Publications
Text queries can be conducted by Author, Title, or Keyword.
Polynomial warping to a brain atlas: A tool for the study of
Alzheimer's Disease
Source: 1999;.
Author: Woods RP, Thompson PM, Mega MS, Toga AW, Mazziotta JC.
Abstract:
We evaluated the use of automated, intensity based polynomial warping as implemented in the Automated Image Registration package (AIR) to create a disease specific MRI atlas of patients with probable Alzheimer's disease. Accuracy was measured using manually identified cortical surfaces as a gold standard. Surfaces evaluated included the Sylvian fissures, the anterior and posterior calcarine sulci, the parieto-occipital sulci, the callosal sulci, and the cingulate sulci. All results obtained with polynomial warping were better than those obtained using the original manual method described by Talairach, et al. Each successively higher order polynomial warp through seventh order produced better overlap of homologous cortical surfaces. Improvements grew smaller, and computation times grew larger with each increase in polynomial order. Most brain regions evaluated had residual variability of less than three millimeters after eight order polynomial transformation. Polynomial warping with AIR is a valid method of anatomic standardization within the context of Alzheimer's disease. The continually evolving disease-specific atlas described here will facilitate the study of Alzheimer's disease by providing a standard anatomic framework into which diverse observations can be integrated and through which various subpopulations can accurately compared or identified. METHODS: All subjects underwent MR scanning using a standardized protocol. Scans from all subjects were manually edited to remove the scalp, skull and dura, as required for the registration and warp routines. Image registration was performed using the AIR package with new modifications to allow 6yh, 7th, and 8th order polynomial warping. Briefly, the algorithm automatically registers images by systematically altering the parameters of a spatial transformation model to minimize a cost function that reflects the degree of similarity between the image and designated target template. RESULTS AND DISCUSSION: The results presented here for patients with Alzheimer's disease support and extend our earlier validation of the use of polynomial warping for registration of normal subjects. In view of the atrophic changes present in the brains of patients with AD, the validity of an automated intensity based warping method was not a forgone conclusion. Our findings indicate that unlike the original landmark based Talairach method, automated polynomial warping is able to reduce residual anatomic variability to RMS errors less than 3 mm in most parts of the brain. Although much of the emphasis here has been on the elimination of anatomic variability between AD disease subjects as an unwanted confound, this variability has intrinsic interest from a morphometric standpoint. Differences in premorbid brain size have been described here in patients with AD, and such morphometric differences may provide important clues about the pathophysiology of the disease or about risks for developing the disease. These and other sytematic differences in brain shape may be corrected by a polynomial warping transformation and go unrecognized if the transformations themselves are not systematically analyzed. We are currently developing methods to allow the derived polynomial transformations to be analyzed for both global and local differences in brain size or shape.
Source: 1999;.
Author: Woods RP, Thompson PM, Mega MS, Toga AW, Mazziotta JC.
Abstract:
We evaluated the use of automated, intensity based polynomial warping as implemented in the Automated Image Registration package (AIR) to create a disease specific MRI atlas of patients with probable Alzheimer's disease. Accuracy was measured using manually identified cortical surfaces as a gold standard. Surfaces evaluated included the Sylvian fissures, the anterior and posterior calcarine sulci, the parieto-occipital sulci, the callosal sulci, and the cingulate sulci. All results obtained with polynomial warping were better than those obtained using the original manual method described by Talairach, et al. Each successively higher order polynomial warp through seventh order produced better overlap of homologous cortical surfaces. Improvements grew smaller, and computation times grew larger with each increase in polynomial order. Most brain regions evaluated had residual variability of less than three millimeters after eight order polynomial transformation. Polynomial warping with AIR is a valid method of anatomic standardization within the context of Alzheimer's disease. The continually evolving disease-specific atlas described here will facilitate the study of Alzheimer's disease by providing a standard anatomic framework into which diverse observations can be integrated and through which various subpopulations can accurately compared or identified. METHODS: All subjects underwent MR scanning using a standardized protocol. Scans from all subjects were manually edited to remove the scalp, skull and dura, as required for the registration and warp routines. Image registration was performed using the AIR package with new modifications to allow 6yh, 7th, and 8th order polynomial warping. Briefly, the algorithm automatically registers images by systematically altering the parameters of a spatial transformation model to minimize a cost function that reflects the degree of similarity between the image and designated target template. RESULTS AND DISCUSSION: The results presented here for patients with Alzheimer's disease support and extend our earlier validation of the use of polynomial warping for registration of normal subjects. In view of the atrophic changes present in the brains of patients with AD, the validity of an automated intensity based warping method was not a forgone conclusion. Our findings indicate that unlike the original landmark based Talairach method, automated polynomial warping is able to reduce residual anatomic variability to RMS errors less than 3 mm in most parts of the brain. Although much of the emphasis here has been on the elimination of anatomic variability between AD disease subjects as an unwanted confound, this variability has intrinsic interest from a morphometric standpoint. Differences in premorbid brain size have been described here in patients with AD, and such morphometric differences may provide important clues about the pathophysiology of the disease or about risks for developing the disease. These and other sytematic differences in brain shape may be corrected by a polynomial warping transformation and go unrecognized if the transformations themselves are not systematically analyzed. We are currently developing methods to allow the derived polynomial transformations to be analyzed for both global and local differences in brain size or shape.