Virtual reality empowered deep learning analysis of brain activity

Abstract

Tissue clearing and fluorescent microscopy are powerful tools for unbiased organ-scale protein expression studies. Critical for interpreting expression patterns of large imaged volumes are reliable quantification methods. Here, we present DELiVR a deep learning pipeline that uses virtual reality (VR)-generated training data to train deep neural networks, and quantify c-Fos as marker for neuronal activity in cleared mouse brains and map its expression at cellular resolution. VR annotation significantly accelerated the speed of generating training data compared to conventional 2D slice based annotation. DELiVR detects cells with much higher precision than current threshold-based pipelines, and provides an extensive toolbox for data visualization, inspection and comparison. We applied DELiVR to profile cancer-related mouse brain activity, and discovered a novel activation pattern that distinguishes between weight-stable cancer and cancer-associated weight loss. Thus, DELiVR provides a robust mouse brain analysis pipeline at cellular scale that can be used to study brain activity patterns in health and disease. The DELiVR software, Fiji plugin and documentation can be found at https://www.DISCOtechnologies.org/DELiVR/. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=169 SRC="FIGDIR/small/540970v1_ufig1.gif" ALT="Figure 1"> View larger version (66K): org.highwire.dtl.DTLVardef@172d5bforg.highwire.dtl.DTLVardef@2f1d80org.highwire.dtl.DTLVardef@139e7a0org.highwire.dtl.DTLVardef@95dce1_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LIDELiVR detects labelled cells in cleared brains with deep learning C_LIO_LIDELiVR is trained by annotating ground-truth data in virtual reality (VR) C_LIO_LIDELiVR is launched via a FIJI plugin anywhere from PCs to clusters C_LIO_LIUsing DELiVR, we found new brain activity patterns in weight-stable vs. cachectic cancer C_LI Supplementary Videos can be seen at: https://www.DISCOtechnologies.org/DELiVR/