Is the segmentation method used when running the model on cell-based phenotypic screening dataset for comparison scaffold or random?