Back to Projects

AI-enabled Imaging, Emerging Imaging

Greater than the sum of its parts: From siloed hardware to software enabled integration

Project ID: 2021_032

1st Supervisor: Jo Hajnal, King’s College London
2nd Supervisor: Shaihan Malik, King’s College London
Clinical Champions: David Edwards, King’s College London

Aim of the PhD Project:

  • To explore adaptive control as a means to make MRI systems more efficient at lower cost
  • To create a demonstrator system that provides proof of concept

Project Description / Background:

The classic approach for delivering high performance from complex machines is precision engineering focused on ensuring each subsystem if optimised for an idealised response and to minimise interaction between subsystems. However, a data and control revolution has started to overturn this paradigm. Mobile phone cameras provide an everyday but impressive example. With each generation of phone, these deliver ever higher performance using essentially the same optics crammed into incredibly small form factors. Avoiding image aberration and achieving sharp focus when the device cannot be held still  has moved from the realms of precision compound lenses (large and expensive) and mechano-optical stabilisation, to harnessing high data rates that provide multiple frames per exposure combined with machine learning based image restoration and adaptive correction. By contrast, medical imaging equipment such as MRI systems continue to rely on highly engineered and expensive subsystems designed to be as close to ideal as possible and managed by a control system that operates each subsystem independently of the other. This approach produces high performance, but at a massive cost.

Although MRI scanners are large machines, they have a lot in common with a mobile phone camera in that many different sub-systems are juxtaposed in a tight space. A fundamental challenge is that MRI scanners have 3 different magnetic field generating systems operating at different temporal rates. The main magnet generates a static field that must be extremely stable over time, gradient coils produce fields with precisely defined spatial patterns that can be changed over milliseconds, and radiofrequency (RF) resonators create magnetic fields that oscillate in the MHz range. These systems are nested one inside another. The fields each generate are harnessed for imaging within the patient, but each produces fields on the outside as well. A modern scanner exploits carefully balanced layers of counter-propagating electrical currents to simultaneously generate strong fields in the patient and to cancel fields in other regions where they would cause interactions between neighbouring sub-systems. The main magnet is actively shielded to reduce fields outside the scanner; gradients also use shielding currents designed to minimize production of time-varying magnetic fields incident on the  main magnet structure in order to minimize generation of parasitic eddy currents; the RF system has an electrical shield to isolate it from the rest.

These isolation strategies come at a massive cost in efficiency. A whole body RF coil has to be so close to the gradient coil that as much as 50% of the power applied to can be wasted in its shield. Manufacturers estimate that gradient systems efficiency can be as low as 20%  as a result of the need to isolate them from the main magnet system.

Removing the shielding current would dramatically increase efficiency of gradient coils, allowing systems to produce stronger fields and/or greatly reduce the (currently highly significant) power requirements. The cost of this is strong eddy currents that disrupt imaging and render standard acquisition protocols unusable. This project will explore how to achieve a new generation system control capability for unshielded field generators and seek to demonstrate its feasibility in proof of concept experiments focusing on the gradient system.

Same anatomy different eddy currents!

Figure 1: Same anatomy different eddy currents!


  1. Welz, A., Cocosco, C., Dewdney, A., Gallichan, D., Jia, F., Lehr, H., … Zaitsev, M. (2013). Development and Characterization of An Unshielded PatLoc Gradient Coil for Human Head Imaging. Concepts in Magnetic Resonance Part B: Magnetic Resonance Engineering, 43(4), 111–125.
  2. Conolly, S., Nishimura, D., Macovski, A., & Glover, G. (1988). Variable-rate selective excitation. Journal of Magnetic Resonance, 78(3), 440–458.

Back to Projects