Analyzing the Effectiveness of Extracting Hierarchical Features in Improving Recognition Rate of Heart Failure in Electronic Healthcare

Abstract: Recently primary recognition of heart failure using applicable software has become a significant subject in electronic health. A vital step of this process is to design a classifier to distinguish the images belonging to normal and abnormal patients. Targets: The aim of this article is to introduce a new classifier for recognizing left ventricle failures which automatically extracts features from raw data. For this purpose the Convolutional Neural Network (CNN) is presented as a member of deep learning family which the feature extraction step is happened automatically in its middle layer. Methodology: The proposed scheme was simulated by using the software package python 3.6 and its performance was evaluated on MICCAI 2009 data set as an official reference in left ventricle segmentation challenge. Founded: The CNN reached the accuracy of 90 percent for total normal and abnormal cases. This result was approximately 8 percent better than the accuracy which had been obtained by using un-hierarchical features in training neural networks. Conclusion: The obtained results confirmed the effectiveness of using hierarchical features and deep learning paradigm against un-hierarchical features in classifying heart failures.

Analyzing the Effectiveness of Extracting Hierarchical Features in Improving Recognition Rate of Heart Failure in Electronic Healthcare Keywords:

Hierarchical Feature Extraction , Deep Learning , Convolutional Neural Network , Heart Failure , Classification.