An intelligent diagnostic system for screening newborns

摘要

The goal of this research is to devise and develop an intelligent system for analyzing heart sound signals, able to support physicians in the diagnosis of heart diseases in newborns. Many studies have been conducted in recent years to automatically differentiate normal heart sounds from heart sounds with pathological murmurs using audio signal processing in newborns. Serious cardia pathology may exist without symp- toms. Since heart murmurs are the first signs of heart disease, we screen newborns for normal (innocent) and pathological murmurs. This thesis presents a variety of techniques in time-frequency domain such as Cepstrum, Shannon energy, Bispe trum, and Wigner Bispe trum for feature extraction. A comparison of these techniques is considered to feature selection which has been used to reduce the size of the feature vector. In the final step, different lassi ers and techniques, e.g., Multi layer perc eptron (MLP), decision tree, Classification and Regression Trees (CART) and ensemble of decision trees, are applied on data in order to a hieve highest performan e. High classifi cation accuracy, sensitivity, and specifity have been obtained on the given data by CART. The validation process has been performed on a balanced dataset of 116 heart sound signals taken from healthy and unhealthy medical cases.