Region-Based Ultrasound Image Segmentation and Motion Estimation for Liver Tracking
Date Issued
2012
Date
2012
Author(s)
Lin, Ching-Kai
Abstract
Medical imaging is an effective modality that enables clinicians to study the shape, size, and dynamics of the target. In this thesis, an image-guided system for liver motion tracking is presented. Traditional medical image-guided systems include CT, MRI, X-ray and ultrasound image (US). While considering frame rate and invasiveness, the ultrasound image-guided system is proposed for tracking the liver motion. The main objective is to use ultrasound image for continuously tracking liver motion. Ultrasound images have lower resolution and ill-defined edges in noisy images.
To remedy the problem, a new processing framework for tracking liver motion for ill-defined edges in noisy images is proposed. The proposed framework includes two stages: region segmenting and motion estimation. The region segmenting is done by the K-means clustering and the motion is estimated by template matching.
1) Region segmenting by K-means clustering. The traditional method is to use edge-based algorithms. However, the ultrasound image quality is too poor to segment the liver region by edge-based algorithms, since the trouble of edge-based algorithms is the presence of noise that results in random variation in level from pixel to pixel. Therefore, the region-based algorithm is proposed to segment the liver region in this thesis.
2) Motion estimation by template matching. After the possible liver location is found, template matching is used to estimate more accurate liver motion. This method can locate the liver more accurately.
The ultrasound scenario is experimentally tested. This thesis achieves liver region segmentation and liver motion tracking by analyzing dynamic information acquired through sequential images. The root mean square (RMS) tracking error of the image segmentation and template matching results is 1.71 mm and 1.05 mm; the percentage is 7.18 and 4.3 percent, respectively.
Subjects
ultrasound image technology
image segmentation
motion estimation
K-means clustering
template matching
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-101-R99921056-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):b87b5bf7ddde47cab5b81e5c479ffef8