2012/02/20 Fixed label map coverage problem.
Files changed: infant-neo-aal.img, infant-1yr-aal.img, infant-2yr-aal.img
There are 3 atlases dedicated for neonates, 1-year-olds, and 2-year-olds. Each atlas comprises a set of 3D images made up of the intensity model, tissue probability maps, and anatomical parcellation map. These atlases are constructed with the help of state-of-the-art infant MR segmentation and groupwise registration methods, on a set of longitudinal images acquired from 95 normal infants (56 males and 39 females) at neonate, 1-year-old, and 2-year-old.
We aim to construct a set of dedicated 0-1-2 infant atlases. Demographic information can be found in Table 1. First, an atlas is constructed for each of the three age groups: neonates, 1-year-olds, and 2-year-olds. Second, we apply state-of-the-art infant longitudinal segmentation and groupwise registration techniques for constructing the brain atlases. Third, a total number of 95 subjects with complete 0-1-2 longitudinal scans were collected for atlas building. T1 and T2 images were acquired with a 3T Siemens scanner. Thus, this dataset has large sample size and relatively high image quality. Finally, in each atlas, we provide the intensity model, tissue probability maps and anatomical parcellation map to meet the need of typical applications.
Table 1. Demographic information of the normal infants used in this study
|Scan||N||Gender||Age at Birth (weeks)||Age at MRI (weeks)||Group|
|First||95||56 m/39 f||37.9±1.8 (33.4 – 42.1)||41.5±1.7 (38.7 – 46.4)||Neonate|
|Second||94.2±3.4 (87.9 – 109.1)||1-year-old|
|Third||146.2±4.9 (131.4 – 163.4)||2-year-old|
In particular, based on the observation that the images acquired at 2-year-olds can be segmented with relative ease and higher accuracy, we use their segmentation results to guide segmentation of images from earlier age groups, i.e., neonates and 1-year-olds. At the same time, longitudinal correspondences across three age groups are also established. With the 2-year-old images as the bridge, the anatomical parcellation, i.e., Automated Anatomical Labeling (AAL) map, is propagated to images of neonates and 1-year-olds. Finally, images at each individual age group are registered cross-sectionally with a groupwise algorithm to form a respective atlas. The obtained infant atlases can be used as references for spatial normalization of a group of infant images, as tissue priors for atlas-based tissue segmentation, and as templates for structural labeling. The effectiveness of our atlases, in comparison with other 3 widely used atlases, is evaluated with typical atlas-based applications. Results indicate that our atlases yield the highest spatial-temporal consistency in spatial normalization and structural labeling of individual infant brain images. Additionally, our atlases give the best performance in atlas-based segmentation of neonatal images.
Cite the work
Feng Shi, Pew-Thian Yap, Guorong Wu, Hongjun Jia, John H. Gilmore, Weili Lin, Dinggang Shen,"", PLoS ONE, 6(4): e18746, 2011.
Fig. 1. The infant 0-1-2 atlas construction framework. From left to right, three main steps are involved in constructing the atlases: longitudinal tissue segmentation (step 1), anatomical labeling (step 2), and unbiased groupwise atlas construction (step 3).
Overview of Infant 0-1-2 Atlases
Fig. 2. The infant 0-1-2 atlases for (A) neonates, (B) 1-year-olds, and (C) 2-year-olds, respectively. In each panel, from top to bottom are the intensity model, three tissue probability maps for GM, WM, CSF, and anatomical parcellation map; from left to right, seven representative slices are shown. Note that region boundaries are shown for better visualization.