Abstract
Accurate localization of the aortic annulus is key to several imaging tasks, like cross-sectional aortic valve plane estimation, aortic root segmentation, and annulus diameter measurements. In this project, we propose an end-to-end trainable deep reinforcement learning (DRL) algorithm aimed at identification of the aortic annulus in patients with aortic dissection. We trained 5 different agents on a dataset of 75 CT scans from 66 patients following a sequential model-upgrading strategy. We evaluated the effect of performing different image preprocessing steps, adding batch normalization and regularization layers, and changing terminal state definition. At each step of this sequential process, the model performance has been evaluated on a validation sample composed of 24 CTA scans from 24 independent patients. Localization accuracy was defined as the Euclidean distance between estimated and target aortic annulus locations. Best model results show a median localization error equal to 2.98 mm with an interquartile range equal to [2.25, 3.81] mm, and a failure rate (i.e., percentage of samples with localization error \(> 10\) mm) of \(0\%\) in validation data. We proved the feasibility of DRL application for aortic annulus localization in CTA images of patients with aortic dissection, which are characterized by a large variability in aortic morphology and image quality. Nevertheless, further improvements are needed to reach expert-human level performance.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
Publicly available at https://github.com/amiralansary/rl-medical.
References
Al, W.A., Yun, I.D.: Partial policy-based reinforcement learning for anatomical landmark localization in 3D medical images. IEEE Trans. Med. Imaging 39(4), 1245–1255 (2019)
Alansary, A., et al.: Evaluating reinforcement learning agents for anatomical landmark detection. Med. Image Anal. 53, 156–164 (2019)
Astudillo, P., et al.: Automatic detection of the aortic annular plane and coronary ostia from multidetector computed tomography. J. Intervent. Cardiol. 2020, 1–9 (2020)
Elattar, M., et al.: Automatic aortic root landmark detection in CTA images for preprocedural planning of transcatheter aortic valve implantation. Int. J. Cardiovasc. Imaging 32(3), 501–511 (2016). https://doi.org/10.1007/s10554-015-0793-9
Gao, X., et al.: Quantification of aortic annulus in computed tomography angiography: validation of a fully automatic methodology. Eur. J. Radiol. 93, 1–8 (2017)
Ghesu, F.C., et al.: Multi-scale deep reinforcement learning for real-time 3D-landmark detection in CT scans. IEEE Trans. Pattern Anal. Mach. Intell. 41(1), 176–189 (2017)
Ioffe, S., Szegedy, C.: Batch normalization: accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167 (2015)
Jones, M., Dobson, A., O’Brian, S.: A graphical method for assessing agreement with the mean between multiple observers using continuous measures. Int. J. Epidemiol. 40(5), 1308–1313 (2011)
Mnih, V., et al.: Human-level control through deep reinforcement learning. Nature 518(7540), 529–533 (2015)
Moccia, S., De Momi, E., El Hadji, S., Mattos, L.S.: Blood vessel segmentation algorithms-review of methods, datasets and evaluation metrics. Comput. Methods Programs Biomed. 158, 71–91 (2018)
Pepe, A., Li, J., Rolf-Pissarczyk, M., et al.: Detection, segmentation, simulation and visualization of aortic dissections: a review. Med. Image Anal. 65, 101773 (2020)
Queirós, S., et al.: Automatic 3D aortic annulus sizing by computed tomography in the planning of transcatheter aortic valve implantation. J. Cardiovasc. Comput. Tomogr. 11(1), 25–32 (2017)
Rubin, G.D., Shiau, M.C., Leung, A.N., Kee, S.T., Logan, L.J., Sofilos, M.C.: Aorta and iliac arteries: single versus multiple detector-row helical CT angiography. Radiology 215(3), 670–676 (2000)
Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction. MIT Press, Cambridge (2018)
Theriault-Lauzier, P., et al.: Recursive multiresolution convolutional neural networks for 3D aortic valve annulus planimetry. Int. J. Comput. Assist. Radiol. Surg. 15, 577–588 (2020). https://doi.org/10.1007/s11548-020-02131-0
Tops, L.F., et al.: Noninvasive evaluation of the aortic root with multislice computed tomography: implications for transcatheter aortic valve replacement. JACC: Cardiovasc. Imaging 1(3), 321–330 (2008)
Yu, C., Liu, J., Nemati, S.: Reinforcement learning in healthcare: a survey. arXiv preprint arXiv:1908.08796 (2019)
Zhang, P., Wang, F., Zheng, Y.: Deep reinforcement learning for vessel centerline tracing in multi-modality 3D volumes. In: Frangi, A., Schnabel, J., Davatzikos, C., Alberola-López, C., Fichtinger, G. (eds.) MICCAI 2018. LNCS, vol. 11073, pp. 755–763 (2018). Springer, Cham. https://doi.org/10.1007/978-3-030-00937-3_86
Acknowledgements
A. Pepe was supported by the TU Graz LEAD project Mechanics, Modeling and Simulation of Aortic Dissection. M.J. Willemink was supported by a Postdoctoral Fellowship Grant from the American Heart Association (18POST34030192). D. Mastrodicasa was supported in part by a grant from National Institute of Biomedical Imaging and Bioengineering (5T32EB009035).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Codari, M. et al. (2020). Deep Reinforcement Learning for Localization of the Aortic Annulus in Patients with Aortic Dissection. In: Petersen, J., et al. Thoracic Image Analysis. TIA 2020. Lecture Notes in Computer Science(), vol 12502. Springer, Cham. https://doi.org/10.1007/978-3-030-62469-9_9
Download citation
DOI: https://doi.org/10.1007/978-3-030-62469-9_9
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-62468-2
Online ISBN: 978-3-030-62469-9
eBook Packages: Computer ScienceComputer Science (R0)
