Figure 1. Details of the automatic 3D dendritic spine segmentation approach.

(A) Workflow of the approach. 1, The starting z-stack (yellow represents the fluorescence intensity). 2, Subtracting the background. 3, Applying DeepD3 to assign prediction values to the pixels of its likelihood to be part of a spine (magenta to white represents lower to higher likelihood). 4, Creating a binary mask from a low prediction value threshold. 5, Applying the mask back to the original z-stack. 6, Extracting the fluorescent intensity from the masked regions. 7, Applying Stardist to perform 2D segmentation of the spines in each frame (different colors represent differently segmented spines). 8, Connecting the 2D segmentation across each frame to create the 3D segmentation. Scale bar: 5 μm.

(B) Example ROIs showing comparison of dendritic spine segmentation results across z-frames for DeepD3 3D segmentation function, DeepD3+Watershed, and DeepD3+Stardist. Scale bar: 500 nm.