PhD defense of Lin Li December 3, 12:00
The PhD defense of Lin Lin "Statistical Methods for Ultrasound Image Segmentation" will take place on December 3, 12:00 in ICT-507AB.
Everybody is welcome to attend!
Ultrasound imaging is commonly used in clinical diagnosis. Ultrasound imaging has several advantages over other medical imaging techniques such as X-ray, computed tomography (CT) and magnetic resonance imaging (MRI). The ultrasound imaging system is cheap, portable and has short acquisition times. Also the ultrasound technique is safe for patients.
At the same time, ultrasound imaging has some limitations which reduces its applicability. The quality of ultrasound images is relatively poor with speckle noise and artifacts. The objects’ edges in an ultrasound image are usually very blurry or missing at some places. Due to the noisy nature of ultrasound images, traditional segmentation algorithms have difficulty in producing desirable results. In this thesis, we focus on using statistical methods for ultrasound images segmentation, and propose three new algorithms.
The first algorithm combines the Chan & Vese algorithm and the Bhattacharyya distance. The Chan & Vese algorithm is a global algorithm, and it divides the image domain into two parts: internal and external regions. By using the Bhattacharyya distance, the proposed algorithm can maximize the difference between image regions and minimize the difference within the image regions. In the second algorithm, a localized region based active contour is used under the assumption that if the global requirement is not fulfilled in the image domain, then it can be satisfied in a small sub region with a high probability. Rayleigh distribution and shifted Rayleigh distribution are used to model the ultrasound image intensity distribution. The third algorithm combines a localized factor into the region scalable fitting algorithm and incorporates the Bhattacharyya distance in this algorithm.
To validate the performance of the proposed three algorithms, synthetic ultrasound images, phantom ultrasound images and patient ultrasound images are used as evaluation images. The segmentation results of the proposed algorithms show that the proposed algorithms are able to deal with ultrasound images with blurry edges and can produce desirable segmentation results. The segmentation results of the proposed algorithms are also compared with other algorithms to prove the efficiency of the proposed algorithms.