Image contamination correction

Some images may have obvious local contaminations (left picture). If not handled properly such contamination can affect downstream normalization and expression level comparison. Assuming the contamination is “additive” on the true signals and bahave like a semi-transparent layer (more discussion in Schadt et al. 2000), we implemented the image gradient correction algorithm in dChip. In “CEL Image” view (check “Use unnormalized data” when “Open Group”, uncheck “Image/Unscrambled”), one can use right-click (left-click to cancel, double-right-click to end) to outline a contaminated image region (left picture):

then select “Image/Gradient Correction” to adjust the background brightness of this region to a similar level as the background of the surrounding region. The background of a CEL is defined as the median of the CEL values in the 7*7 square centering around this CEL, and is calculated for CELs in the outlined region as well as the surrounding region extending out by 7 CELs (middle picture). Then a CEL value in the outlined contaminated region is adjusted by the difference between its background and the median background of the extended surrounding region (right picture).

When the contamination totally disguises the real signals (left picture):

we may use “Image/Replace Value” to replace each CEL value in the outlined region by the median value of this CEL in all other arrays (right picture).

Select “Undo Last” to reverse the last “Gradient Correction” or “Replace Value” operation, and “Save DCP file” to store the corrected image back into DCP files. Re-normalizing arrays (check “ignore the normalized data” in the “Analysis/Normalize” dialog) is needed, especially after the baseline array is corrected.