Exemplar 3 (psychosis)
Using existing MRI datasets to standardise analyses and develop capacity for remote analyses of existing datasets: psychosis exemplar
Aims: To integrate large existing datasets of serial MRI scans and behavioural data coupled to the NeuroGrid image analysis services into a Grid-based database, test image normalisation techniques, and develop a generic ontology for a psychosis database, for use in multi-centre studies.
Background and rationale: There is now strong evidence that structural and functional brain changes occur in psychosis (schizophrenia and bipolar disorder). Thus, psychosis is a brain disease. Some of these features may predate onset of illness and may progress as a function of illness severity: we anticipate that the accuracy of their measurement will continue to improve rapidly in the coming years. If so, brain changes may prove to be key outcome measures to validate treatment studies, which rest currently entirely on symptom ratings. There is therefore an urgent need to increase the reliability, sample size, and hence power, of imaging studies to develop the necessary measures of brain change that can be used in cohort studies and clinical trials.
We will use existing data from the MRC-funded Edinburgh High Risk Project (EHRS; PI Johnstone, image analysis supervised by Lawrie; MRC programme grant G9825423) to develop a remotely accessible database. The toolkit developed in the core Grid services part of the project will then be applied to this database to solve a set of problems that are key to undertaking successfully large scale clinical neuroimaging studies. The EHRS is one of only two studies worldwide which has longitudinal data gathered from individuals before and after their first onset of psychosis. It has identified a number of clinical, cognitive and imaging trait markers of schizophrenia in a cohort of approximately 200 subjects who have been examined repeatedly over up to 10 years. Two scanners have been used for MRI (from 1994-9 and 1999-2004) with sequences that optimised grey/white contrast for particular machines. Differences in coil homogeneity and tissue signal intensities, and co-registration problems have been identified. Data collection has been completed in a complementary study of patients with familial schizophrenia and their unaffected first degree relatives, patients with bipolar disorder (BPD) and their unaffected relatives, and healthy controls (McIntosh; MRC Clinical Training Fellowship G84/5699). Both studies have systematically acquired overlapping demographic, diagnostic, symptomatic, MRI, neuropsychological, obstetric and familial mental health data (including that from national archives). Merging the two datasets would develop a comprehensive psychosis ontology, in a database of approximately 350 individuals and 700 MRI scans, as a template for future collaborative studies. Protocols for scanner harmonisation based on two scanner specific templates, as well as subject specific templates, are being developed. This will benefit from refinement with the NeuroGrid image analysis toolkit and validation studies on different distributed datasets. The EHRS will benefit from the construction of a Grid-based database and associated increased computational power e.g. to run complex analyses of connectivity on fMRI data using the toolkit. It may also prove possible for us and other researchers to ‘rescue’ data from previous cohort studies and trials, which have been disrupted by unavoidable changes in MRI hardware. Applying the combined methodological expertise across NeuroGrid sites to the problem of harmonisation will allow us to plan image acquisition and analysis for the multi-centre prospective MRI study of MCI, and contribute to the development of appropriate technology for additional studies across a range of neuropsychiatric disorders.
Methods: We will develop a Grid-federated database architecture to provide distributed data storage, beneath an overarching Grid-layer which allows the database to be transparent and accessible. The interface for the psychosis database will be developed in collaboration through the Edinburgh and Oxford e-Science Centres. Local and remote (Oxford, London) analyses will be conducted to investigate optimal approaches to image standardisation. We will use a range of existing MRI image processing software (SPM, Freeborough et al, FSL (SIENA)) incorporated into the NeuroGrid image analysis toolkit, to identify the key factors that impact on scanner incompatibility and to refine processing strategies that minimize artefactual temporal effects in the 20 EHRS healthy controls. Using the methods for the assessment of volume change over time incorporated into the NeuroGrid toolkit including tools developed at FMRIB and Queen Square), we will establish reliable measures of disease progression in subjects at high risk of schizophrenia and BPD.
Contribution to the development of Neuro-Grid: the exemplar tests capabilities of NeuroGrid to deal with retrospective data, assimilate material into databases and use of the toolkit for normalisation and analysis.
