Active contour models for knee cartilage and meniscus ultrasound image segmentation / Amir Faisal

• Main Author: Amir, Faisal
• Format: Thesis
• Published: 2017
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://studentsrepo.um.edu.my/7900/2/All.pdf; http://studentsrepo.um.edu.my/7900/4/faisal.pdf; http://studentsrepo.um.edu.my/7900/

Quantification of the cartilage degeneration as well as the meniscus degeneration and displacement requires segmentation of various parts of the knee joints in the twodimensional ultrasound images in order to provide a direct measurement of the cartilage thickness and the meniscus area and position, respectively. The goal in the knee cartilage ultrasound image segmentation is to locate the boundaries of a monotonous hypoechoic band between hyperechoic lines of the soft tissue-cartilage interface and of the cartilagebone interface. Hence, the true thickness between the two interfaces can be computed based on the segmented images. Meanwhile, the goal in segmenting the meniscus ultrasound image is to locate the femoral condyle, the meniscus, and the tibial plateau simultaneously. This thesis presents active contour models for knee cartilage and meniscus ultrasound image segmentation. Cartilage boundary segmentation using locally statistical level setmethod (LSLSM) and cartilage thickness estimation using the normal distance are presented. In addition, multiple active contours using scalable local regional information on expandable kernel (MLREK) have been proposed to capture multiple, separate objects of the femoral condyle, the meniscus, and the tibial plateau. Segmentation performance is then validated using Dice coefficient and Hausdorff distance metrics. Segmentation results of the presented methods are compared to the existing active contour methods in the attempt of segmenting the knee cartilage andmeniscus in the ultrasound images, which show an improvement on the segmentation performance offered by the proposed methods. The choice of various parameters in MLREK in response to the segmentation outcome is then investigated. A demonstration on how to choose the threshold value to adapt the kernel size in order to successfully reach the boundary concavity is given. The ability of multiple contours in preventing merging and overlapping in the shared boundaries of separate regions is shown. A flexibility in setting each contour with different parameter values for multiple structure segmentation is also illustrated. MLREK has shown to perform multiple object segmentation all at once in an ultrasound image. Application of the presented methods to segment a set of the knee cartilage and meniscus ultrasound images illustrates a good and consistent segmentation performance. The reproducibility of the ultrasound-based cartilage thicknessmeasurements using intraclass correlation coefficient and agreement between pairs of the measurements by the normal distance and the manual measurement using Bland-Altman analysis are determined. The cartilage segmentation possiblewith LSLSMhas allowed the obtained segmentation results to be used formaking the cartilage thickness computation. The robustness of themethods described against various thickness of the cartilage and various shapes and areas of themultiple objects indicates a potential of the methods to be applied for the assessment of the cartilage degeneration as well as the meniscus degeneration and displacement. The cartilage degeneration and the meniscus degeneration and displacement typically seen as changes in the cartilage thickness and the meniscus area and position can be quantified over time by comparing the cartilage thickness and the meniscus area and position at a certain time interval.