Saturday, December 21, 2019

What Is The Five Public Available Brain Image Betration...

The proposed approach was evaluated using 5 public datasets based on two types of experiments. It was compared to two related registration methods and three state-of-the-art methods in terms of accuracy, solution regularity and computational cost. \Subsection{Data} 5 public available brain MR image datasets (BrainWeb, CUMC, IBSR, LPBA, OASIS) were selected in our experiments. These datasets have been used in image registration projects [#Klein2009, #Ou2014, #Hellier2001, #Rohlfing2012] for performance evaluation. In these datasets, the whole head was captured by a variety of imaging scanners and protocols. Accordingly, images had different size, voxel spacing and contrast level. All images were provided with corresponding label†¦show more content†¦These labels were combined to produce an initial non-background binary mask. We then used the c3d tool provided by the ITK-SNAP package [#Yushkevich2006] to sequentially perform a three-step operation on the initial mask: dilate by one voxel, fill holes with face connectivity and erode by one voxel in each dimension. Such an operation resulted a solid mask for the brain only region. Structures outside the mask were stripped to generate the brain images. An additional bias field inhomogeneity correction algorithm was performed based on the N3 algorithm [#Sled1998] using the c3d tool. †¢ CUMC: 12 images were acquired on a 1.5 T Siemens scanner by the Columbia University Medical Center. These images had relatively large variations in contrast. 128 class labels were provided covering detailed brain structures. This dataset was used for evaluation in [#Klein2009] and a cached copy http://www.synapse.org/#!Synapse:syn3217817 was maintained by the author of [#Klein2009]. We used this copy in our experiment because the original data source was inaccessible. †¢ IBSR: the Center of Morphometric Analysis at the Massachusetts General Hospital provided the Internet Brain Segmentation Repository (IBSR) dataset. It contained 18 images with three types of voxel spacings. A skull stripped version [#Rohlfing2012] including label modification was provided as an update to the original data at http://www.nitrc.org/projects/ibsr. 43

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.