Cell identification in whole-brain multiview images of neural activation

TL;DR

This paper introduces a scalable multiview imaging analysis pipeline that enhances, fuses, and accurately identifies brain cells in whole-brain microscopy images, demonstrated on mouse data with high precision.

Contribution

The paper presents a novel multiview semantic deconvolution method combined with hierarchical registration for improved cell identification in large brain images.

Findings

Achieved an F1 score of 0.89 on test data

Effectively fuses multiview images into a single 3D volume

Demonstrated scalability to whole-brain datasets

Abstract

We present a scalable method for brain cell identification in multiview confocal light sheet microscopy images. Our algorithmic pipeline includes a hierarchical registration approach and a novel multiview version of semantic deconvolution that simultaneously enhance visibility of fluorescent cell bodies, equalize their contrast, and fuses adjacent views into a single 3D images on which cell identification is performed with mean shift. We present empirical results on a whole-brain image of an adult Arc-dVenus mouse acquired at 4micron resolution. Based on an annotated test volume containing 3278 cells, our algorithm achieves an $F_1$ measure of 0.89.