Figure 1

Applying steerable filters to a DiI image of the developing mouse optic chiasm. (A) A 150 μm-thick horizontal section of an E13.5 mouse brain into whose left eye DiI had been injected. An image of the chiasm taken with fluorescence was superimposed, revealing the DiI labelled axons. A grid whose baseline ran between the centers of each eye was used to align images from multiple embryos. The red shaded rectangle highlights the region that was selected for statistical comparison in sets of mice of different genotypes. Scale bar = 200 μm. (B) Magnification of the DiI labelled chiasm from A. Scale bar = 200 μm. (C) 2nd-derivative Gaussian filters of different orientations used to analyze images: among these examples, the top filter would respond best along the axon in D. Scale bar = 1 μm. (D) High magnification of a segment of axon from B. Arrows, which are associated with each pixel along the centre of the axon and point away from the injected eye, are aligned with the filter orientations that gave the best responses. Scale bar = 2 μm. (E) The result of applying the full algorithm to the image in B: red arrows represent direction, green arrows represent curvature. Only a small subset of arrows has been plotted, for clarity. Scale bar = 200 μm. (F) Examples of two arrows at different positions on a single axon (surrounding axons removed for clarity) showing how vector length is proportional to curvature, with the direction of the arrow indicating the direction of curvature. Scale bar = 10 μm.