Artificial Neural Network Application in Classifying the Left Ventricular Function of the Human Heart Using Echocardiography

This paper proposes a methodology to assist Acute Care Physicians in diagnosing the heart condition of a patient with high accuracy within the least possible time. As the initial step in diagnosing the heart condition, we focused on automatically classifying the Left Ventricular Systolic function of the Human Heart into two categories as normal or abnormal. The proposed solution is a combination of image processing techniques and Artificial Neural Networks (ANN). Images obtained from Echocardiography videos were subjected to a series of image pre-processing techniques. The two parameters; Left Ventricular Internal Systolic diameter (LVIDs) and Left Ventricular Internal Diastolic diameter (LVIDd) were extracted from the echocardiography videos using a feature extraction algorithm. These two parameter values were used to calculate the Ejection Fraction (EF). LVIDs, LVIDd and EF values were then used as inputs to the Artificial Neural Network. A feed-forward back propagation neural network was trained as a classifier to distinguish between normal and abnormal Left Ventricular Systolic function. The cardiologist's decision was used as the expected output for the training of the Artificial Neural Network. A dataset of 50 images that included both normal and abnormal heart condition was used. The trained Artificial Neural Network could classify the images with an accuracy of 98%.