|Title||Image Segmentation Using Varying Ellipses|
|Publication Type||Conference Paper|
|Year of Publication||2010|
|Authors||Kazemzadeh, F., T. M. Haylock, and A. R. Hajian|
|Conference Name||Canadian Student Conference on Biomedical Computing and Engineering|
|Publisher||University of Waterloo|
|Conference Location||Waterloo, Ontario, Canada|
Highly accurate segmentation techniques are needed in the field of medical imaging. Segmenting medical images is difficult due to varying contrast and the high level of noise inherent in many medical imaging modalities. A segmentation routine based on growing ellipses is shown to be able to segment an object from background in these images. The algorithm increases the size of each ellipse until a threshold is met, indicating the edge of the object. The object's center position is initialized by the algorithm operator and a suitable threshold is found iteratively and interactively. Using standard image processing test images, results show as high as 98.5% successful segmentation. Success is gauged by comparing results to a manually segmented image and false positive segmentation tends to be low. Suitable images for ellipse segmentation are symmetric and well enclosed, which are typical characteristics found in a medical image.