Computational Analysis of Prostate TUmouRs
Research type
Research Study
Full title
Computational Analysis of Prostate TUmouRs (CAPTUR study)
IRAS ID
288185
Contact name
Tristan Barrett
Contact email
Sponsor organisation
Cambridge University Hospitals NHS Foundation Trust and the University of Cambridge
Duration of Study in the UK
5 years, 1 months, 30 days
Research summary
In recent years there has been increasing interest in the application of computational methods in medicine, which can be attributed to the exponential increase in computer power and refinement of computational algorithms. This emerging field is particularly applicable to evaluation of radiological studies as the digital images obtained in the course of normal clinical care are inherently amenable to computational analysis.
One relatively simple application of computational methods is the extraction of image information about patterns of pixels in the images that is not otherwise perceivable by the human eye, termed texture analysis. This allows characterisation of tumour heterogeneity, characterisation of parts of the tumour with different biological characteristics (tumour habitats) and in some cases prediction of tumour genomics (radiogenomics). Another rapidly developing branch of computational analysis is machine learning, a subset of artificial intelligence based on algorithms that are able to learn from provided data without pre-defined rules of reasoning, in particular deep learning based on multi-layer convolutional neural networks. Machine learning promises automation of many mundane tasks in radiology workflow which currently are not routinely performed due to the required time commitment, such as tumour segmentation, image co-registration, or volume measurements. A further area of interest is in determining image quality, which is crucial to accurate image interpretation.
In this study we propose to perform retrospective computational analysis of existing MRI imaging data obtained in the course of routine clinical care of patients with prostate tumours treated in our institution, in correlation to information on tumour histology, grading, genomics, and clinical outcomes. Computational analysis will result in image masks of the relevant regions for the evaluation of texture analysis and convolutional neural network training. The aim of the study is to develop algorithms and predictive models to improve radiological workflow, allow non-invasive tumour characterisation and predict patient outcomes.
REC name
N/A
REC reference
N/A