Towards 10-minute Magnetic Resonance Scanning in Children - Developing Accelerated Imaging Using Machine Learning

Magnetic Resonance Imaging (MRI) scans play a vital role in helping many ill children, by finding out what the problem is and helping plan their treatment. MRI is safe because it does not use radiation. MRI scans produce good-quality pictures or images of many parts of the body, including the brain, heart, spine, joints and other organs. The main problem is they take a long time - often over an hour. During the scan, the child has to keep very still and may even need to hold their breath many times. This is especially hard for children and unwell patients. Hence, younger children under 8 years old need a general anaesthetic, to put them to sleep during the scan. In many childhood diseases, for example in cancer, children may need many MRI scans to follow up disease progression and treatment. Being put to sleep for all of these scans is not pleasant for the child and may occasionally cause problems. It also puts a lot of pressure on hospitals who need to find the doctors, beds, equipment and funds for this.

One way of overcoming these problems would be to speed up the MRI scans so the children do not have to keep still or hold their breath. The simplest way of doing this is to collect less data for each image, but this causes so much distortion in the images that they cannot be used. There are some ways of converting these into useful images, but these are complicated and take too long to use in a hospital.

Machine Learning is an upcoming way of teaching computers to find complicated patterns in large amounts of information. Recent advances mean that computers are now so powerful that they can learn effectively. Machine Learning has been successfully used for analysing many types of images, for example to perform de-noising, interpolation, image classification and border identification. Despite its popularity, only a few recent studies have shown its potential for reconstruction of MRI images. This is partly due to the greater complexity of the problem and importantly, the large amounts of data required to 'learn' the solution. At Great Ormond Street Hospital, we have MRI images from over 100,000 children and scan an additional 10,000 children each year, all of which we could use to help train and test Machine Learning technologies.

I have already shown that basic Machine Learning techniques can remove distortions from MRI scans of the heart, so I am well placed to develop Machine Learning techniques to reconstruct MRI images from other children's diseases, as well as developing more advanced Machine Learning techniques. I showed Machine Learning to be faster than existing reconstruction methods and the images were of better quality than more conventional state-of-the-art techniques. However, much more work is needed to get Machine Learning working reliably in children's scans and to make the most of the possible benefits.

If we can use fast scanning with Machine Learning we could shorten scan times from 1 hour to about 10 minutes for children having MRI scans. They would not have to keep completely still for the scan and would not have to hold their breath, therefore reducing the need to put patients to sleep. This would make MRI scanning far less difficult and daunting for children, and would eliminate the cost and side effects from the anaesthetic. Quicker scans would help reduce waiting lists and costs for the NHS. It would also mean that MRI scanning would be used far more often, so it could help many more children. Additionally, these techniques could enable MRI scans to become affordable in some countries for the first time.

Magnetic Resonance Imaging (MRI) plays a vital role in diagnosis, management and monitoring of many childhood diseases. However, a complete examination is slow and requires the use of general anaesthesia (GA) in children under 8years. This fellowship will develop Machine Learning (ML) techniques to speed up the acquisition of MRI data, reducing total scan times from ~1hour to ~10minutes. Motion correction techniques will be incorporated to reduce the need for patient cooperation, making it possible to dispense with GA. In addition, fast and accurate image interpretation tools will be created, to help efficient and reproducible analysis and disease classification.

Impact on Patients;
Combined, this will make MRIs safer, more comfortable and less daunting, improving children's experience. The work will also increase accessibility and tolerance of MRI in paediatric diseases, allowing it to be used more routinely, both for initial diagnosis and follow-up of childhood illness. Together with improved image quality, reduced motion artefact, as well as higher resolution imaging and/or acquisition of additional data, these have the potential to improve diagnostic accuracy, monitoring and hence outcomes in childhood diseases.

Impact on Clinical Research;
In the short-term, these technologies will position the UK at the forefront of clinical research, and encourage pharmaceutical companies to use MRI in clinical trials. This is because they will reduce scan costs and simplify recruitment by enabling research scans to be performed on clinical patients (due to more efficient protocols). Additional populations may be recruited, where MRI may not have been ethically acceptable or practical before because of the need for GA or long scan times-improving understanding of other diseases. Additionally, more scans may be performed, improving the quality of data, allowing higher resolution imaging or acquisition of additional data.

Impact on NHS;
In the mid-term, benefits to the NHS include reduced risk and cost of GA, allowing redeployment of valuable anaesthetic staff and reducing pressure on equipment and hospital beds. Reduced requirements for patient compliance will reduce unsatisfactory scans and recall rates. Shortened scan times will reduce scan costs and allow an increase the number of patients that can be examined each day, ultimately reducing waiting lists. The technologies will help improve diagnosis and management of paediatric disease, preventing cost and side effects of ineffective treatment. By achieving high quality, fast scanning using existing scanner technology, these techniques may also reduce the need to upgrade scanners.

Other Clinical Applications;
The proposed techniques can also be applied to MRI of other systems, e.g. musculoskeletal or gastrointestinal imaging, as well as to adult MRI protocols, where similar benefits to patients, research studies and the NHS can be achieved. In the long term, it is possible these technologies may enable MRI to replace CT in some cases-eliminating the need for ionising radiation and CT contrast agent, reducing risk to patients, and freeing up CT resources.

Worldwide and Industrial Impact;
These technologies have the potential to transform paediatric MRI from an expensive, infrequently used technique, into a high throughput imaging technique that is routinely used clinically and in research studies in paediatric disease. Reduced scan times and costs would also have worldwide healthcare impact, including in countries where MRI is currently still too expensive for widespread use. This makes these technologies attractive to MRI scanner manufacturers, opening up new markets. Additionally, existing rolling software updates to existing MRI scanners enable worldwide deployment of these new techniques. Existing links with Siemens Healthcare (who are project partners), will allow short term translation to other sites and ultimately integration into the vendor's product.