Journal: bioRxiv

Article Title: CochleaNet: deep learning-based image analysis for cochlear connectomics and gene therapy

doi: 10.1101/2025.11.16.688700

Figure Lengend Snippet: a , Schematic overview of the workflow from sample preparation, to imaging, to data pre-processing, to prediction with CochleaNet, and to subsequent data analysis. b-b’’ , Intact mouse cochlea immunolabeled for parvalbumin (PV, red: labels spiral ganglion neuron (SGN) somata and neurites as well as inner hair cells (IHCs)), vesicular glutamate transporter 3 (Vglut3, blue: labels IHCs), and C-terminal Binding Protein 2 (CtBP2, white, staining RIBEYE of the presynaptic IHC ribbons). c-c’’ , CochleaNet segmentation of SGNs based on PV, IHCs based on Vglut3 and ribbon synapse detections based on CtBP2, from three dedicated deep neural networks.

Article Snippet: For ribbon synapses, intensity-based detection of CtBP2 immunofluorescence spots was performed in the Imaris software ( Meyer et al , 2009 ), followed by manual removal of false positive detections.

Techniques: Sample Prep, Imaging, Immunolabeling, Binding Assay, Staining

Journal: bioRxiv

Article Title: CochleaNet: deep learning-based image analysis for cochlear connectomics and gene therapy

doi: 10.1101/2025.11.16.688700

Figure Lengend Snippet: Three intact cochleae from untreated mice imaged with the high-resolution light-sheet microscope and analyzed with CochleaNet. a-c, Visualization of three cochlea with rendering of the unprocessed PV (red), Vglut3 (blue) and CtBP2 (white) signal (top left), the unprocessed signal with an overlay of SGN and IHC segmentation (top right), the processed signal, where off-target signal was removed and PV signal in SGNs as well as Vglut3 and CtBP2 signal in IHCs were increased to highlight the cells that were analyzed (bottom left), and overlay of the processed signal with segmentations (bottom right). Signal removal and intensity modulation was performed in Imaris using the segmentation masks from CochleaNet. Note that the CochleaNet networks were applied to the unprocessed signal; differentiation of signals in the cells of interest and background / off-target signal is one of the main analysis challenges.

Article Snippet: For ribbon synapses, intensity-based detection of CtBP2 immunofluorescence spots was performed in the Imaris software ( Meyer et al , 2009 ), followed by manual removal of false positive detections.

Techniques: Microscopy

Journal: bioRxiv

Article Title: CochleaNet: deep learning-based image analysis for cochlear connectomics and gene therapy

doi: 10.1101/2025.11.16.688700

Figure Lengend Snippet: a , Gerbil cochlea imaged in the high-resolution microscope with PV (red), Vglut3 (blue) and CtBP2 (white) staining; right panels show the respective signals at higher magnification. b , SGN and IHC segmentation as well as synapse detection (small spheres in zoom-in) from CochleaNet. c , Quantification of gerbil SGN, IHC, and ribbon synapse counts with literature values. d , SGN counts of a gerbil treated with f-Chrimson therapy, for injected (left) and non-injected (right) cochlea, with reference value from the untreated animal. See for the corresponding SGN densities. e , Efficiency of f-Chrimson expression derived from GFP intensities, analyzed across the tonotopic axis.

Article Snippet: For ribbon synapses, intensity-based detection of CtBP2 immunofluorescence spots was performed in the Imaris software ( Meyer et al , 2009 ), followed by manual removal of false positive detections.

Techniques: Microscopy, Staining, Injection, Expressing, Derivative Assay