Figure 2.

Training and inference workflow for the segmentation of cell organelles without the use of structural FL using the channel-wise approach (top) and multi-channel approach (bottom). (I) Training: (a) Training set of multi-modal fluorescent images (three channels represented as red blue and green), (b) Training set annotations of the organelles segmentations, (c) Out-of-the-box pre-trained Cellpose model (Vanilla Cellpose), and (d) Finetuned model trained for each of the individual channels (channel-wise) or trained with a subset of the channels (multi-channel). (II) Inference: (a) Multi-modal fluorescent image (three channels), (b) Models selected from the model zoo corresponding to the image’s cell line and FL channel combination, (c) Spatial flows and probability maps output by the finetuned models for each of the channels, (d) Channel-wise averaging of the maps, and (e) Integration into the segmentation labels.