Segmentation of bone structures in Magnetic Resonance Images (MRI) for human hand skeletal kinematics modelling

Kurzfassung

Roboticists have shown a great interest in designing and building robotic hands, mainly because of the complex structure and high range of functionalities that the human hand has. Even though many hand models have been produced, being already used in many different applications, they are still not precise and skilled enough to be considered a proper replica of the human hand, due the size and exibility issues. In order to build an accurate representation of the human hand kinematics, in vivo 3D MRI (Magnetic Resonance Imaging) scans of the hand in different postures are used, here for the investigation of bone poses and finger movements. This thesis deals with an important step in deriving an accurate human hand model using MRI imaging: the precise segmentation of bone structures. In this case, the task of segmenting multiple bone structures is of great diffculty, as it has to take into consideration the characteristic problems encountered with MR imaging techniques. The segmentation algorithm focuses on the boundary between the bone marrow (yielding high gray level intensity values) and the cortical bone (represented by dark regions). Therefore, methods for reducing noise and enhancing edges are considered, and the extraction of the structures of interest is accomplished by using implicit active contours. After initialization is performed using Fuzzy C-means or thresholding techniques, a hybrid level set method, which includes region-based and edge-based information, is used for the segmentation of the bone volumes. The whole algorithm is implemented in MATLAB and comes in the form of a Graphical User Interface (GUI) which makes the interaction with the user much easier. For validation, the segmentation results are compared with manually acquired ground truth, providing an average sensitivity of 93% and a Dice similarity coefficient of 95%. The segmented bone structures are exported in 3D coordinates with respect to a fixed reference system, as it is requested for the further steps in the human hand modelling: pose estimation and registration of different postures.