(a), (c), and (e) are three adjacent images of a 3-D MRI volume with number 77, 78 and 79 respectively. We can see they vary lightly from each other. (b), (d), (f) show their segmentation results.

Contexts in source publication

Context 1

... 3-D MRI volumes, the change in size and intensity for each structure varies lightly from one image to the adjacent. In other words, the differences between a few adjacent images are very small, it can be seen clearly from Fig. 5a, c and e. So after one slice of a 3-D MRI volume has been segmented by the method described in Sec- tion 2, its neighboring images can be segmented simply by propagating its informa- tion. The information is consisted of watershed lines, mean values and variances of homogeneous ...

Context 2

... seg , if its mean value is greater than M min and smaller than M max of its class, and its variance is less than V max of its class, the region contain only one tissue. We can assign all pixels in it to its class. Otherwise, the region is inhomogeneous and we use the kNN classifier with the training set obtained in I key to classify pixels in it. Fig. 5b, d, and f show the segmentation results of the images in Fig. 5a, c, and e respectively. Only the image in Fig. 5d is segmented by the three stage method, others are segmented by the information of Fig. 5d. Using the extension method, less than half of the slices in a volume need to be ap- plied with the watershed algorithm, FCM cluster ...

Context 3

... max of its class, and its variance is less than V max of its class, the region contain only one tissue. We can assign all pixels in it to its class. Otherwise, the region is inhomogeneous and we use the kNN classifier with the training set obtained in I key to classify pixels in it. Fig. 5b, d, and f show the segmentation results of the images in Fig. 5a, c, and e respectively. Only the image in Fig. 5d is segmented by the three stage method, others are segmented by the information of Fig. 5d. Using the extension method, less than half of the slices in a volume need to be ap- plied with the watershed algorithm, FCM cluster and MCD estimator and the rest slices will contain only a small ...

Context 4

... max of its class, the region contain only one tissue. We can assign all pixels in it to its class. Otherwise, the region is inhomogeneous and we use the kNN classifier with the training set obtained in I key to classify pixels in it. Fig. 5b, d, and f show the segmentation results of the images in Fig. 5a, c, and e respectively. Only the image in Fig. 5d is segmented by the three stage method, others are segmented by the information of Fig. 5d. Using the extension method, less than half of the slices in a volume need to be ap- plied with the watershed algorithm, FCM cluster and MCD estimator and the rest slices will contain only a small number of pixels that need classifying by the ...

Context 5

... class. Otherwise, the region is inhomogeneous and we use the kNN classifier with the training set obtained in I key to classify pixels in it. Fig. 5b, d, and f show the segmentation results of the images in Fig. 5a, c, and e respectively. Only the image in Fig. 5d is segmented by the three stage method, others are segmented by the information of Fig. 5d. Using the extension method, less than half of the slices in a volume need to be ap- plied with the watershed algorithm, FCM cluster and MCD estimator and the rest slices will contain only a small number of pixels that need classifying by the kNN. Thus, the overall execution time and computation of 3-D MRI volumes is ...