FIG. 2.

Data analysis pipeline from raw diffraction patterns to a Laue-symmetrized anisotropic diffuse map. Numbers (1)–(6) correspond to the same image pre-processing substeps as mentioned in Sec. II. Following this pipeline, the (a) crystal diffraction and (b) non-crystal background patterns are applied with the user-defined detector mask and a deeper bad pixel removal step based on pixel positions and intensities. The non-crystal background patterns are then scaled with the exposure time and subtracted from crystal diffraction patterns, giving rise to the (c) background subtracted patterns, followed by multiple pixel intensity and position corrections to produce the (d) corrected diffraction patterns. Bragg peaks are predicted in positions and then replaced with median intensities to generate (e) patterns without Bragg peaks, followed by image scaling and the radial profile variance removal method which end up with the final pre-processed diffraction patterns (f). These patterns are merged into a (g) 3D diffraction volume using indexing results and orientations from the goniometer. This 3D volume is then applied with Laue symmetrization to generate the (h) Laue-symmetrized diffraction volume, followed by the isotropic component subtraction step which produces the final (i) Laue-symmetrized anisotropic diffuse map. For improved visualization, panels (g)–(i) were created using more finely sampled diffraction volumes than were used in data quality evaluation and modeling.