et al. Automated cardiovascular magnetic resonance image analysis with fully convolutional networks. J Cardiovasc Magn Resonan. 2018;20:1–12. https://doi.org/10.1186/S12968-018-0471-X.
Böttcher B, Beller E, Busse A, Cantré D, Yücel S, Öner A, et al. Fully automated quantification of left ventricular volumes and function in cardiac MRI: clinical evaluation of a deep learning-based algorithm. Int J Cardiovasc Imaging. 2020;36:2239–47. https://doi.org/10.1007/S10554-020-01935-0.
Steeden JA, Quail M, Gotschy A, Mortensen KH, Hauptmann A, Arridge S, et al. Rapid whole-heart CMR with single volume super-resolution. J Cardiovasc Magn Reson. 2020;22:56. https://doi.org/10.1186/s12968-020-00651-x.
Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 9351, Springer Verlag; 2015, p. 234–41. https://doi.org/10.1007/978-3-319-24574-4_28.
Li L, Jamieson K, Rostamizadeh A, Talwalkar A. Hyperband: a novel bandit-based approach to hyperparameter optimization. J Mach Learn Res. 2018;18:1–52.
Abadi M, Agarwal A, Barham P, Brevdo E, Chen Z, Citro C, et al. TensorFlow: large-scale machine learning on heterogeneous distributed systems. 2015.
Montalt-Tordera J. TensorFlow MRI 2021. https://doi.org/10.5281/zenodo.5151590.
He K, Zhang X, Ren S, Sun J. Delving deep into rectifiers: surpassing human-level performance on ImageNet classification. In: Proceedings of the IEEE International Conference on Computer Vision, Institute of Electrical and Electronics Engineers Inc.; 2015, p. 1026–34. https://doi.org/10.1109/ICCV.2015.123.
Kingma DP, Ba JL. Adam: A method for stochastic optimization. In: 3rd International Conference on Learning Representations, ICLR 2015—Conference Track Proceedings, International Conference on Learning Representations, ICLR; 2015.
Silversmith W. cc3d: connected components on multilabel 3D images 2021.
Gijsen FJH, Van De Vosse FN, Janssen JD. The influence of the non-Newtonian properties of blood on the flow in large arteries: steady flow in a carotid bifurcation model. J Biomech. 1999;32:601–8. https://doi.org/10.1016/S0021-9290(99)00015-9.
Caballero AD, Laín S. A review on computational fluid dynamics modelling in human thoracic aorta. Cardiovasc Eng Technol. 2013;4:103–30. https://doi.org/10.1007/S13239-013-0146-6.
Bonfanti M, Balabani S, Greenwood JP, Puppala S, Homer-Vanniasinkam S, Díaz-Zuccarini V. Computational tools for clinical support: a multi-scale compliant model for haemodynamic simulations in an aortic dissection based on multi-modal imaging data. J R Soc Interface 2017;14. https://doi.org/10.1098/RSIF.2017.0632.
Gabe IT, Gault JH, Ross J, Mason DT, Mills CJ, Schillingford JP, et al. Measurement of instantaneous blood flow velocity and pressure in conscious man with a catheter-tip velocity probe. Circulation. 1969;40:603–14. https://doi.org/10.1161/01.CIR.40.5.603.
Payer C, Štern D, Bischof H, Urschler M. Multi-label whole heart segmentation using CNNS and anatomical label configurations. Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 2017;10663 LNCS:190–8. https://doi.org/10.1007/978-3-319-75541-0_20.
Sun C, Shrivastava A, Singh S, Gupta A. Revisiting unreasonable effectiveness of data in deep learning era. In: Proceedings of the IEEE International Conference on Computer Vision, vol. 2017–October, Institute of Electrical and Electronics Engineers Inc.; 2017, p. 843–52. https://doi.org/10.1109/ICCV.2017.97.
Krizhevsky A, Sutskever I, Hinton GE. ImageNet classification with deep convolutional neural networks. Adv Neural Inform Process Syst. 2012;25.
Bhalodia R, Elhabian SY, Kavan L, Whitaker RT. DeepSSM: a deep learning framework for statistical shape modeling from raw images. Lecture Notes Comput Sci. (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 2018;11167 LNCS:244–57.
Raju A, Miao S, Cheng C-T, Lu L, Han M, Xiao J, et al. Deep implicit statistical shape models for 3D medical image delineation 2021. AAAI. 2022;36:2135–43.
Zhu JY, Park T, Isola P, Efros AA. Unpaired image-to-image translation using cycle-consistent adversarial networks. In: Proceedings of the IEEE International Conference on Computer Vision 2017;2017–October:2242–51. https://doi.org/10.48550/arxiv.1703.10593.
Yan W, Wang Y, Gu S, Huang L, Yan F, Xia L, et al. The domain shift problem of medical image segmentation and vendor-adaptation by Unet-GAN 2019. https://doi.org/10.48550/arxiv.1910.13681.
Liu L, Wolterink JM, Brune C, Veldhuis RNJ. Anatomy-aided deep learning for medical image segmentation: a review. Phys Med Biol. 2021;66:11TR01. https://doi.org/10.1088/1361-6560/ABFBF4.