Effective enhancement of classi?cation of respiratory states using feed forward back propagation neural networks

摘要

In biomedical signal analysis, Arti?cial Neural Networks are frequently used for classi?cation, owing to their capability to resolve nonlinearly separable problems and the ?exibility to implement them on-chip processor, competently. Arti?cial Neural Network for a classi?cation task attempts to hand design a network topology and to ?nd a set of network parameters using a back propagation training algorithm. This work presents an intelligent diagnosis system using arti?cial neural network. Features were extracted from respiratory effort signal based on the threshold-based scheme and the respiratory states were classi?ed into normal, sleep apnea and motion artifacts. The introduced neural classi?er was then trained with different back propagation training algorithms and the classi?ed output was compared with the hand designed results. Five different back propagation training algorithms were used for training, such as Levenberg–Marquardt, scaled conjugate gradient, BFGS algorithm, one step secant and Powell–Beale restarts. Our results revealed that the system could correctly classify at an average of 98.7%, when the LM training method was used. Receiver Operating Characteristic (ROC) analysis and confusion matrix showed that the LM method conferred a more balanced and an apt classi?cation of sleep apnea and normal states.