NNERVE: neural network extraction of repetitive vectors for electrodiagnosis

Neural networks have been proposed for numerous different applications in computer based medical systems. The authors present a system for decomposition and interpretation of the electromyogram. They show that signal analysis requires a multi-stage hybrid approach, using more than one neural network, and separating stages that deal with the signal itself, signal features, and the final diagnostic interpretation. The use of neural networks for decomposition provides a high degree of robustness in the face of the noise and variability present in real clinical signals, and requires a novel algorithm called pseudo-unsupervised training. The use of neural networks for signal interpretation allows this stage to be developed even in the absence of clear diagnostic interpretive criteria, and allows the discovery of new signal features and classification schemata.<
>