The application of unsupervised artificial neural networks to the sub-classification of subjects at-risk of Huntington's Disease

The contingent negative variation (CNV), which is an evoked response in the human electroencephalogram (EEG), was measured for a number of Huntington's disease patients (HDs) and subjects at-risk of developing HD (ARs), and for equal numbers of matched normal subjects. The sampled voltage response values and the duration of the CNV were then used as input data to Kohonen and ART2 unsupervised artificial neural networks to classify the subjects. The two methods gave similar results for the HDs vs normals which also agreed with the results of a cluster analysis. The results of attempting to identify abnormal ARs showed that the ART2 results showed partial agreement with the results of the Kohonen network and cluster analysis. The application of these unsupervised neural networks to the sub-typing of clinical categories appears to offer a relatively simple tool suitable for hardware implementation.<
>