PET-MR Motion Correction Based Purely on Routine Clinical Scans

Lead Research Organisation: King's College London
Department Name: Imaging & Biomedical Engineering


Globally, cancer is one of the most dangerous and prevalent diseases affecting mankind. Each year, around 14 million people are diagnosed with cancer and 8 million people die from the disease. One of the primary means of diagnosing cancer and monitoring/planning its treatment is through the use of PET imaging. However, the problem of patient motion, such as that caused by breathing, seriously hinders the effectiveness of current PET scanners.

One way of estimating breathing motion, so that its effects can be corrected for, is through the use of a second imaging modality. Traditionally, PET imaging has been combined with CT scanning because of its usefulness in reconstructing PET images. However, CT scanning delivers a significant radiation dose to the patient and is too slow to be able to estimate breathing motion effectively. Recently, a new generation of PET scanners have been introduced to the market, which combine PET with MR imaging. MR offers a highly promising source of information for estimating breathing motion. However, techniques to fully exploit the rich information provided by MR scanning are currently lacking.

The aim of this proposal is to devise techniques for estimating breathing motion from MR and PET imaging data, for the purpose of motion correcting both the PET and MR data. If successful, the outcome of this research will be to improve the quality of PET-MR imaging. This will enable cancer to be diagnosed earlier, allowing potentially more effective treatment. In addition, it will be possible to monitor the progress of cancer treatment more reliably, and even make the treatment more targeted and effective.

A key benefit of our proposed approach is that it will not require clinical workflows to be altered. Normally, MR and PET scans are performed for clinical purposes: the PET images are used to diagnose/assess tumours, whereas MR images provide the anatomical context. The two modalities are therefore complementary, and both are clinically useful. All previous approaches to motion correcting PET using MR data have, as well as being limited in their effectiveness, required extensive use of MR scanner resources. Effectively, the MR scanner would be used as an expensive motion correction device. As a result none of these techniques are currently in widespread clinical use. Our proposed approach makes use only of images that are routinely acquired for clinical purposes, and would therefore be much more likely to be used clinically.

Planned Impact

One of the primary groups of beneficiaries of the methods developed in this project are patients who have, or are suspected of having, cancer, as well as their relatives and carers. Secondary beneficiaries will be clinicians involved in diagnosing and treating cancer. In addition, the research will also have an impact on businesses that are developing innovative solutions to the challenge of improving cancer diagnosis and treatment. Other beneficiaries of this work include researchers in medical imaging and computer science.

Patients who are suspected of having developed cancer will benefit from improved diagnosis as a result of the higher spatial resolution of PET imaging made possible by this research. Patients who have already been diagnosed with cancer will benefit from more reliable disease staging and improved information for guiding treatment. For example, as well as improved PET-MR imaging, one secondary output from the techniques we develop will be information about the expected respiratory motion of the patient. Therefore, as well as providing more detailed images for planning tumour margins, the motion information could be used to incorporate respiratory motion into these tumour margins, further improving the treatment, e.g. motion compensated radiotherapy.

In addition, clinicians and the NHS will benefit from the availability of improved cancer imaging. This will enable more efficient and targeted use of clinical resources. For example, improved diagnosis will result in earlier and potentially more effective treatment, which will obviate the need for expensive treatment when the disease is more advanced. Improved staging will lead to more reliable information about the success or otherwise of treatment, again potentially reducing unnecessary use of expensive clinical resources. Improved treatment planning will lead to a higher treatment success rate, reducing the need for future treatment.

Furthermore, we strongly believe that the application in medical imaging of dimensionality reduction techniques such as manifold alignment is a highly promising area for future research. Therefore, the techniques we develop in this project could have an impact on other clinical application areas, such as image-guided interventions. This would open up a new area for potential impact on the public, the NHS and researchers in medical fields.

Finally, the proposed research will strengthen the international position of UK healthcare and biomedical industries in the area of cancer imaging, and potentially in other areas too.
Description When patients breathe during scanning, this can cause errors in the resulting images. This is true for both magnetic resonance (MR) imaging, which is used for visualising the anatomy, and positron emission tomography (PET), which is commonly used for cancer diagnosis. Much research effort has been focused on devising methods to correct for these errors, but they all involve acquiring extra data for motion correction purposes, or altering the clinical scanning protocol in some way. As a result they have yet to be translated into regular clinical use. We developed methods that can be used to perform motion correction of simultaneously acquired MR and PET scanning data from routinely acquired data only. Therefore, no changes need to be made to the standard clinical protocols. This is a significant step forward to making motion correction available in clinics worldwide.
Exploitation Route Healthcare technology companies who manufacture MR scanners may be interested in our technique for self-gating. Similarly, companies who manufacture PET scanners may be interested in our parameter-reduced reconstruction approach. Both methods are of interest to the academic community for developing the the state of the art in the fields of MR and PET reconstruction.
Sectors Healthcare,Pharmaceuticals and Medical Biotechnology

Description Research was carried out to demonstrate that motion correction of magnetic resonance (MR) scans can be performed without any additional motion-related acquisitions, i.e. just by analysing the structure of the motion-corrupted data acquired. We made exciting progress in this area and have shown how this principle can be applied to the commonly used radial acquisition trajectory, resulting in significant improvements over state-of-the-art techniques. We then built upon this early work and demonstrated that simultaneously acquired PET and MR data can be analysed in the same way, enabling the potential for robust and accurate motion correction of both modalities without altering clinical scanning protocols in any way. This research culminated in a publication in one of the highest impact machine learning journals (IEEE Transactions on Pattern Analysis and Machine Intelligence).
First Year Of Impact 2019
Sector Healthcare,Pharmaceuticals and Medical Biotechnology
Title Code and data from Chen et al TMI 2016 
Description Code and data for generation of randomised realistic synthetic dynamic MR data 
Type Of Material Database/Collection of data 
Year Produced 2016 
Provided To Others? Yes  
Impact Unknown 
Title Data from Clough et al PAMI 2019 
Description MR slice data as used in Clough et al IEEE Transaction on PAMI 2019. 
Type Of Material Database/Collection of data 
Year Produced 2019 
Provided To Others? Yes  
Impact None yet 
Title PET-MR motion data 
Description This database contains the data used in the research article 'Respiratory Motion Correction of PET using MR-Constrained PET-PET Registration', by Balfour et al [BioMedical Engineering OnLine 2015, 14:85]. This data includes motion-affected PET images simulated from real dynamic MR image volumes. This dataset includes data from MR scans of 4 healthy volunteers. The database, including a more detailed description of the data it contains, can be found at: 
Type Of Material Database/Collection of data 
Year Produced 2015 
Provided To Others? Yes  
Impact The database was used in the following publication: 'Respiratory Motion Correction of PET using MR-Constrained PET-PET Registration', by Balfour et al [BioMedical Engineering OnLine 2015, 14:85].