چكيده به لاتين
Intravascular ultrasound (IVUS) is a medical imaging tool used for determining the degree of stenosis of the artery vessel and the atherosclerotic plaque assessment. In the longitudinal view of the acquiesced image sequence, the site of the diseased segment and its length can be estimated and precise information is provided about the plaque structure by the short axis view. Because of cardiac dynamics, imaging catheter has a complex motion during a pullback in the coronary artery. Thus, the considerable displacement between adjacent frames is resulted, leading to global and local misalignments of vessel structures in longitudinal views. Detecting these motion artifacts in IVUS image sequences is one the main fields for IVUS image processing. This field can be divided into three main branches including, heart motion phase detection, rigid registration and non-rigid registration. The first group detects IVUS frames with the same heart phase. The second group proposes methods to detect the rigid motion between IVUS frames. The third group includes non-rigid registration methods and their applications. In this thesis, we used the multi-subband spatio-spectral transforms to detect and compensate the motion artifacts in IVUS sequences. In order to detect the heart motion phase, we used the phase information of the dual-tree complex wavelet transform. We analyzed this method using both synthetic and real datasets and compared it with a gold standard dataset. In all cases the proposed method resulted in better performance compared to the other studied methods. In case of rigid registration of IVUS sequences, we proposed a new windowed gradient descent method. The proposed method could achieve CAR values higher than 0.2 and SR values higher than 0.25 and outperformed other methods. In case of the non-rigid registration methods, we proposed a new model based on dual-tree complex wavelet transform. The proposed method is tested on the CUMC12 public MRI dataset and provided mean of TO value higher than 0.5233 and outperformed well-known IRTK, SyN and SPM-DARTELL methods. We also used the Curvelet transform to propose a new method for non-rigid registration of IVUS sequences which could achieve a Hausdorff error lower than 0.79.
