Skip to content

Image Analysis

Building analysis into the image processing pipeline through machine learning to detect healthy anatomy and anomalies, monitoring disease progression and improving navigation

What does this
Methodology Involve?

Common operations applied to medical images involve various forms of image registration and segmentation. Often, these are considered as ad hoc steps within some form of processing pipeline, with little or no attempt to consolidate or generalise the principles. A more rigorous approach involves formulating all these operations within a unifying probabilistic generative model of the data. Of particular importance to medical imaging is the incorporation of spatial deformations within the model, along with methods for encoding inter-subject variability. There is still much work to be done in the area of biomedical image analysis, image datasets are large and models are complex, so numerous approximations are necessary – although a continuation of Moore’s Law would allow much more to be achieved. In general, those models that most accurately encode real biological phenomena (basic science applications) are likely to be the most useful for making predictions on which real-world medical decisions may be made (applied science). In addition to addressing the immediate short term needs of biologists and clinicians, imaging scientists should also be laying the groundwork for a future in which harnessing `big data’ in the NHS will revolutionise healthcare.

Partners and Facilities

This methodology connects with Clinical Themes across the CDT, with a focus on Neuroimaging and Cancer Imaging.

Studying Under this theme

Projects listed below are available for applicants to include in their 5 project choices. Successful candidates will meet with both primary supervisors from their selection as part of the induction process before being assigned to an allocated research project. To view all available project videos please visit our YouTube channel.

Theme Leader

Prof. John Ashburner
UCL Wellcome Trust Centre for Neuroimaging

I am one of the developers of the SPM software, which is used internationally by several thousand brain imaging researchers. Contributions to SPM relate to generative models for brain image registration and tissue segmentation (as well as informatics issues such as DICOM and other image file formats, ontology descriptions of processing procedures, user interfaces etc).

My current interests involve applying the computational anatomy framework of Grenander and others, in conjunction with pattern recognition approaches, to obtain accurate characterisations of anatomical differences among populations of subjects. This work has potential for translation into clinical practice, and relies on the establishment of large databases of scans. The emphasis is on models with predictive accuracy, rather than simple idealisations.