Neural networks in computer aided clinical electromyography

Abstract

In concentric needle electromyography, quantitative measurements are applied on the motor unit action potentials, which are recorded from the biceps muscle of normal subjects and patients suffering from neuromuscular disorders. An unsupervised learning neural network is employed for the classification of neuromuscular disorders. The results suggest that unsupervised learning has certain advantages in cases where the classes of the training data are unknown in number, or are not easily separated. Higher diagnostic yield is achieved through unsupervised learning, when compared to supervised learning. The significant finding, however, is not the higher diagnostic yield, but the flexibility offered by the unsupervised learning artificial neural networks for producing subcategories of various diseases that cannot be seen by simply studying the motor unit action potentials.