Articular cartilage (AC) is a flexible and soft yet stiff tissue that can be visualized and interpreted using magnetic resonance (MR) imaging for the assessment of knee osteoarthritis. Segmentation of AC from MR images is a challenging task that has been investigated widely. The development of computational methods to segment AC is highly dependent on various image parameters, quality, tissue structure, and acquisition protocol involved. This review focuses on the challenges faced during AC segmentation from MR images followed by the discussion on computational methods for semi/fully automated approaches, whilst performances parameters and their significances have also been explored. Furthermore, hybrid approaches used to segment AC are reviewed. This review indicates that despite the challenges in AC segmentation, the semi-automated method utilizing advanced computational methods such as active contour and clustering have shown significant accuracy. Fully automated AC segmentation methods have obtained moderate accuracy and show suitability for extensive clinical studies whilst advanced methods are being investigated that have led to achieving significantly better sensitivity. In conclusion, this review indicates that research in AC segmentation from MR images is moving towards the development of fully automated methods using advanced multi-level, multi-data, and multi-approach techniques to provide assistance in clinical studies.
Purpose: Knee Osteoarthritis (OA) progression and monitoring are possible by evaluating changes in subchondral bone tissues from magnetic resonance images but highly dependent on the accurate segmentation and quantitative measurement techniques. Existing methods to segment bone from MR images are either insensitive or rely on supervised computational techniques that require training models. Thus, the aim of this work is to develop an automated and unsupervised bone segmentation technique that can be used in large-scale longitudinal/multicentre studies. Methods: In this work, an automatic and unsupervised bone segmentation approach is developed and tested on 8 MR Datasets (DESS MR Seq., No. of Slices ¼ 160, TR/TE ¼ 16.3/4.7ms, ST¼ 0.7 mm, Res ¼ 0.365 Â 0.456 mm, FOV ¼ 140 Â 140 mm, FA ¼ 25 , MS ¼ 384 Â 384, BW ¼ 185 Hz/px obtained from OA Initiative) consisting of 1280 Slices. Segmentation of bones including femur and tibia is achieved in the two-fold process i.e. 1) Pre-processing and Pre-segmentation and 2) boundary correction leading to accurate segmentation. The pre-processing steps involve enhancement of knee MR images using local Gray level S-curve transformation technique that improves gradient magnitude of the image and leading to provide sharp edges and high contrast between adjacent tissues. Furthermore, an unsupervised method called distance regularised level set evolution (DRLSE) is used for pre-segmentation that evolves curve in the region of interest. For DRLSE, level set function (LSF) evolution occurs according to the initialization of two level set functions with the very small dimension corresponding to femur and tibia bone by obtaining the location using multi-resolution 3D scale-space technique. In the first process, all the slices in a single dataset are pre-segmented that results in some leakages associated at boundaries of bone. In order to resolve this issue, a boundary displacement technique is applied on two consecutive pre-segmented slices and distance between individual points of the current (reference slice with perfect boundaries) and successive (adjacent slice with leakages) boundary contours are measured to classify leakages with a threshold of 5 pixels. Once classified as leakage point, this was replaced with the corresponding points of the reference slice and the rest boundary is kept unaltered that is leading to provide accurate boundary correction. If there are any edge gaps between the contours, a linear interpolation method is applied to fill those gaps. In the same way, all the slices in each dataset are segmented and the results obtained from the segmentation are compared with manually segmented bones by experts. For the comparison, sensitivity, specificity, accuracy and dice similarity coefficient (DSC) are measured with their coefficient of variations as the number of datasets are rather small.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.