Off-campus Michigan Tech users: To download campus access theses or dissertations, please use the following button to log in with your Michigan Tech ID and password: log in to proxy server

Non-Michigan Tech users: Please talk to your librarian about requesting this thesis or dissertation through interlibrary loan.

Date of Award


Document Type

Master's Thesis

Degree Name

Master of Science in Medical Informatics (MS)

College, School or Department Name

School of Technology

First Advisor

Jinshan Tang


Since ultrasonic energy was first used in the human body for medical purposes in the late 1940s, ultrasound imaging has become one of the most widely used medical imaging techniques because of its non-invasive and real-time imaging nature. However, one big challenge with ultrasound imaging techniques is speckles, which can be described as a kind of granular artifacts existing in ultrasound images. Speckles decrease ultrasound image resolution and thus significantly increase the difficulty of automatic image processing such as segmentation, classification, and so on. Therefore, speckle reduction has become an important research topic in ultrasound imaging.

In this master thesis, a cluster driven anisotropic diffusion (CDAD) filter for speckle reduction in ultrasound images is developed. The proposed anisotropic diffusion (AD) filter is based on multiplicative noise model and is driven by the K-means clustering algorithm. Compared with other AD filters, such as speckle reducing anisotropic diffusion (SRAD), the proposed method has many benefits. For example, instead of manually selecting a homogeneous sample region for the computation of diffusion coefficients, the proposed algorithm selects the homogeneous region automatically (based on K-means clustering results). The clustering result is also applied to influence the weights of speckle smoothing and edge enhancement in the proposed AD filter so as to improve the AD filter’s performance. In addition, in order to improve the computation efficiency, this thesis investigates Graphic Processing Units (GPU) technology for algorithm acceleration. Several GPU frameworks are investigated and the best one is selected based on the comparison of the performances. Based on the comparison of contrast values in both homogeneous regions and the set of edge points, CDAD filter shows better performance in speckle reduction and edge preservation than SRAD. Besides the 2-D cluster driven AD filter, we also extend it to 3-D cases, and GPU implementation on 3-D cluster driven AD filter is also investigated.