Frans Coenen: Abstract

It can be argued that the state-of–the-art with respect to algorithms for knowledge Discovery, whether biologically inspired or otherwise, is well advanced. There is still more work to be done but we have some excellent algorithms, techniques and methods available to us. Imbedding such algorithms into real applications however remains problematic, we can do the knowledge discovery, but “contorting” data to fit the required input representations and the understanding of the output still presents us with difficulties. In short, end-to-end knowledge discovery remains a challenge. This is particularly the case given complex data, data that is not (at least in the first instance) presented in a “standard” tabular format. One form of complex data is image data, both 2D and 3D.

This presentation presents a review of experience gained regarding the practical embedding of knowledge discovery into medical image diagnosis systems. More specifically experience is presented with respect to knowledge discovery in 2-D Magnetic Resonance Imagery (MRI) brain scan data, 2-D fundus images of the human retina, 2D mammograms, 3-D Optical Coherence Tomography (OCT) of the eye and 3D MRI brain scan data. It will be demonstrated that data preparation is all-important and that this preparation is dictated by both what we want to find out and what knowledge discovery algorithms we intend to deploy. More particularly it will be suggested that graph representations, and consequently graph based knowledge discovery, are an ideal choice.

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