A Machine Learning Workflow of Multiplexed Immunofluorescence Images to Interrogate Activator and Tolerogenic Profiles of Conventional Type 1 Dendritic Cells Infiltrating Melanomas of Disease-Free and Metastatic Patients

Table 3

Optimal parameters settings by selecting the models with the highest F1-Score.

Model	Patch size	Epoch	Batch size	Augmentation	Learning rate	Training time
U-Net3-class	1024 × 1024	500	2	5	0.01	2 h 50 min
Stardist	1024 × 1024	250	2	5	0.0003	1 h 44 min
SplineDist	512 × 512	200	4	5	0.0003	32 h 36 min
Cellpose	1024 × 1024	20,000	2	5	0.0002	2 h 37 min