Quantification of Articular Cartilage from MR Images Using Active Shape Models

Abstract

Osteoarthritis is the most common cause of disability in the developed world. One the most important features of the disease is the progressive thinning and eventual loss of articular cartilage which can be visualised using Magnetic Resonance (MR) imaging. A major goal of research in osteoarthritis is the discovery and development of drugs which preserve the articular cartilage. To guide this research, accurate and automatic methods of quantifying the articular cartilage are needed. All previous attempts to do this have used manual or semi-automated data-driven segmentation strategies. These approaches are labour-intensive and lack the required accuracy. We describe a model-driven approach to segmentation of the articular cartilage using Active Shape Models (ASMs) and show how measurements of mean thickness of the cartilage can be obtained. We have applied the technique to 2D slices taken from T _1 -weighted 3D MR images of the human knee. We give results of systematic experiments designed to determine the accuracy and reproducibility of the automated system. In summary, the method has been shown to be sufficiently robust and accurate for use in drugs trials.