45 0

Fully Automatic and Efficient Midsagittal Plane Extraction, Alignment, and Skull Stripping in Neuroimages

Fully Automatic and Efficient Midsagittal Plane Extraction, Alignment, and Skull Stripping in Neuroimages
Other Titles
신경영상에서 자동적이고 효율적인 두개골 분리와 정중시상 두개골평면 추출 및 정렬
Hafiz Zia Ur Rehman
Prof. Sungon Lee
Issue Date
Automatic detection of the midsagittal plane (MSP) in neuroimages can significantly aid in registration of medical images, asymmetric analysis, and alignment or tilt correction (re-center and re-orientation) in brain magnetic resonance images (MRIs). Similarly, skull stripping (or brain extraction) from a volumetric dataset is a crucial pre-processing step in most of the brain MRI studies such as cortical surface reconstruction, brain volumetric measurement, and tissue identification. Manual MSP extraction, misalignment correction, and skull stripping, although doable, are extremely time-consuming and laborious to perform on a huge scale. It also demands an urbane knowledge of brain anatomy. Therefore, it is neither sufficient nor efficient. Firstly, an automatic technique for image alignment using principal component analysis (PCA) has been developed in this dissertation, which also addresses a critical problem of 180˚ rotation in PCA principal axes. Secondly, a fully automatic and computationally efficient algorithm has been proposed for MSP extraction that can substantially assist in registration, symmetric/asymmetric analysis, and alignment correction in brain MRI volumes. The parameters of MSP are estimated in two steps. In the first step, symmetric features and PCA-based technique are employed to vertically align the bilateral symmetric axis of the brain. In the second step, PCA is utilized to achieve a set of parallel lines (principal axes) from the selected two-dimensional (2-D) elliptical slices of brain MRIs, followed by a plane fitting using orthogonal regression. Thirdly, 3D-UNet has been used for skull stripping in brain MRIs. The 3D-UNet is an extended version of the previously proposed 2D-UNet which is based on a deep learning network, specifically, convolutional neural network (CNN). The designed algorithms have been applied and validated on different simulated and real publicly available image datasets. The results revealed that the proposed algorithms are efficient as compared to the existing methods and exhibit superior performance in terms of accuracy and precision. The contribution of this dissertation will allow a significant enhancement of the accuracy and precision in the quantification of medical imaging in skull stripping and registration.
Appears in Collections:
Files in This Item:
There are no files associated with this item.
RIS (EndNote)
XLS (Excel)


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.