Automatic articular cartilage segmentation with multiple models

In this paper a method for cartilage segmentation of human knee from MRI images using multiple models is presented. Initially we trained a model with three types of knee MRI scans using existing set of large data called as training set. This training set includes features of pixels and their classes such as background and cartilage. Multiple k-NN models based on MRI scan type and slice number are used to segment cartilage from knee MRI scan. Multiple models are required for different types of MRI scans which have different levels of intensities. Each MRI scan has around 20 slices in which few slices in middle have more cartilage pixels than other slices. The performance of proposed method is evaluated on knee MRI scan and comparison is carried out with manual segmentation by a radiologist. It is revealed that proposed technique improves accuracy and processing time during segmentation of cartilage.