Fig. 4: Isotropic restoration via Self-Net facilitates morphology reconstruction of densely-packed neurites. | Light: Science & Applications

Fig. 4: Isotropic restoration via Self-Net facilitates morphology reconstruction of densely-packed neurites.

From: Deep self-learning enables fast, high-fidelity isotropic resolution restoration for volumetric fluorescence microscopy

Fig. 4

a XY MIP of an axo-axonic cell (AAC). Scale bar: 50 μm. b YZ MIP of the image volume indicated by the blue box in (a). Raw and Self-Net data are shown for comparison. Scale bar: 10 μm. c Comparison of the axial MIP images of the raw, CARE, and Self-Net data indicated by the blue box in (b). Scale bar: 5 μm. d Intensity profiles along the corresponding lines indicated by the colored arrowhead pairs in (c). e Color-coding isotropy ratios of the images shown in c. f Axial FWHMs and isotropy ratios of the axon fibers in the image data shown in (c). g Automated skeletonization of the image data shown in (c). The yellow arrowheads indicate typical skeletonization errors when using the raw and CARE data. h Automated skeleton extraction accuracy (n = 5 subblocks of 3003 voxels). i Volume rendering of a local dense image block containing four interwoven neurons. Raw and Self-Net+ data were shown for comparison. j Local enlarged views indicated by the blue box in i. Intensity profiles of the colored dashed line are shown in the bottom right. Scale bar: 5 μm. k Average axial FWHMs and signal-to-background ratios (SBRs) for the raw data and Self-Net and Self-Net+ output (n = 5 measurements). l Erroneously-traced fiber length for each neuron in the raw and Self-Net+ data (n = 3 annotators). The gold-standard tracing results were derived from the reconstruction consensus among three neuroanatomical experts, who independently performed tracing through slice-by-slice editing in the HR lateral views of the raw data. m, n Two types of typical reconstruction errors occur in using the raw data. The dark cyan dashed lines indicate the tracing results of the same annotator using the raw and Self-Net+ data more than 1 week apart. The tracing results using the Self-Net+ data were consistent with the gold-standard tracing results. Top left: enlarged views of the regions indicated by the pink arrowheads. Scale bar: 10 μm. Experiments were repeated with three other dense image stacks, achieving similar results

Back to article page