An automatic technique for left ventricle segmentation from msct cardiac volumes

In this research, an automatic technique to segment the left ventricle from the heart information in multislice computed tomography images is proposed. A preprocessing stage is considered as a necessary preliminary task for diminishing the artifacts impact in the image analysis. With this idea, a similarity enhancement that combines a smoothed version of the original volume with a processed volume using mathematical morphology is used. This preprocessing approach is compared with respect to other strategies. After, a volume-of-interest is defined in order to isolate the cavity using two cropping planes detected with least squares support vector machines. Finally, the segmentations are obtained using both a region growing algorithm and a level sets algorithm. The robustness of each enhancement strategy is validated by performing the segmentation of images. This evaluation considered the Dice score, and both volume and surface errors. A clinical dataset from 12 patients is used in the inter- and intra subject evaluation. During intra-subject validation the proposed scheme achieves the best results, while a modified version of this scheme achieved the best performance during inter-subject validation.