Back to Projects

AI-enabled Imaging, Emerging Imaging

AI enabled ultrahigh field body MRI for bone marrow cancer imaging

Project ID: 2020_021

1st Supervisor: Shaihan Malik, King’s College London
2nd Supervisor: Jorge Cardoso, King’s College London
Clinical Champion: Vicky Goh, King’s College London

Aim of the PhD Project:

  • Perform pioneering high resolution bone marrow cancer imaging in the pelvis at 7T
  • Create a high-resolution body imaging capability for 7T MRI
  • Develop AI-based methods to remove the need for patient-specific calibrations that currently severely limit the clinical utility of these machines outside of the head or small joints

Project Description / Background:

MRI plays an important role in the management of cancer patients. However, its sensitivity for detecting cancer is limited in some body areas e.g. lung, skeleton, by their intrinsically low MRI signal.  New ultrahigh field strength (7T) scanners can potentially achieve much higher sensitivity and spatial resolution than existing technology. We hypothesise that for bone-marrow imaging this improved sensitivity will detect a low burden of bone marrow cancer or metastases, changing patient management at an earlier stage. At present 7T-MRI is routinely successful for brain but less successful for body imaging. The much higher resonant frequency at 7T compared to 1.5T clinical scanners (300MHz vs 64MHz) leads to uneven propagation of radio-frequency magnetic fields into the body, resulting in unusable images. Parallel transmission systems (PTx), adapting to each patient, can improve this in principle1.

A key feature of PTx methods is that they are specific to each type of pulse sequence, since they concern how the RF system is used during the MR acquisition process. Within our own group for example, we have developed new methods for uniform T1-weighted imaging, 2 T2-weighted TSE,3,4 FLAIR5 and rapid gradient echo sequences6. Although these methods can produce high quality imaging, they require calibration measurements (i.e. B­1 and B0 field maps) and in-situ calculations that may take a few minutes to complete and require the presence of a specialist technical operator. While this works well in controlled conditions, it is not compatible with a clinical workflow where it would be desirable to remove the need for a specialist operator and to drastically reduce or even eliminate the time spent on calibration/calculation.

The ‘individual’ approach does overlook a key redundancy: though different in detail, human subjects are similar in many ways. Hence it is likely that the space of optimal solutions forms a low dimensional sub-space of the possible solutions. The existence of such a low-dimensional solution space has been demonstrated recently by the proposal of ‘universal pulses’7 for B1+ correction for 7T brain imaging; these pulses are optimized by including calibration data from many subjects at once and perform reasonably well across a population of different test subjects. This was also demonstrated in our recent work on FLAIR5. ‘Universal’ methods exist at the opposite end of the spectrum to the adaptive methods and are overly reductive for many applications in which inter-subject variation is larger. However their existence points towards a new paradigm for MRI scanner technology in which models with enough flexibility to generate optimal settings for really diverse subject groups could be ‘learned’, using current calculation methods to generate the training data.  Artificial intelligence (AI) based MRI scan optimization methods are now beginning to emerge in related areas, but are so far limited to simpler cases such as 3T MRI8, RF shimming9 and power estimation for inversion pulses.10 We will create a comprehensive ‘learning examination’ approach to 7T body MRI in which each new image obtained contains information that is used to inform the subsequent ones, producing high quality imaging with high time efficiency.

References:

  1. Padormo, F., Beqiri, A., Hajnal, J. V. & Malik, S. J. Parallel transmission for ultrahigh-field imaging. NMR Biomed. 29, 1145–1161 (2015).
  2. Malik, S. J., Keihaninejad, S., Hammers, A. & Hajnal, J. V. Tailored excitation in 3D with spiral nonselective (SPINS) RF pulses. Magn. Reson. Med. 67, 1303–1315 (2012).
  3. Malik, S. J., Beqiri, A., Padormo, F. & Hajnal, J. V. Direct signal control of the steady-state response of 3D-FSE sequences. Magn. Reson. Med. 963, 951–963 (2014).
  4. Sbrizzi, A. et al. Optimal control design of turbo spin-echo sequences with applications to parallel-transmit systems. Magn. Reson. Med. 373, 361–373 (2017).
  5. Beqiri, A., Hoogduin, H., Sbrizzi, A., Hajnal, J. V. & Malik, S. J. Whole-brain 3D FLAIR at 7T using direct signal control. Magn. Reson. Med. 80, 1533–1545 (2018).
  6. Beqiri, A., Price, A. N., Padormo, F., Hajnal, J. V. & Malik, S. J. Extended RF shimming: Sequence-level parallel transmission optimization applied to steady-state free precession MRI of the heart. NMR Biomed. 30, e3701 (2017).
  7. Gras, V., Vignaud, A., Amadon, A., Le Bihan, D. & Boulant, N. Universal pulses: A new concept for calibration-free parallel transmission. Magn. Reson. Med. 77, 635–643 (2017).
  8. Tomi-Tricot, R. et al. SmartPulse, a machine learning approach for calibration-free dynamic RF shimming: Preliminary study in a clinical environment. Magn. Reson. Med. 1–16 (2019). doi:10.1002/mrm.27870
  9. Ianni, J. D., Cao, Z. & Grissom, W. A. Machine learning RF shimming: Prediction by iteratively projected ridge regression. Magn. Reson. Med. 1–11 (2018). doi:10.1002/mrm.27192
  10. Bollman, S. et al. Improving FLAIR SAR efficiency by predicting B1-maps at 7T from a standard localizer scan using deep convolutional neural networks. in Proc. Intl. Soc. Mag. Reson. Med. 1129 (2018).
  11. Malik, S. J., Beqiri, A., Padormo, F. & Hajnal, J. V. Direct signal control of the steady-state response of 3D-FSE sequences. Magn. Reson. Med. 73, 951–963 (2015).
  12. Homann, H. et al. Toward Individualized SAR Models and In Vivo Validation. 1776, 1767–1776 (2011).
  13. Gal, Y., Hron, J. & Kendall, A. Concrete Dropout. arXiv Prepr. (2017).

Back to Projects