Classification of heart diseases in ultrasonic images using neural networks trained by genetic algorithms

Recent studies show the effectiveness of neural-network-based computer-aided-diagnosis schemes for automated detection of various diseases, such as malignant breast mass and lung nodules. In this paper we describe a method for automated classification of ultrasonic heart (echocardiographic) images. The feature of the method is to employ an artificial neural network (NN) trained by genetic algorithms (GA's) instead of backpropagation. With the GA the optimal weighting coefficients of the NN are determined. Also the method shows a faster convergence for obtaining the optimal solution in NN training. Experiments on different data sets show the superiority of the GA-based method over backpropagation for classification.