Daniel Amsel

Master's Student

Thesis Topic: Deep learning-based accelerated T1 mapping for cardiac MRI

Supervisors: Marc Vornehm, Jens Wetzl (Siemens Healthineers), Florian Knoll

Description

Magnetic Resonance Imaging (MRI) has the unique ability to distinct between T1 and T2 contrasts, which allows for a number of non-invasive quantitative measures [5]. In cardiac MRI, one example for such a measure could be the extracellular volume (ECV), used for example for the diagnosis of edema and interstitial fibrosis [5]. For ECV calculation, T1 maps of the heart are required before and after injection of a contrast agent [5]. Oftentimes the Modified Look-Locker inversion recovery sequence (MOLLI), followed by an exponential fitting [9] are used for T1 map computation. Although this procedure overcomes the challenges of cardiac and breathing motion during acquisitions, it is still very restrictive in terms of scan time and resolution, due to the requirement of multiple single-shot measurements during a single breath-hold. Multiple solutions to improve both the scan-time and the T1 map resolution have been proposed. The T1 mapping sequence itself can be modified to require less single-shot acquisitions with the downside of decreasing the T1 accuracy [10, 1]. Recently deep learning-based approaches were published to correct the noisy T1-maps reconstructed from acquisitions with accelerated sequences, counteracting this disadvantage [6, 4]. Another possible solution is to use higher acceleration factors during the single-shot acquisitions. This allows for either a reduction in scan-time or acquisitions with higher resolution. Both classical machine learning methods[3] as well as deep learning-based methods[8, 7] have been proposed regarding this modification.

In this work, a fully deep learning-based pipeline for accelerated T1 mapping in cardiac MRI will be developed and implemented. The pipeline consists of three separate neural networks, solving the sub-problems of reconstruction [11], registration [2] and exponential fitting. The networks are concatenated to allow for end-to-end training. Goal of the pipeline is to enable accelerated acquisitions of the inversion recovery single shots to allow for the reconstruction of T1 maps with higher resolution. The pipeline will be trained supervised, using data provided by Siemens Healthineers. For evaluation, the pipeline will be compared to a clinically approved, state-of-the-art T1 mapping method in terms of T1 accuracy. Additional evaluation experiments regarding the reconstruction of higher resolution T1 maps will be performed.

The thesis will include the following points:

Literature research of the topic of (accelerated) parametric T1 mapping for cardiac MRI
Developing an end-to-end deep learning-based pipeline for accelerated T1 mapping that performs reconstruction, registration and parameter extraction
Comparison of the method to a state-of-the-art method for T1 mapping for cardiac MRI

References

[1] K. Chow, J. A. Flewitt, J. D. Green, J. J. Pagano, M. G. Friedrich, and R. B. Thompson. Saturation recovery singleshot acquisition (SASHA) for myocardial T1 mapping. Magnetic Resonance in Medicine, 71(6):2082–2095, 2014. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/mrm.24878.

[2] A. V. Dalca, G. Balakrishnan, J. Guttag, and M. R. Sabuncu. Unsupervised learning for fast probabilistic diffeomorphic registration. In A. F. Frangi, J. A. Schnabel, C. Davatzikos, C. Alberola-Lòpez, and G. Fichtinger, editors, Medical Image Computing and Computer Assisted Intervention – MICCAI 2018, pages 729–738, Cham, 2018. Springer International Publishing.

[3] M. Doneva, P. Börnert, H. Eggers, C. Stehning, J. Sénégas, and A. Mertins. Compressed sensing reconstruction for magnetic resonance parameter mapping. Magnetic Resonance in Medicine, 64(4):1114–1120, 2010. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/mrm.22483.

[4] R. Guo, H. El-Rewaidy, S. Assana, X. Cai, A. Amyar, K. Chow, X. Bi, T. Yankama, J. Cirillo, P. Pierce, B. Goddu, L. Ngo, and R. Nezafat. Accelerated cardiac T1 mapping in four heartbeats with inline MyoMapNet: a deep learning-based T1 estimation approach. Journal of Cardiovascular Magnetic Resonance, 24(1):6, Jan. 2022.

[5] P. Haaf, P. Garg, D. R. Messroghli, D. A. Broadbent, J. P. Greenwood, and S. Plein. Cardiac T1 mapping and extracellular volume (ECV) in clinical practice: a comprehensive review. J Cardiovasc Magn Reson, 18(1):89, Nov. 2016.

[6] H. Jeelani, Y. Yang, R. Mathew, M. Salerno, and D. Weller. Fast and Robust T1-mapping using Convolutional Neural Networks. In Joint Annual Meeting ISMRM-ESMRMB ISMRT 31st Annual Meeting, 2019.

[7] H. Jeelani, Y. Yang, R. Zhou, C. M. Kramer, M. Salerno, and D. S. Weller. A Myocardial T1-Mapping Framework with Recurrent and U-Net Convolutional Neural Networks. In 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI), pages 1941–1944, Apr. 2020. ISSN: 1945-8452.

[8] Y. Jun, H. Shin, T. Eo, T. Kim, and D. Hwang. Deep model-based magnetic resonance parameter mapping network (DOPAMINE) for fast T1 mapping using variable flip angle method. Medical Image Analysis, 70:102017, May 2021.

[9] D. R. Messroghli, A. Radjenovic, S. Kozerke, D. M. Higgins, M. U. Sivananthan, and J. P. Ridgway. Modified Look-Locker inversion recovery (MOLLI) for high-resolution T1 mapping of the heart. Magnetic Resonance in Medicine, 52(1):141–146, 2004. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/mrm.20110.

[10] S. K. Piechnik, V. M. Ferreira, E. Dall’Armellina, L. E. Cochlin, A. Greiser, S. Neubauer, and M. D. Robson. Shortened Modified Look-Locker Inversion recovery (ShMOLLI) for clinical myocardial T1-mapping at 1.5 and 3 T within a 9 heartbeat breathhold. Journal of Cardiovascular Magnetic Resonance, 12(1):69, Nov. 2010.

[11] A. Sriram, J. Zbontar, T. Murrell, A. Defazio, C. L. Zitnick, N. Yakubova, F. Knoll, and P. Johnson. End-to-end variational networks for accelerated MRI reconstruction. In A. L. Martel, P. Abolmaesumi, D. Stoyanov, D. Mateus, M. A. Zuluaga, S. K. Zhou, D. Racoceanu, and L. Joskowicz, editors, Medical Image Computing and Computer Assisted Intervention – MICCAI 2020,
pages 64–73, Cham, 2020. Springer International Publishing.