A deep learning pipeline for accurate and automated restoration, segmentation, and quantification of dendritic spines

Sergio Bernal-Garcia, Alexa P. Schlotter, Daniela B. Pereira, Aleksandra J. Recupero, Franck Polleux, Luke A. Hammond,
A deep learning pipeline for accurate and automated restoration, segmentation, and quantification of dendritic spines,
Cell Reports Methods,
Volume 5, Issue 10,
2025,
101179,
ISSN 2667-2375,
https://doi.org/10.1016/j.crmeth.2025.101179.
(https://www.sciencedirect.com/science/article/pii/S2667237525002152)
Abstract: Summary
Quantification of dendritic spines is essential for studying synaptic connectivity, yet most current approaches require manual adjustments or the combination of multiple software tools for optimal results. Here, we present restoration enhanced spine and neuron analysis (RESPAN), an open-source pipeline integrating state-of-the-art deep learning for image restoration, segmentation, and analysis in an easily deployable, user-friendly interface. Leveraging content-aware restoration to enhance signal, contrast, and isotropic resolution further enhances RESPAN’s robust detection of spines, dendritic branches, and soma across a wide variety of samples, including challenging datasets with limited signal, such as rapid volumetric imaging and in vivo two-photon microscopy. Extensive validation against expert annotations and comparison with other software demonstrate RESPAN’s superior accuracy and reproducibility across multiple imaging modalities. RESPAN offers significant improvements in usability over currently available approaches, streamlining and democratizing access to a combination of advanced capabilities through an accessible resource for the neuroscience community.
Keywords: deep learning; dendritic spines; neuronal morphology; image analysis; content-aware restoration; image segmentation; synaptic connectivity; fluorescence microscopy; open-source software; two-photon microscopy