This paper will present an algorithm for Heart Rate Variability HRV signals classifications. In this algorithm we used Radon transform of binary matrix of scatter-gram of heart rate HRV signals to extract features of binary matrix. Artificial neural network (ANN) technique with back-propagation networks (BPN) was used for binary matrix features classifications. Radon transform with 90 projections was selected because it presented the best inverse Radon transform that gave a closer image of the original scatter-gram. The optimum numbers of neurons in the hidden layer of BPN is 145 was obtained. Two databases were formed, one for training and the second for testing the accuracy of the BPN to recognize on types of heart rate variability. The two database consist of HRV signal pathologies, sympathetic activity, normal cardiac, parasympathetic activity, arrhythmia, availability problem with breath, existence of stress and the composition of these pathologies. This algorithm present the accuracy of diagnosis for sympathetic activity, normal cardiac, parasympathetic activity, arrhythmia, availability problem with breath and existence of stress were 97,396%, 98,438%, 100%, 94,792%, 87,3265% and 91,146% respectively.

Keywords : Artificial Neural Network; Heart Rate Variability; Scatter-Gram; Radon Transform; Inverse Radon Transform; Back-Propagation Networks.