| Abstrakt: |
The left ventricle dysfunction is the root cause for major cardiac arrests across globe. Heart functional indices such as End diastolic volume, systolic volume and Ejection Fraction are usually computed from manual segmentation of left ventricle wall (LV) in short axis MR images. This is very time consuming and suffers inter-intra observer variation due to change in shape, size and contrast of LV wall in a cardiac cycle. In this paper, deformable models are improved as compared to existing approaches with respect to their adaptation on the number of iterations per image by automatic stopping of curve evolution. This has reduced the time consumption for segmentation. The proposed strategy is based on the integrated approach of machine learning and adaptive deformable flow model with no user interaction for LV initialization. CNN is used for LV detection and shape inference which makes the approach fully automatic. The proposed methodology has improved the existing parametric deformable models in terms of the time consumption and accuracy. The proposed integrated framework has significantly improved the segmentation of LV wall for better clinical inference. For assessing the accuracy of segmentation between proposed method and ground truth, validation metrics are calculated such as percentage of good contours, Dice Metric, APD, Conformity coefficient, Pearson's Coefficient for EDV and ESV. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink