Truy xuất hình ảnh dựa trên nội dung để chẩn đoán hình ảnh tưới máu cơ tim sử dụng mạng nơ-ron tích chập tự mã hóa sâu

Journal of Nuclear Cardiology - Tập 30 - Trang 540-549 - 2022
Akinori Higaki1,2, Naoto Kawaguchi3, Tsukasa Kurokawa1, Hikaru Okabe1, Takuro Kazatani1, Shinsuke Kido1, Tetsuya Aono1, Kensho Matsuda1, Yuta Tanaka1, Saki Hosokawa1, Tetsuya Kosaki1, Go Kawamura1, Tatsuya Shigematsu1, Yoshitaka Kawada1, Go Hiasa1, Tadakatsu Yamada1, Hideki Okayama1
1Department of Cardiology, Ehime Prefectural Central Hospital, Matsuyama, Japan
2Department of Cardiology, Pulmonology, Hypertension & Nephrology, Ehime University Graduate School of Medicine, Toon, Japan
3Department of Radiology, Ehime University Graduate School of Medicine, Toon, Japan

Tóm tắt

Hình ảnh tưới máu cơ tim (MPI) bằng chụp cắt lớp phát xạ đơn photon (SPECT) đóng vai trò quan trọng trong chiến lược điều trị tối ưu cho bệnh nhân mắc bệnh tim mạch vành. Chúng tôi đã kiểm tra tính khả thi của việc trích xuất đặc trưng từ MPI bằng cách sử dụng mô hình tự mã hóa tích chập sâu (CAE). Tám trăm bốn mươi ba cặp hình ảnh tưới máu cơ tim trong tình trạng stress và nghỉ đã được thu thập từ những bệnh nhân liên tiếp thực hiện chụp xạ tim tại bệnh viện của chúng tôi từ tháng 12 năm 2019 đến tháng 2 năm 2022. Chúng tôi đã huấn luyện mô hình CAE để tái tạo dữ liệu hình ảnh cặp đầu vào, giúp cho bộ mã hóa xuất ra một vector đặc trưng 256 chiều. Các vector đặc trưng được giảm chiều thêm thông qua phân tích thành phần chính (PCA) để trực quan hóa dữ liệu. Truy xuất hình ảnh dựa trên nội dung (CBIR) được thực hiện dựa trên độ tương đồng cosine của các vector đặc trưng giữa hình ảnh truy vấn và hình ảnh tham chiếu. Mức độ nhất quán giữa kết quả tìm kiếm của bác sĩ chẩn đoán hình ảnh giữa MPI truy vấn và MPI được truy xuất đã được đánh giá bằng độ chính xác nhị phân, độ chính xác, độ thu hồi và điểm F1. Một biểu đồ phân tán ba chiều với PCA đã chỉ ra rằng các vector đặc trưng giữ lại thông tin lâm sàng như điểm số chênh lệch tổng hợp phần trăm, sự hiện diện của thiếu máu, và vị trí của sẹo do các bác sĩ chẩn đoán hình ảnh báo cáo. Khi CBIR được sử dụng như một công cụ chẩn đoán dựa trên độ tương đồng, độ chính xác nhị phân đạt 81,0%. Kết quả cho thấy tính hữu ích của việc học đặc trưng không giám sát cho CBIR trong MPI.

Từ khóa

#hình ảnh tưới máu cơ tim #chụp cắt lớp phát xạ đơn photon #mạng nơ-ron tích chập #truy xuất hình ảnh dựa trên nội dung #phân tích thành phần chính #học đặc trưng không giám sát

