Atlas of a Rhesus Lateral Geniculate Nucleus (LGN)

Projection columns representing directions in visual space: red, contra-eye parvo; yellow, ipsi-eye parvo; green, ipsi-eye magno; blue, contra-eye magno. Columns are retinotopically located at intersections of the light grey background polar grid (more clearly visible on the high-resolution image): inclinations every 15°; eccentricities spanning roughly equal distances in the nucleus (0, 1.0, 1.9, 3.4, 6.3, 11.3, 19.2, 30.7, 46.6, and 67.6 degrees).

The heart-shaped central figure intersecting the oval blind spots separates regions of visual space represented by six (central) and four (peripheral) geniculate layers. Cut faces of semitransparent views pass though the blind spot: upper panels show the 17° isoeccentricity surface; lower panels show the minus 7° isoinclination surface. The posterior pole (foveal representation) is in the uppermost part of each panel; the lower-left panel shows the ventral surface. Projection columns are retouched to improve clarity.

Made with the assistance of Janet Sinn-Hanlon of the Visualization, Media, and Imaging Laboratory of the Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign.

This atlas was created by Ed Erwin, Frank Baker, William Busen, and Joseph Malpeli for the study described in Erwin et al., 1999, Relationship between laminar topology and retinotopy in the rhesus lateral geniculate nucleus: results from a functional atlas, Journal of Comparative Neurology, 407: 92-102, 1999. with support from a grant from the National Eye Institute (NIH RO1 EY02695). That paper describes the limitations of the atlas, and the user would be wise to take note of these. The atlas provides only data files; software for analysis or visualization must be obtained by the user. Three-dimensional rendering of such large files requires fairly powerful computing resources, but one can still obtain insightful views of the nucleus if the linear resolution of the arrays is reduced by a factor of 2 or 3, thus reducing the size of the data set by a factor of 8 or 27. The three-dimensional renderings presented in Erwin et al. (1999) were generated with the Analyze® Biomedical Imaging Application Package (Mayo Foundation); we have also viewed two-dimensional cuts through the LGN with Matlab.

The following text describes the format of the data files containing the atlas.

Good luck - Ed Erwin, Frank Baker, Bill Busen & Joe Malpeli

The first four data arrays described below have dimensionality 240 (medial-lateral) x 280 (dorsal-ventral) x 320 (anterior-posterior), with each voxel representing a 25 x 25 x 25 micron volume. The first 240 entries represent a medial-lateral line of voxels, the next 240 entries represent an adjacent medial-lateral line of voxels in the same coronal plane, and so on (280 times) until that coronal plane is complete. Then this sequence is repeated for the next coronal plane, and so on (320 times) until the volume is complete. The Horsley-Clarke position of the origin of this coordinate system is 8.5 mm lateral, -1.5 mm dorsal, and 3.5 mm anterior, with increasing array indices corresponding to increasing distance in the lateral, dorsal and anterior directions. These arrays are supplied in binary integer files (little-endian format - low byte is least significant) without headers.

Note that visual space was mapped in spherical polar coordinates, the coordinate system E, Figure 2 of Bishop et al., 1962 (J. Physiol. 163: 466-502). Retinotopy is mapped continuously across the LGN, spanning interlaminar spaces and optic-disk gaps. It often extends slightly beyond the outer borders of the LGN, to an extent that occasionally differs for ECC.DAT and INCL.DAT.

The number of cells per voxel as functions of Horsley-Clark anterior position is given below. These are equations of the curves shown in Figure 2 of Malpeli et al. (1996, J. Comp. Neurol. 375: 363-377), except that the factor of 0.000625 (0.025²) has been incorporated because their figure gave cells/mm² for a 25 micron-thick slab, instead of cells/voxel. 

     magno:   cells / voxel =  0.000625 (ax^3 + bx^2 + cx + d), where
                    a = -3.6220071 
                    b =  95.829137
                    c = -852.45404
                    d =  2732.8129
     parvo:    cells / voxel =  0.000625 (ax^6 + bx^5 + cx^4 + dx^3 + ex^2 + fx + g), where  
                    a =  0.2893099
                    b = -14.997893
                    c =  317.08614
                    d = -3499.3549
                    e =  21249.945
                    f = -67304.569
                    g =  87499.769

Last modified: January 30, 2008