MinIP(713686035, Turbo)

U-Net is pretty much the go-to model architecture for biomedical image segmentation. Yet stock U-Net has not yet been shown to be particularly effective when applied to images of neurons.

On the other hand, recurrent variants might well be a fruitful avenue of research. For example, there is a 2018 pre-print1 which seems promising. The recurrent aspect allows the model to maintain an internal memory/state about remote areas of a given neuron. Neurons are very, very long compared to other cells. The recurrence can span such distances. In the pre-print recurrent U-Net is applied to retinas.



This is text with a 1 rando foot note

  1. Text must start in column 1. Tigue