UCL/Sheffield Brain Tumour Initiative
Research type
Research Study
Full title
UCL/Sheffield Brain Tumour Initiative
IRAS ID
221600
Contact name
MJ Cardoso
Contact email
Sponsor organisation
University College London
Duration of Study in the UK
4 years, 11 months, 28 days
Research summary
Manual segmentation of tumours from MRI images is currently a necessary step in the pipeline for quantitative neuro-oncologic assessment. Recent computational advances have raised the possibility of designing fully-automated algorithms that are sufficiently accurate to replace some of the work of manual image annotation. We are currently working on developing such a tool, while ensuring it fully incorporates complementary information from a variety of image modalities, works on multiple tumour types/grades, and can be used on images from multiple sites. The tool must also be robust to missing data (for example, scans that were not acquired due to artefacts or patient movement).
The machine learning techniques we use require ample training data (i.e. labelled medical images) with sufficient anatomical variability to learn the true anatomical appearance of healthy and pathological tissues. These algorithms require training on scans from multiple centres, as, in order to be of clinical use, it will need to work on a variety of scanners and imaging parameters. We aim to design a tumour segmentation tool that can automatically segment brain tumours from healthy tissue by learning from previous examples.REC name
North West - Liverpool Central Research Ethics Committee
REC reference
18/NW/0151
Date of REC Opinion
8 Mar 2018
REC opinion
Favourable Opinion