Application of artificial neural networks in automatic cartilage segmentation

Magnetic resonance imaging of articular cartilage has recently been recognized as the best non-invasive tool to visualize the cartilage morphology, biochemistry and function. In this paper, the challenging issue of automatic determining the cartilage volume is tackled. First, algorithms based on classical segmentation methods such as thresholding, poly-fitting, and average weight calculating are combined and tailored to develop a clustered segmentation method. Second, artificial neural network (ANN) is applied to improve the developed method by better coping with the nonlinearity and unidentified MRI image noises. This ANN is then applied with the active contour models (Snake) to provide the desirable outcome. Computational examples are given to justify our analysis and demonstrate the proposed method.