A constructive neural network for detecting left ventricular hypertrophy

Neural networks apply non-linear statistics to classification problems. One such problem is a medical diagnosis of a left ventricular hypertrophy. The objective of this paper is to develop a model based on self constructed artificial neural networks to support a medical diagnosis. The database used was composed of 101 individuals registered at the university hospital of Federal University of Pernambuco. Input data included: anthropometric, biographical, the biochemistry series, hormones, results of the ABPM-24h, echocardiography and electrocardiographic variables. Many combinations of the available variables were tested to select those that produced the best performance in terms of correct classification rate, sensitivity, specificity and area under ROC curve. The results were compared to those obtained with a classical approach used in medical area, logistic regression. The echocardiography findings were used in this study as a gold standard. Results show a good accuracy rate in classification using the neural system and encourage future improvements.