Acid-base balance is a dynamic process and the transition from one classification to another does not occur abruptly and, although the general principles for classifying such data are standard, the details of each algorithm can be subtly different. This research aimed to explore the use of self-organising maps as applied to the medical diagnostic problem of acid-base disorders. The relevance of using self-organising networks for the classification of this data particularly focused on the mapping from the multi-dimensional data to the network nodes. This maintained the spatial organisation of the data allowing a natural graphical representation which directly related to the more familiar diagrams used as aids in this subject. Furthermore, the network was not bound by rigid classificatory criteria and thus could give some meaningful category even to data samples excluded by the standard algorithms. It was, thus, an improvement on such rule-based diagnostic criteria. Other methods were investigated including supervised networks and clustering algorithms but were not as successful when presented with mixed samples and relied more strongly on the initial data classifications further strengthening the argument for the self-organising methods. The self-organising map has proved to be of potential use in the classification of acid-base disorders. It has several advantages over standard simple classificatory algorithms and such features can be used to provide a meaningful graphical representation of the data. Various methods can be used to overcome problems involved with the labelling of these maps and comparisons have been made between them. The use of these maps also shows potential in dealing with temporal data series.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:662110 |
| Date | January 2001 |
| Creators | Smith, Sharon |
| Publisher | University of Edinburgh |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/1842/12142 |
Page generated in 0.002 seconds