INTRODUCTION: Chagas' disease is an endemic rural disease prevalent in much of Central America and South America, and approximately half of the patients infected with the parasite Trypanosoma cruzi show no clinical, electrocardiographic and radiological cardiac involvement. This paper, however, proposes a technique for the diagnosis of Chagas' disease based on ECG signals, which extracts relevant information from these signals. METHODS: Two approaches are studied and implemented. Both approaches use heart rate variability (HRV) signals, and classification by a neural network, more specifically, the Kohonen self-organizing map. The HRV, which reflects sympathetic and parasympathetic autonomic neural modulation of the heart, is evaluated based on continuous series of RR intervals, calculated from 5-minute records of conventional ECG. In the first approach, statistical/temporal indexes obtained directly from the HRV signals are used as neural network inputs for training and testing of the classification method. In the second approach, derivative of Gaussian (DoG) wavelet scalograms are used to evaluate the HRV signals. Scalographic indexes are used as neural network inputs for training and testing of the algorithm. Kohonen topological maps are used to compare the ability of these HRV indicators of discriminating between patients with Chagas heart disease, Chagas indeterminate heart disease, and normal subjects. Conventional temporal indicators and indicators obtained from DoG scalograms are compared. RESULTS: Results of the application of the proposed methods to HRV signal databases, and performance comparisons, are presented. The results show that scalographic indicators have superior discriminatory power than conventional time-domain indicators. Based on an analysis of statistical significance, we show that the average power of the high-frequency band of the scalogram power spectral density is the indicator with greatest discriminatory power (p < 0,05 for all 3 cases). CONCLUSION: The proposed method has the ability to discriminate between normal subjects, subjects with Chagas cardiomyopathy, and subjects with the indeterminate form of Chagas' disease. It was observed that scalographic neural networks present greater discrimination ability than temporal neural networks.
Chagas' disease; Scalogram; Wavelets; Neural networks