Figure 7. Cross-atlas registration and neuronal cell nuclei detection for assignment of automatically detected cells to major brain regions in developing zebrafish.

Developing larval (five dpf) zebrafish brains were divided into 10 quantifiable anatomically relevant regions by registering sample brains to the Vibe-Z volume-based zebrafish brain atlas and referencing to an immunohistochemistry BrdU based neuroanatomy atlas. Sample landmark selection is shown for equivalent slices between the ViBE-Z atlas data and a five dpf micro-CT for registration-based segmentation (A). Overlay of ViBE-Z atlas embryo (red) over five dpf data (blue) post-registration (B). Semi-automated brain segmentation presented for a five dpf sample in its anatomic context in a representative slice (C). Neuronal cell nuclei detection and validation using a supervised random forest classifier (D). Nuclear detection visualized as a point map for a typical five dpf sample in its anatomic context (E). Semi-automated brain segmentation overlaid onto classifier-based nuclei detection (F). Brain density was calculated by counting every nucleus within a ~ 22 micron (30 voxel) radius surrounding each voxel of the brain (G)..This dimension waschosen because it covers about 5 cell diameters, an estimate of the width of a small brain region.

Figure 7—figure supplement 1. Validation of automated detection of neuronal cell nuclei in larval (five dpf) zebrafish.

75 × 75×75 µm regions were selected on each fish to validate detection of nuclei: sample regions R1, R2 and R3 are presented (A). 3D visualizations of thresholded original data (B–a) and nuclear probabilities based on the trained classifier (B–b) in Region two are shown. 3D rendering of validation is presented in (B–c); green denotes true positive, blue false positive, orange false negative. Good agreement between manual and automated segmentation is shown on regional 3D rendering and f1 scores (C) across all specimens.

Figure 7—figure supplement 2. Comparison of object detection between micro-CT (five dpf) and transmission electron microscopic (5.5 dpf) sections.

PTA and osmium staining in (A) micro-CT and (B) transmission electron microscopy images (Hildebrand et al., 2017) both allow the nuclei of individual cells to be distinguished. Automated object detection applied to these images show similar regional patterning and counts of nuclei in corresponding brain slices (C, D).