Left Ventricle Segmentation Using a Combination of Region Growing and Graph Based Method

• Published in: Iranian journal of radiology 2017-04, Vol.In press (In press)
• Main Authors: Ghelich Oghli, Mostafa, Mohammadzadeh, Maryam, Mohammadzadeh, Vahid, Kadivar, Sakineh, Mohammad Zadeh, Ali
• Format: Article
• Language: English
• Subjects: Algorithms; Automation; Computation; Image segmentation; International conferences; Methods; NMR; Nuclear magnetic resonance; Registration; Wall thickness
• ISSN: 1735-1065; 2008-2711
• DOI: 10.5812/iranjradiol.42272

Background: Left ventricle segmentation plays an essential role in computation of cardiac functional parameters such as ventricular end diastolic and end systolic volumes, ejection fraction, myocardial mass, and wall thickness and also wall motion analysis. Manual segmentation is also time consuming and suffers from inter and intra observer variability. Several approaches have been proposed that segment the left ventricle (LV) by automatic and semi-automatic methods, but the problem is still open due to the huge shape variety of the left ventricle and motion artifact. Materials and Methods: A robust semi-automatic approach is hereby presented for addressing the left ventricle segmentation problem. The presented method combines region information of the left ventricle with gradient and edge information in a graph framework. The LV region information is captured using our previously presented region growing method and is embedded into livewire framework. Results: The modified livewire that is presented here shows a great success in quantitative criteria over the publically available MICCAI 2009 left ventricle segmentation challenge database that contains 45 normal and abnormal cases. We have computed dice metric (DM) and average perpendicular distance (APD) for the proposed method and it outperformed the state of the art results over all papers that used the same database. Validation metrics, dice metric and average perpendicular distance were computed as 0.95 mm and 1.48 mm versus those of 0.87 - 0.93 mm and 1.76 - 1.81 mm obtained by other methods, respectively. Conclusion: Using semi-automatic approaches for cardiac segmentation yields satisfying results and this is because of incorporating radiologist’s experiences into the segmentation procedure. Maintaining image information to reduce user interaction is our goal for further researches.