visual selection of healthy cells

[outofculture]

• cellpose's 3d algorithm fails with our brightfield data. our XZ and YZ slices are riddled with diffraction that make cell identification extremely difficult
• cellpose's 3d algorithm works well with our fluorescence data, but it's pretty slow (138s to process a 117x1024x1024 stack on an RTX 4070)
• cellpose's 3d annotation is inadequate and tedious. it doesn't merge/copy between layers
• cellpose's cyto3 model matches on unhealthy cells as much as healthy ones
• building a 3d model without cellpose in the mix would require a great deal of training data

[outofculture]

attempting to revive analysis/scripts/builder as an alternate ui for data annotation