2D electrophoresis image segmentation within a pixel-based framework

The current paper follows a complete pixel-based approach from aligning to a resulting spot volume table, with several novel steps along the workflow. The workflow encompasses image pre-processing, pixel-based analyses, segmentation where the experimental design is utilised, and final data table reporting. Our approach employs optical flow estimation for alignment, and technical variation is removed prior to the statistical validation. The statistical data analyses are based on ANOVA adjusted by rotation testing and on resampled PLS regression at pixel level. A novel approach for segmentation is performed based on the statistical output performed on pixel level (regression coefficients and their p-values or t-values), and refolded to image representation (meta image). A double threshold is used to distinguish random false positive pixels from structured information of protein spots. Finally, the end user is presented a list of protein spots of interest at spot volume level rather than pixel level where the spot detection procedure has taken into account the relevance in light of the experimental design.