|
|
Yi-Yu Chou, Natasha Lepore, Greig I. de Zubicaray, Stephen E. Rose,
Owen T. Carmichael, James T. Becker, Arthur W. Toga, Paul M. Thompson. 2006.
Automated Ventricular Mapping via Multiple Surface-based Atlases.
13th Annual Meeting for the Organization of Human Brain Mapping.
| "By integrating multiple labels, random digitization errors from individual hand-traced segmentations are reduced. The resulting average models are robust to registration inaccuracies, which may result from non-global minima of the intensity-based cost function. This work complements other multi-atlas or meta-analytic studies [2,3] that combine multiple segmentations to generate results with higher fidelity to an expertly defined gold standard." |
| "In this study, we established experimentally how label propagation and decision fusion can be combined to automate the segmentation of MR images of the human brain. Although the principle has been described previously (Rohlfing et al., 2004a; Svarer et al., 2005; Klein et al., 2005), we present the first comprehensive assessment using carefully validated input data of 30 subjects." |
| "In recent work by Klein and Hirsch (2005), an automatic method is described which enables identification of large cortical structure volumes and can also be improved using fusion of multiple classifiers (Klein et al., 2005)." | "Label volumes represent classifiers that assign a structure label to every voxel in the corresponding MR image volume. To combine the information from multiple individual propagated label volumes into a consensus segmentation, the classifiers were fused: the consensus class of each voxel was defined as the modal value of the distribution of the individual label assignments (vote rule decision fusion as described by Kittler et al., 1998, and used by Hammers et al., 2003; Klein et al., 2005)." |
| "A recent paper by Klein et al. (2005) also showed improvements in label consistency when fusion methods were used on atlases generated by their automated approach. This group reports that maximum label agreement was 82.50% for 10 classifier subjects. Applying the same measure to our data, we find a value of 85.1% for fusion of 9 classifiers. The difference may be attributed to the smaller number of contrived boundaries in our source labels and to the use of a higher dimensional registration procedure. We have also used a larger number of labeled brains and so have been able to move closer to convergence of the labels." |
Talks given at Cold Spring Harbor Laboratory (Aug. 2007), Columbia University (Feb. 2005, Oct. 2006, May 2007), the International Conference on Complex Systems 2004, and Neuroscience 2002.
Ph.D. thesis, Cornell (May, 2004)
Automated brain labeling with Mindboggle.
Poster (full size) for Human Brain Mapping 2003
Arno Klein and Joy Hirsch. 2003. Mindboggle: new developments in automated brain labeling.
9th Annual Meeting for the Organization of Human Brain Mapping.
Poster (full size) for Human Brain Mapping 2002
Arno Klein and Joy Hirsch. 2002. Fully-automated nonlinear labeling of human brain activity.
8th Annual Meeting for the Organization of Human Brain Mapping.
Poster (full size) for Human Brain Mapping 2001
Arno Klein and Joy Hirsch. 2001. Automatic labeling of brain anatomy and fMRI brain activity.
7th Annual Meeting for the Organization of Human Brain Mapping.