The proposed algorithm was done through the following steps. Firstly the MR images are filtered using Gaussian filter in order to remove the noise while preserving the image’s edges1. The images are then analyzed to define the heart region, considered as the region of interest (ROI), through assuming the heart ventricles as two intersected circles1. The circular Hough transform was then used in order to locate the two circles corresponding to the heart ventricles. After defining the ROI, an initial contour extraction step was done to roughly define the shape and position of the LV related for each slice. In this step, we propose using the intensity time curve information for initialization then considering the LV as circle or ellipse as in the literature2,3. The proposed algorithm for initial contour generation can be summarized as follow: the intensity time curve was calculated, the average intensity in each time frame, for each slice. The best frame for extracting the initial contour should be found at the maximum peak of the first one third, as maximum filling for the LV occur at the first one third of the cardiac cycle, of the curve peaks. This frame was then converted to binary image using T value calculated based on the maximum value of threshold using Otsu’ method and intensity value at 80% of the area under probability density function curve. A combination of morphological operators (opening and closing) is then applied to remove the unwanted pixels. Though, the initial contour is extracted. A registration step was then implemented in order to compensate for the motion artifacts caused by breath and heart contraction movements. Finally, the levelset algorithm was employed in segmenting the final boundary of the LV at each frame. The main advantage of levelset is its flexibility and convenience in the implementation of active contours4. To the best of our knowledge, most of the literature uses the final contour of the previous frame as the initialization contour for the segmentation process. In this step, we propose defining the levelset initial contour, for each frame, based on the polar representation of images in order to increase the segmentation accuracy. The levelset initialization algorithm can be described as follows:

1-Transform each Cartesian frame into the polar coordinates representation using the center of the previously extracted initial contour.

2-Convert the gray level polar coordinate image to binary image using T threshold value.

3-Keep objects found at the first third, position of the LV, of the binary image. The largest connected object was survived.

4-Transform the image back to the Cartesian coordinates, where the boundary of survived object is considered as the initial contour of this frame.

Two short axis view datasets of cardiac magnetic resonance (CMR) perfusion imaging were used for performance evaluation. The datasets consist of 10 image sequences for 6 different patients; each has 3 slices (basal, middle, apex).The average accuracy of the proposed segmentation algorithm improved from 0.77 to 0.82 and from 6.8 to 6.3, measured using Dice metric (DM) and Hausdorff (HD) similarity metrics respectively, as seen in Figure1. Figure 2 shows examples of our segmentation results.In this study we have presented a modified workflow to segment the LV automatically using levelset method. The system introduces a new step to extract a real shape of the LV for each slice instead of assuming it circle or ellipse. We also present a new initialization for the levelset method based on polar representation of the images. The accuracy of segmentation results improved from 0.77 to 0.82, measured using DM.