Segmentation Tracking and Evaluation in Tendon Medical Images

  • 莊 柏逸

Student thesis: Doctoral Thesis


Tendinopathy is a common clinical issue in recent years Ultrasound and H&E stained tendon microscopy images are commonly used for the clinical diagnosis of tendinopathy severity Due to property variations of ultrasound images traditional methods cannot effectively segment the finger joint’s tendon structure Moreover speckle noise and out-of-plane issues make the tendon tracking process difficult In microscopy as most of the reported tendon tissue evaluations are manual objective quantitative analyses remain scant In this thesis we developed three tracking and segmentation methods for ultrasound and microscopy images In axial view ultrasound images an adaptive texture-based active shape model (ATASM) method is proposed for segmenting the tendon and synovial sheath Adapted weights are applied in the segmentation process to adjust the contribution of energy terms depending on image characteristics at different positions The pathology is then determined according to the wavelet and co-occurrence texture features of the segmented tendon area In order to automatically track the tendon motion we developed a new method called optical-flow-trend-based multi-kernel block matching (OFTB-MKBM) that combines the advantages of optical flow and multi-kernel block matching in the sagittal view ultrasound images For every pair of adjacent image frames the optical flow is computed and used to estimate the accumulated displacement The proposed method selects the frame interval adaptively based on this displacement Multi-kernel block matching is then computed on the two selected frames To reduce the tracking errors the detailed displacements of the frames in between are interpolated based on the optical flow results In microscopy we develop two automatic systems for tendon nuclei segmentation from micro- and macro-view images In micro-view microscopy the proposed sampling-based thresholding segments the nuclei by feature sampling from a small number of selected nuclei For complex images with more nuclei the Laplacian-based thresholding is proposed to segment the nuclei based on the nuclei boundary information The system selects the thresholding result depending on the number of segmented nuclei The segmented nuclei are then classified as normal or abnormal based on their characteristics In macro-view microscopy we first segment the vessel and calcified regions according to the color information The remaining regions are then classified into normal or abnormal tissues based on their saturation values In the experiments the results of proposed ATASM had fewer errors with respect to the ground truth than the traditional active contour model (ACM) and active shape model (ASM) The segmentation results were all acceptable in data of both clear and fuzzy boundary cases in 74 images And the symptom classifications of 42 cases were also a good reference for diagnosis according to the expert clinicians’ opinions The mean absolute error of OFTB-MKBM evaluated by using cadaver data was less than 0 05 mm The proposed OFTB-MKBM also tracked the motion of phantom and tendons in vivo and achieved the better tracking results than optical flow and multi-kernel block matching The average sensitivity of micro-view segmentation method in the microscopy achieved 90 85% with respect to expert judgments The results are better than the ones using maximum entropy threshold and level set segmentation structure The experiment also showed that the proposed micro- and macro-view classifications have good correlation with each other
Date of Award2017 Sep 6
Original languageEnglish
SupervisorYung-Nien Sun (Supervisor)

Cite this