Muhammad Hanif , Ahmad Nizar (2017) Pattern recognition of heart valve in echocardiogram using convolutional neural network / Muhammad Hanif Ahmad Nizar. Masters thesis, University of Malaya.
Abstract
One of the causes of heart failure is valvular heart disease which can be diagnosed using echocardiogram. However, using this machine requires a degree of skill in order to locate the heart valves. We propose the use of a trained neural network to locate the position of the aortic valve from an echocardiogram image as an assistive technology during echocardiogram examination. The neural network AlexNet was used in this study which was trained using a deep learning platform, NVIDIA DIGITS. 58 of patients’ echocardiogram were used to train the AlexNet which were obtained from the National Heart Institute. After training the AlexNet, it was tested with 25 images and the resulted images were validated with a sonographer. Testing the AlexNet within the deep learning platform showed it was able to achieve an accuracy of 99.87% for aortic valve and 99.69% for background images. The qualitative comparison from the sonographer with the resulted image was that the trained neural network was able to localize the image accurately but may not be able to segment the valve precisely. This study was able to demonstrate the possibility of utilizing neural network to develop assistive technology for medical devices such as echocardiogram. Recommendations to increase the performance of the neural network such as using a neural network with more layers and providing a larger dataset for training were also explained.
Actions (For repository staff only : Login required)