Tài liệu tham khảo

Hachamovitch R, Rozanski A, Shaw LJ, Stone GW, Thomson LE, Friedman JD. Impact of ischaemia and scar on the therapeutic benefit derived from myocardial revascularization vs. medical therapy among patients undergoing stress-rest myocardial perfusion scintigraphy. Eur Heart J 2011;32:1012‐24. Seifert R, Weber M, Kocakavuk E, Rischpler C, Kersting D. Artificial intelligence and machine learning in nuclear medicine: Future perspectives. Semin Nucl Med 2021;51:170‐7. Guner LA, et al. An open-source framework of neural networks for diagnosis of coronary artery disease from myocardial perfusion SPECT. J Nucl Cardiol 2010;17:405‐13. Betancur J, et al. Deep learning for prediction of obstructive disease from fast myocardial perfusion SPECT: A multicenter study. JACC Cardiovasc Imaging 2018;11:1654‐63. Kaplan Berkaya S, Ak Sivrikoz I, Gunal S. Classification models for SPECT myocardial perfusion imaging. Comput Biol Med 2020;123:103893. Sengupta PP, et al. Proposed requirements for cardiovascular imaging-related machine learning evaluation (PRIME): A checklist. JACC Cardiovasc Imaging 2020;13:2017‐35. Pasini A. Artificial neural networks for small dataset analysis. J Thorac Dis 2015;7:953‐60. Müller H, Michoux N, Bandon D, Geissbuhler A. A review of content-based image retrieval systems in medical applications: Clinical benefits and future directions. Int J Med Inform 2004;73:1‐23. Dhara AK, Mukhopadhyay S, Dutta A, Garg M, Khandelwal N. Content-based image retrieval system for pulmonary nodules: assisting radiologists in self-learning and diagnosis of lung cancer. J Digit Imaging 2017;30:63‐77. Daoud MI, Saleh A, Hababeh I, Alazrai R, Content-based image retrieval for breast ultrasound images using convolutional autoencoders: A feasibility study. BioSMART 2019 - Proc. 3rd Int. Conf. Bio-Engineering Smart Technol. (2019). https://doi.org/10.1109/BIOSMART.2019.8734190. Klocke FJ, et al. ACC/AHA/ASNC guidelines for the clinical use of cardiac radionuclide imaging–executive summary: A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (ACC/AHA/ASNC Committee to Revise the 1995 Guideli. Circulation 2003;108:1404‐18. Cerqueira MD, et al. Standardized myocardial segmentation and nomenclature for tomographic imaging of the heart. J Cardiovasc Magn Reson 2002;4:203‐10. Nakajima K, et al. Creation and characterization of Japanese standards for myocardial perfusion SPECT: Database from the Japanese Society of Nuclear Medicine Working Group. Ann Nucl Med 2007;21:505‐11. Moroi M, Yamashina A, Tsukamoto K, Nishimura T, J-ACCESS Investigators. Coronary revascularization does not decrease cardiac events in patients with stable ischemic heart disease but might do in those who showed moderate to severe ischemia. Int J Cardiol 2012;158:246‐52. Maron DJ, et al. International study of comparative health effectiveness with medical and invasive approaches (ISCHEMIA) trial: Rationale and design. Am Heart J 2018;201:124‐35. Lever J, Krzywinski M, Altman N. Points of significance: principal component analysis. Nat Methods 2017;14:641‐2. Higaki A, Mahmoud AUM, Paradis P, Schiffrin EL. Automated detection and diameter estimation for mouse mesenteric artery using semantic segmentation. J Vasc Res 2021;58:379‐87. Higaki A, Uetani T, Ikeda S, Yamaguchi O. Co-authorship network analysis in cardiovascular research utilizing machine learning (2009–2019). Int J Med Inform 2020;143:104274. Vollmer S, Mateen BA, Bohner G, Király FJ, Ghani R, Jonsson P, et al. Machine learning and artificial intelligence research for patient benefit: 20 critical questions on transparency, replicability, ethics, and effectiveness. BMJ 2020;368:15. Latif A, Rasheed A, Sajid U, Ahmed J, Ali N, Ratyal NI et al. Content-based image retrieval and feature extraction: a comprehensive review. Math Probl Eng 2019, (2019). Higaki A, Kurokawa T, Kazatani T, Kido S, Aono T, Matsuda K et al. Image similarity-based cardiac rhythm device identification from X-rays using feature point matching. Pacing Clin Electrophysiol 2021;44:633‐40. Siradjuddin IA, Wardana WA, Sophan MK, Feature extraction using self-supervised convolutional autoencoder for content based image retrieval. ICICOS 2019 - 3rd Int. Conf. Informatics Comput. Sci. Accel. Informatics Comput. Res. Smarter Soc. Era Ind. 4.0, Proc. (2019). https://doi.org/10.1109/ICICOS48119.2019.8982468. Choe J, et al. Content-based Image retrieval by using deep learning for interstitial lung disease diagnosis with chest CT. Radiology 2021. https://doi.org/10.1148/radiol.2021204164. Le QV Building high-level features using large scale unsupervised learning. in 2013 IEEE International Conference on Acoustics, Speech and Signal Processing 8595–8598 (IEEE, 2013). https://doi.org/10.1109/ICASSP.2013.6639343.