Automatic matching of homologous histological sections

Abstract

The role of neuroanatomical atlases is undergoing a significant redefinition as digital atlases become available. These have the potential to serve as more than passive guides and to hold the role of directing segmentation and multimodal fusion of experimental data. Key elements needed to support these new tasks are registration algorithms. For images derived from histological procedures, the need is for techniques to map the two-dimensional (2-D) images of the sectional material into the reference atlas which may be a full three-dimensional (3-D) data set or one consisting of a series of 2-D images. A variety of 2-D-2-D registration methods are available to align experimental images with the atlas once the corresponding plane of section through the atlas has been identified. Methods to automate the identification of the homologous plane, however, have not been previously reported. In this paper we use the external section contour to drive the identification and registration procedure. For this purpose, we model the contours by B-splines because of their attractive properties the most important of which are: 1) smoothness and continuity; 2) local controllability which implies that local changes in shape are confined to the B-spline parameters local to that change; 3) shape invariance under affine transformation, which means that the affine transformed curve is still a B-spline whose control points are related to the object control points through the transformation. In this paper we present a fast algorithm for estimating the control points of the B-spline which is robust to nonuniform sampling, noise, and local deformations. Curve matching is achieved by using a similarity measure that depends directly on the parameters of the B-spline. Performance tests are reported using histological material from rat brains.

Statistical parametric mapping applied to an autoradiographic study of cerebral activation during treadmill walking in rats

“…Registration methods for autoradiographs have used, for example, artificial landmarks (Goldszal et al, 1995), external section contour (Cohen et al, 1998), principle axes transformation (Hess et al, 1998;Hibbard et al, 1987), consistent matrix transformation (Andreasen et al, 1992;Hess et al, 1998), or multi-modal warping based on mutual information metric as a mapping cost function (Kim et al, 1997). We chose a non-warping geometric model restricted to rigid-body transformations that operated on an automated intensity-based registration algorithm (Thevenaz et al, 1998).…”

B-spline snakes: a flexible tool for parametric contour detection

“…In this context, note also that elastic deformation transformations have been studied as a quantitative description of the callosal shape with respect to the Talairach atlas [15]. We would also like to mention the work performed in [12], that uses B-spline for contour description in the context of an identification and registration procedure of neuroanatomical atlases.…”

“…The argument is essentially the same for more general curves in the plane, which are described using two splines instead of one. Specifically, we represent a general B-spline snake as follows: (12) where and are the and spline components, respectively; these are both parameterized by the curvilinear variable . The exact value of , which marks the end of the curve, is dictated by the desired resolution of the final discrete curve; by convention, we do only render the curve points for integer.…”

Piecewise affine registration of biological images for volume reconstruction

“…Hibbard and Hawkins (1988) for instance use principal axes to align digital autoradiograms. Better performances in terms of precision can be achieved by matching contours (Cohen et al, 1998;Zhao et al, 1993), edges (Kay et al, 1996;Kim et al, 1995) or points (Rangarajan et al, 1997). While it could be argued that these techniques enable a better control over the registration process, the segmentation of feature elements can sometimes be as difficult a problem as the overall registration process itself, with similar drawbacks: non-reproducibility if the feature are extracted manually, lack of precision for fully automated segmentation, etc.…”