The fluorescence images were deconvolved with a constrained, iterative approach originally designed for UNIX systems (Carrington et al., 1995). The algorithm was rewritten using FFTW, a free, fast-Fourier transform library, and implemented as a multiuser client/server system on computers running the Fedora operating system (Red Hat), either stand alone or configured in a Beowulf cluster. Each image was dark current and background subtracted, flat-field corrected, and then deconvolved. After deconvolution, images were thresholded to eliminate nonspecific binding. Voxels that fell below a threshold were considered to be nonspecific bindings and were set to zero; all other voxels remained unchanged. This threshold was derived from analysis of control images containing no primary antibody and only secondary label. The intensity that eliminates 99% of voxels in the control images became the threshold intensity.
