Fig. 4
From: A cell-level quality control workflow for high-throughput image analysis
Detection of staining artifacts, a comparison between our cell-level QC and a patch-level QC application recently integrated into common image analysis platforms. a From left to right, four different types of artifacts are displayed with their: top - raw image, middle - single cell QC result, and bottom - patch-level defocus score generated by a deep learning approach (Yang 2018). In the cell QC mask, artifacts are marked in green and good quality cells are labeled in blue. In the results generated by the patch-level approach, the patch outlines denote the predicted defocus level by hue (red for least defocus) and prediction certainty by lightness (increased lightness for increased certainty). b A comparison between the ratio of defocused patches (y-axis) and ARcell (x-axis) for images collected from assay α and β. Each dot represents an image with both the size and intensity proportional to the prediction certainty of defocus level. Patches with defocus level greater than 5 are considered out of focus. Images within the red dashed circle show high defocus patch ratio but low prediction certainty. More comparisons using different defocus level cutoffs can be found in Supplementary Fig. 8 along with labeled examples