Exemplar 2 (stroke)
Development of an infrastructure for image management and observer interpretation in large multicentre trials – ACCESS
Aim: To explore use of an infrastructure for improved use of imaging in studies where traditional methods are no longer adequate, including: a) efficient interpretation and storage of large image datasets from multicentre randomised controlled trials, b) very large studies of observer reliability of image interpretation, c) establishing large “living archives” of images linked to key metadata for diseases which require long term study to understand their true natural history and the effects of treatment, and d) for knowledge transfer.
Research to be undertaken:Very large clinical trials are now needed to understand the disease, and the effects of treatment in individual patients, to improve outcome after stroke.
These trials depend on imaging as part of the assessment of the patient and also to find features that indicate if the patient is likely to respond to that particular treatment or not.
The facilities for handling images for large clinical trials are not very good at present.
Some signs of early stroke are very subtle and it might be possible to improve detection of abnormalities with image recognition algorithms.
The aims of the Stroke Exemplar are to improve infrastructure for handling imaging in large studies including:
- efficient interpretation and storage of large image datasets from multicentre randomised controlled trials,
- very large studies of observer reliability to improve image interpretation,
- establishing large "living archives" of images linked to key metadata for diseases which require long term study to understand their true natural history and the effects of treatment, and
- for knowledge transfer.
The development of a structure for trial image metadata, based on a careful description of the metadata in the two exemplar trials, is a key part of the project in conjunction with Dr Proctor. We also plan to test whether it is possible to train an image recognition algorithm using high contrast stroke (diffusion-weighted magnetic resonance) images for use on low contrast CT scan images to improve detection of subtle abnormalities.
We will use two ongoing large multicentre trials to develop and test our methods:
- IST3 - The Third International Stroke Trial (PI Prof Peter Sandercock, Edinburgh) - testing thrombolysis with alteplase (a clot-busting drug) in 6000 patients with ischaemic stroke in XXX centres in YYY countries worldwide - www.IST3.com
- ENOS - 'Efficacy of Nitric Oxide in Stroke' (PI Prof Philip Bath, Nottingham) - testing glyceryl trinitrate skin patches (a nitric oxide donor) in 5,000 patients with ischaemic and haemorrhagic stroke in 200 centres in 35 countries worldwide - www.enos.ac.uk
What are the problems to be overcome?
All the scans have to be adjudicated by expert readers – a quick and efficient method of scan presentation to the expert reader is required
Scans coming from many different scanners and via different transfer methods may have different file formats
Anonymisation of scan data is crucial
Long term data storage is important - these trials are expensive and hard to do and images provide a very important resource which may need to be re-evaluated in light of new knowledge many years from now - the data need to be accessible and current.
Solving these problems will make future trials more efficient also.
Contribution to the development of NeuroGrid: Establishing and testing mechanisms for interpretation and curation of image data which are essential to the infrastructure of many multicentre trials in common brain disorders. This includes testing issues of security, confidentiality, accessibility, linking to metadata, 'hooks' to computational image analysis methods, resolving issues of scale (receiving images from very many sites, not just a few), and ensuring commonality of key data requirements for future meta-analyses.
