Multiple MLP Neural Networks Applied on the Determination of Segment Limits in ECG Signals

The electrocardiogram (ECG) has a characteristic morphology composed by various waves, corresponding to the activities in different regions of the human heart. These waves have expected ranges of duration and amplitude, and large deviations from such values indicate a series of heart diseases. This work proposes an algorithm, based on multiple multi-layer perceptron (MLP) neural networks, to automatically determine the onset and offset of each component wave, as a first step for implementing a fully automated diagnosis system. Data obtained from the MFT-BIH database have been used, comprising a series of long term measurements in patients and also manual definition of the limit points performed by clinical physicians. The results clearly show the applicability of the MLP model in this biomedical task. Also, the combination of the results provided by all trained neural networks, instead of only the best one, has proven to improve the overall performance of the system.