The performance of this convolution neural network architecture is evaluated with recent architectures using various similarities and distance-based majors as well as ejection fraction correlation with ground truth segmentation labelled images.
Deepa Madathil, Associate Professor, Jindal Institute of Behavioural Sciences, O.P. Jindal Global University, Haryana, India.
Pallavi Kulkarni, School of Electronics Engineering, Vellore Institute of Technology, Vellore, Tamilnadu, India; Pimpri Chinchwad College of Engineering and Research, Ravet, Pune, Maharashtra, India
Echocardiography is a widely used ultrasound imaging technique for cardiac health diagnosis. Echocardiography segmentation is a crucial process to evaluate multiple cardiac parameters like ejection fraction, heart wall thicknesses, etc. Recently machine learning techniques especially deep learning using convolution neural network models are finding increasing applications for echo image analysis including its segmentation. In this paper, we have presented a unique convolution neural network (CNN) model for automatic left ventricle (LV) segmentation of echo images. Denoising and feature extraction processes are integrated with the CNN model to enhance its prediction accuracies after training.
The proposed system is trained on two-dimensional sequence images of 70 patients and tested on data of 12 patients. An automatic method for evaluation of ejection fraction is appended using the LV segmentation predictions generated by the CNN model.
The performance of this CNN architecture is evaluated with recent architectures using various similarities and distance-based majors as well as ejection fraction correlation with ground truth segmentation labelled images. CNN layer visualisation methods are applied to obtain deeper insight into the trained network.
Published in: International Journal of Biomedical Engineering and Technology
To read the full article, please click here.