Automated Segmentation of Whole Cardiac CT Images based on Deep Learning

Abstract

Segmentation of the whole-cardiac CT image sequence is the key to computer-aided diagnosis and study of lesions in the heart. Due to the dilation, contraction and the flow of the blood, the cardiac CT images are prone to weak boundaries and artifacts. Traditional manual segmentation methods are time-consuming and labor-intensive to produce over-segmentation. Therefore, an automatic cardiac CT image sequence segmentation technique is proposed. This technique was employed using deep learning algorithm to understand the segmentation function from the ground truth data. Using the convolution neural network (CNN) on the central location of the heart, filtering ribs, muscles and other contrasting contrast are not an obvious part of the removal of the heart area. Staked denoising auto-encoders are used to automatically deduce the contours of the heart. Therefore, nine cardiac CT image sequence datasets are used to validate the method. The results showed that the algorithm proposed in this paper has best segmentation impact to such cardiac CT images which have a complex background, the distinctness between the background and the target area which is not obvious; and the internal structure diversification. It can filter out most of the non-heart tissue part, which is more conducive to the doctor observing patient's heart health.

Semantic Cardiac Segmentation in Chest CT Images Using K-Means Clustering and the Mathematical Morphology Method

“…The experiment was trained on CT images with 20 volumes of size 300 × 300 × 188 and achieved an average dice similarity coefficient of 90.8%. Ahmed et al [24] conducted whole-heart segmentation using supervised deep learning by defining a CNN network and using stacked denoising auto-encoders. The experiment was trained on CT images of eight subjects and achieved an average accuracy of 93.77%.…”

“…Later, the deep learning approach showed promising successful performance [7,9,10]. The deep learning approach has two learning manners, including supervised learning [22][23][24][25][26], which requires a ground truth to align the loss function, and unsupervised learning [27][28][29], which learns the features without ground truth labeling.…”

“…Labeling the ground truth for cardiac segmentation requires the extensive manual annotation of images by the radiologist, which can be time-consuming and labor intensive [7][8][9][10][22][23][24][25][26]. Due to the difficulty in obtaining annotated data and the required expertise as an annotator, an unsupervised approach has been considered in this study.…”

A Review on Brain Glioma Segmentation in MRI images using Deep Learning

“…LGG [17] Figure 2: HGG [17] Precise segmentation of gliomas and their structures is not only critical for care planning but also for follow-up evaluation. Manual segmentation, however, is tedious and vulnerable to inter-and intra-rater errors which are hard to characterize [5], [6]. However, most doctors typically use rough measures for the assessment process.…”

A Heart Segmentation Algorithm Based on Dynamic Ultrasound

“…Tong et al [13] constructed 3D deep supervised U-Net to segment all hearts. Ahmed et al [14] built a deep network to segment the heart. Gao and Lu [15] focused on fetal baseline to realize classification and extraction.…”