Unsupervised learning in computer aided macroelectromyography
Date
1991Author
Schizas, Christos N.Pattichis, Constantinos S.
Livesay, R. R.
Schofield, I. S.
Lazarou, K. X.
Middleton, Lefkos T.
ISBN
0-8186-2164-8Publisher
Publ by IEEESource
Proceedings of the 4th Annual Symposium on Computer-Based Medical SystemsPages
305-312Google Scholar check
Keyword(s):
Metadata
Show full item recordAbstract
Normals and patients from three disorders have been selected for investigation: (1) motor neurone disease (MND) (2) Becker muscular distrophy (AMD) and (3) spinal muscular atrophy. The data from 36 macroelectromyograms were used for analysis. The results suggest that unsupervised learning neural networks generally produce better results than those produced by the supervised learning neural networks. No conclusion about the optimum size of the output grid can be reached from the results since the examined models for the 10 × 10 and 8 × 8 cases produced similar results. It is expected, however, that an optimum grid size should exist. This size will depend on the size and the variability of the training set. More epochs can improve the performance of a model up to a certain level, beyond which the number of epochs will have no positive effect.