Automated segmentation and quantitative analysis of the hip joint from magnetic resonance images

Abstract

The aim of this thesis is to develop a novel computer-aided system with advanced medical image processing approaches. This will allow automatic segmentation and quantification of the osteochondral elements (i.e. the articulating bones and cartilages) from high-resolution three-dimensional (3D) magnetic resonance (MR) images of the hip joint.This research is motivated by the importance of early detection of structural changes and degeneration of the bones and articular cartilages for good patient outcomes, particularly for early and pre-osteoarthritic conditions such as cam-type femoroacetabular impingement (FAI). MR imaging provides an optimal tool for in vivo assessment of the hip joint structure, including the bones and cartilages. This has generated extensive interest in the development of MR technologies to analyse cartilage morphology and assess biochemical compositions of the hyaline cartilage to facilitate early diagnostic and treatment for hip osteoarthritis (OA) and FAI.Quantitative analyses can provide useful morphometric data from complex MR data. In the hip joint, the segmentation of bones and cartilages is an essential prerequisite, which must be accurate, reliable and reproducible, for quantitative measurements. However, this is difficult and is traditionally performed using time-and expert-intensive manual or semi-automatic methods.The hypothesis behind this research is that accurate and reproducible quantitative data can be automatically obtained from high-resolution 3D MR images of the hip joint, through the use of advanced image processing techniques. To this end, this research focuses on two specific aims: Aim 1 -to develop and evaluate a fully automated segmentation approach to deliver accurate and reproducible bone and cartilage segmentations from high-resolution 3D MR images of the hip joint and Aim 2 -to automatically extract reliable and reproducible morphometric data based on the segmented subchondral bones and articular cartilages.The development of an automatic segmentation scheme for the bones and cartilages with high precision and reproducibility is firstly needed in order to provide a basis for subsequent quantitative measurements, which deliver reliable morphometric data of the segmented bones and cartilages for the use in early OA and FAI studies. To attain these aims, images were acquired using different MR sequences from a mixed demographic (male or female and young adult to elderly) of participants with a variety of femoral head-neck junction presentations but no apparent hip OA. Different sequence scans were used to image the same participant for the associated reproducibility experiments.Two state-of-the-art methods (multi-atlas-based and active shape model (ASM) based algorithms) i were developed and evaluated for automatic segmentation of proximal femurs and innominate bones from large field of view (FOV) MR images (water-excitation dual echo steady state (DESS) and multi echo data image combination (MEDIC)). The validation results have indicated accurate and robust...

Fully automatic system to detect and segment the proximal femur in pelvic radiographic images for Legg–Calvé–Perthes disease

“…We present the results in Several methods have been proposed for segmenting the proximal femur in radiological images. However, most of these methods employed CT or MRI [38][39][40][41][42][43] or implied mostly for hips with normal anatomy. 44,45 This paper demonstrated the potential of deep learning computer vision methods for segmenting radiographs in LCPD.…”

Segmentation of Femurs in X-ray Image with Generative Adversarial Networks

“…To determine the accurate edge in blurring region is also challengeable. Most proposed segmentation methods are performed on CT and MRI images [4,5,6,7,8,9,10,11], rather than on X-ray images. Chen et al [12] developed a model-based approach for automatically extracting femur contours from standard hip X-ray images.…”