In July 2017 we were delighted to welcome Dr Dan Marcus from Washington University to work with the Translational Imaging Group on a 5 month research placement.


Dan is Director of the Neuroinformatics Research Group (NRG), the Neuroimaging Informatics and Analysis Centre and the Director of XNAT from the NRG lab, all at Washington University.

Dan is one of our long-term collaborators and his groups XNAT system is being used for much of the groups research output. XNAT allows us to build and test image processing pipelines which enable the development of robust code for novel algorithms which help diagnose and treat conditions such as Alzheimer’s, epilepsy and more.

XNAT is an imaging informatics platform first released in 2004, and we wanted to hear more from the researcher who helped to develop this pioneering tool responsible for over 13 years of safe medical data sharing.






Could you tell us a little bit of background regarding XNAT and the history of its creation at Washington University in 2004?

A: XNAT is an imaging informatics platform, or put more simply an image database designed to cope with large medical imaging datasets. We developed the first instance of the platform in 2004, borne from the development of imaging studies at that time. Historically medical imaging research projects were very small scale, limited to around a dozen patients. Personal file storage was also an issue, with most of the results being kept on local computers, which made cross institution collaborations very difficult. It was around this time that large scale imaging studies were first established which created the need for more efficient way of working.


For our lab this started with a project involving the Alzheimer’s Imaging Centre for which we needed a storage method and sharing process which could cope with 100,000’s of medical data files being access from numerous hospitals and institutions. We quickly realised that getting an efficient pipeline in place was the most important challenge - getting the raw imaging data from the hospitals and converting it into a state in which computer scientists can analyse the images start performing analytics and start producing results.


Very early on we decided to make XNAT open source so that researchers globally could set up their own data sharing systems. We ensured that the platforms were compatible with all the major imaging sources such as DICOM and NIFTY and ensured that databases could support both imaging and non-imaging methods, such as behavioural statistics, to make the final output as flexible as possible for researchers wishing to build their own data structure.


Q: What are some of the most interesting projects you’ve been involved in with XNAT?

A: I would definitely say the Human Connectome Project was career changing to work on. This project saw us obtain funding from the US Department of Health to produce very high-resolution neuroimaging data sets for 1200 base-line patients – i.e. patients with ‘normal’ brain function. The purpose of the project was to help us understand brain connectivity and how the various areas of the brain respond to each other according to tasks and stimuli, allowing us to create a detailed map of this extremely complex organ. Data collection and sharing was an essential part of the project, particularly as we made the final results publicly available. The project has served as blueprint for how to map other complex connectivity networks, which can be useful for everything from traffic planning to social networks.


Our next challenges for this project will focus on creating the same detailed maps for different demographic groups and various disease states to allow us to improve our understanding of how brain function adapts and changes throughout our lives. We hope this will increase our understanding of neurological conditions to provide better treatment options.


The second is the creation of the Dementia Platform UK, an MRC-funded project to create the world’s largest study group for dementia research. It brings together 38 of the UK’s existing population studies and 10 academic partners to create a rich and robust data set to test and improve imaging algorithms. XNAT has been used as a base data federation for this project, and we’re now at the point where the network is fully operational at the participating institutes. This means users can create their own bespoke pipelines to analyse data from the network and make requests for specific data sets as required. Washington University have been involved in getting this created but management will be UK-led after this point.

Q: Would you like to tell us about some of your other projects outside of XNAT?

A: Some other areas of work include algorithm development for tumour segmentation, breast tumour detection and connectome analysis.

Q: How and why did your visit with the Translational Imaging Group come about?

A: I’ve been collaboration with UCL and specifically the Translational Imaging Group for many years, as early and prolific adaptors of the XNAT database doing some really exciting work on data sharing set ups. I’m going to be working with the lab until mid-December and during that time I’m hoping to offer my expertise on setting up new XNAT platforms, helping to troubleshoot any problems and hopefully foster a culture of sustained collaboration between UCL and Washington University. I’m also really interested in helping to set up the pipelines for the new PET MRI scanners here at UCL to see how we can do the same back home.


Q: What do you hope to achieve during your time here?

A: This really follows on from our earlier discussion, but in addition to the UCL / Washington University collaboration I’m really keen on seeing how we can use XNAT to support even larger data sets for the current drive towards big data research. We’re talking about working with really large data sets, 10’s of 100,000’s of cases, linked with images and clinical outcomes across multiple hospital sites and research groups – establishing this will be a real achievement and can hopefully be used as a blueprint for other ambitious big data sets in other countries.

Q: Are there any other significant or interesting projects you can see developing at UCL?

The work on resting state fMRI to map brain areas is really interesting, and something we’ve had over 1000 cases of at Washington University. I’m really keen to compare the tools being used to monitor fMRI between our two institutes. Historically it’s been difficult to identify localised areas of brain function; patients would need to spend hours in a traditional MRI scanner performing a variety of physical and cognitive tasks which can be especially difficult when working with very sick patients. fMRI allows us to map brain connectivity in a resting state and at a much quicker rate. By mapping this connectivity we are able to see which areas of the brain correlate to each other in these complex networks, particularly important when we are talking about resecting the brain for patients with brain cancer or epilepsy.


The GIFT-Cloud project is another interesting development as it’s core example of how the XNAT framework can be used for successful multi-institution sharing on an international scale.


Q: Where do you see the field of medical imaging progressing in the next 10 years?

A: I definitely see the development of large open access data sets becoming standard practice going forward. Longitudinal patient cohort studies will give us unprecedented insight into clinical developments and patterns if we’re able to upload and analyse imaging data throughout the patient pathway.

In addition, these kinds of datasets are essential for developing robust automatic imaging algorithms, artificial intelligence techniques and deep learning frameworks which will all have a big impact on diagnosis and treatment, hopefully leading to more advanced analytics being used clinically.


Q: How do you see the issues around patient data and data protection affecting medical imaging research in the long term? Do you see many differences between US and UK practice?

A: I think the main concern here comes from data mining operations in which data usage hasn’t been communicated to patients properly. In my personal experience patients are generally supportive of contributing to better tools and treatment of their condition when approached directly with clear guidance on how the data will be used. The new large-scale data capture methods we have spoken about are perhaps where concerns arise. Particularly in the US there is the specific concern that large-scale data projects could be used against patients by their medical insurers, for example if imaging data shows a predisposition to a certain disease or condition.


However, worth noting here is that XNAT provides automatic anonymization at the point of data entry. This means that any patient identifiers are stripped out and data records are instead given a unique code allowing the record to be tracked in the system. This allows our researchers to perform large scale algorithm testing which can offer aggregate results across the data set without there being any connection to specific patients.

It’s certainly an issue to be aware of and to ensure we’re communicating about clearly to patients and the public. For example, international standard automatic anonymization is something we’ve built into the new Dementia Platform UK. Their website openly discusses data usage and data protection and I think this kind of transparency is essential to support other large-scale data networks in the future.

Q: Looking at your online profiles it looks like you are Director of many groups and projects, what is your advice on managing multiple teams/work flows like this?

A: Having a great team and strong collaborators makes it possible! In addition, the vast majority of our projects use XNAT so improvements to processes and best practice outcomes can be applied across the board to save time.


Q: Finally, is there any one piece of advice you would offer up and coming medical imaging researchers today?

A: Know how to write code! A solid understanding of big data and its principles is going to be fundamental, and not just for the medical imaging field. Being an effective communicator is also really important – make sure you know how to write and talk about your research effectively.