Fig. 1: Overview of D-LMBmap.
From: D-LMBmap: a fully automated deep-learning pipeline for whole-brain profiling of neural circuitry
a, The pipeline of LSFM whole-brain imaging, which includes tissue clearing, axon staining and microscope imaging. The brain can be imaged in both autofluorescence (488 nm) and stained-specific (647 nm) channels. b, Whole-brain axon segmentation. An axon segmentation DNN is trained to segment axons in thousands of 3D cubes from the LSFM brain and then the whole-brain axons can be reconstructed by a combination of segmented cubes. c, Brain-style transfer. For the LSFM brain imaged in the autofluorescence channel (left), a style-transfer DNN is trained to learn the image style of a brain atlas, where each LSFM image slice can be transferred in atlas style (for example, Allen atlas), as well as preserving their original structures (right). d, Brain region segmentation. Major brain regions can be automatically segmented by a DNN, using either the original or style-transferred brains. e, Whole-brain 3D registration and projection mapping. The original LSFM brain, major brain regions and style-transferred brain are set as the input of the DNN for the whole-brain registration with the brain atlas. After that the registered brain is combined with the whole-brain axons to achieve the projection mapping and axon quantification. f, The software interface of D-LMBmap for whole-brain projection mapping and visualization.
