Dr. Frans Coenen: Abstract

Medical diagnosis is often supported by the analysis of images obtained using (say) magnetic resonance or optical coherence tomography. Often clinicians wish to identify the presence or absence of a feature (tumour v. non-tumour). In data mining terms this is an image classification problem. The principal research issue to be addressed is how to represent images so that salient features are maintained while at the same time ensuring tractability.

This presentation considers a number of mechanisms whereby the desired medical image classification can be achieved. The mechanisms are founded on ideas concerning time series analysis, histograms, frequent sub-tree mining and shape description.

To act as a focus the presentation is directed at two applications. The first is concerned with the analysis of MRI brain scans in terms of a specific feature (the corpus callosum) that occurs within such scans whose size and shape is conjectured to indicate conditions such as epilepsy and autism, and abilities such as musical or mathematical ability. The second is directed at the analysis of retina imagery to identify the presence of a condition known as Age related Macular Degeneration (AMD), the leading clause of adult blindness in the UK.

Organised by BCS SGAI The Specialist Group on Artificial Intelligence http://www.bcs-sgai.org