Anatomic variability as measured with a 3D reconstructed Talairach atlas

Source: 1993 Nov;.
Author: Toga AW, Jones AS, Rothfeld JM, Woods RP, Payne BA, Huang C, Mazziotta JC, Cai R.

Abstract:
Accurate interpretation of PET or other scans of functional anatomy often requires correlations with standardized templates. MRI can be used to identify the anatomy on an individual basis and stereotactic atlases can be used as a common reference coordinate system. In addition, the atlas templates can aid in the segmentation of anatomic structures. The ability to compare brains from different subjects depends on the goodness of fit between scans and the sterotactic atlas. However, no single representation, such as an atlas, can prove accurate even within a homogeneous population of subjects. The ongoing development of a digital neuroanatomic atlas in the rat exemplifies how neuroanatomic variability is being used to generate a digital atlas of the average brain, incorporating the concept of confidence limits to most accurately represent a population. In the human, localization of structures using an atlas can be dramatically improved using transposition and regional scaling procedures based upon anatomic reference points. Different reference and coordinate systems vary in their localization of structures. For example, the 3D representation of the basal ganglia varies considerably when identified by Andrews and Watkins, Talairach and Tournoux, and Burzaco. Stereotaxic localization varies within an individual stereotaxic system also. Talairach stereotaxic space is commonly used in anatomical and multimodality mapping studies. The accuracy of sulcal patterns and other cortical landmarks has been evaluated in several studies, but its accuracy in localizing subcortical brain structures and morphometric accuracy following 3D reconstruction has not been established. Therefore, we set out to examine the anatomic variability of MRI scans as compared with the Talairach data using 3D reconstructed models of several subcortical structures. METHODS: As part of this project we created a computerized 3D version of the Talairach and Tournoux stereotactic atlas of the human brain. The plates were digitized and registered to each other using the numerical frame surrounding each plate. The cortical surface of each plate was contoured and triangulated to form a surface model while maintaining the original coordinate system. Three models were created; coronal, horizontal, and sagittal. MRI scans were obtained from 30 normal subjects. Surface models of subcortical structures from the previously positioned MRI data and the reconstructed Talairach atlas were generated using the same procedure as described above. Structures selected for 3D reconstruction, modelling and morphometric measurement were the putamen, head of caudate, thalamus, ventricular system, and the anterior commissure. RESULTS AND DISCUSSION: Overall, there was good agreement between the MRI data and the Talairach atlas for the structures measured. This is not surprising given that the bicommissural line was used as a reference. It has a constant relationship with intracerebral structures that is better than bony or other anatomic landmarks. The trend for more complete databases of the human brain will enable us to quantitate as well as visualize complex anatomic structures. Comparisons across individuals and modalities will require that we retain information about inherent variability. These results will form the basis for a collection of a homogeneous population, a description of normal morphometric variability and a framework for the development of a comprehensive and representative digital atlas of the human brain.