Figure 1.

The architecture of the image analysis pipelines. (a) For training the random forest-based RF-BF-2 model, a training pipeline was used which also generated the helper models RF-FL-1 and RF-BF-1. (b) For training the U-Net3-BF-1 model the training pipeline was used which also generated the helper model U-Net3-FL-1. (c) U-Net type architecture was used for the DL pipeline. (d) The cell detection pipeline using the trained models was implemented in Aparecium MembraneTools module. In this pipeline, Stage 1 corresponds to data pre-processing, Stage 2 to cell segmentation and Stage 3 to data post-processing. In (a), the U-Net3-BF-1 model is used and in (b) the binary mask is corrected with the quality mask.