Loading...
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 45363 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Fatemizadeh, Emad
- Abstract:
- Breast cancer is considered as the most common type of cancer in women worldwide and mammography is currently utilized as the principal method for screening the breast cancer. Breast Magnetic resonance imaging (MRI) can be used as a complementary imaging technique besides mammography. MRI technique involves scanning a patient before and repeatedly after the injection of the contrast agent (DCE-BMRI). This examination often takes 7-10 minutes and any movement of the patient’s breasts due to breath, heartbeat or deliberate movement, made in this relatively long acquisition period, leads to a distortion in images called motion artifact. This problem makes the quantitative analysis of the images inaccurate and unreliable. Thus, before any diagnostic procedures, the respective critical artifact should be compensated in a pre-processing step by applying image registration. In this research, a consistent registration algorithm based on a parametric approach in order to find the elastic transformation model is presented in which the geometric model is estimated both in forward and backward directions so that the respective transformation are inverse of each other. In this algorithm, an energy function as the linear combination of dissimilarity measure, inverse consistency and also regularization constraint is defined and the minimization of the function is searched. However some dissimilarity measures have a biased force resulted from the deformity. So we theoretically show that SSD and SAD measures are biased towards large gradients and large intensity differences that affect the estimation quality of motion artifact in a bad manner. Thus, we try to compensate the respective bias by independent calculation of the force magnitude and direction by means of some hard and soft thresholds. This way we will be able to reduce the dependency of the algorithm to some special pixels. Results of applying proposed modified similarity measures on 3D breast MR images and also 3D artificial images demonstrate better accuracy in the performance of the algorithm compared to some standard approaches. This improvements are actually clear in special areas with lower gradients or low intensity differences. On the other side, MR images contain intensity bias which are considered to be a low frequency behavior. Therefore, in order to remove such intensity bias, pixels are to be independent in a transform domain. To this end, residual complexity similarity measure based on DCT and WAVLET transforms are suggested. Furthermore, to increase the sparsity of the difference image a thresholding method is proposed so that the difference image is forced to be sparse. As the results on breast MR images show, by such way we can reach to an optimized registration
- Keywords:
- Breast Magnetic Resonance Images ; Consistent Elastic Registration ; Force Bias ; Intensity Bias
- محتواي کتاب
- view