A comparative study of CNN and FCN performance for short-axis left ventricle segmentation in cardiac cine MR sequences

Automated segmentation of the left ventricle (LV) is a common task in the clinical diagnosis of heart disease. In this work, we investigate a deep learning-based method that can perform accurate automated segmentation of the left ventricle from short-axis cardiac MRI images. The designed FCN network constructs the final segmentation from several bottleneck layers, which are considered different representations to the input with different dimensionalities. Experimental results show a strong agreement between reference and automatically extracted contours, with an accuracy of 97,27%, a DSC of 98.20% and an HD of 6.158 mm. To evaluate the performance of our method and prove its technical and clinical contribution, a comparative study of our FCN architecture with the CNN architecture is carried out first. Then, we compared it with state-of-the-art methods using the same database. This comparison showed that FCN is the most suitable for this task. Our method offers clinicians the opportunity to solve the time, labor and error problems associated with manual measurements in current clinical routines.