DeepSpineNet

Table of Contents

Description

DeepSpineNet is a GUI application designed to reconstruct dendritic shafts and spines from confocal microscopy images semi-automatically. Like most commercial applications, the reconstruction workflow has to be supervised by an expert. Nevertheless, DeepSpineNet has taken a significant step towards the automation of the segmentation task. Commercial tools, currently endorsed by the community, are heading towards automation of the segmentation process. Currently, they follow three main approaches:

  1. Some tools rely on the user to guide the reconstruction process. The reconstructions obtained in this way are accurate but require a great effort by the user.
  2. Others approximate the surface of the dendrite employing simple shapes at the cost of sacrificing accuracy in the reconstruction.
  3. The last group of applications only requires the user to adjust a limited number of parameters. Parameter selection is not always obvious, and multiple attempts are often required to obtain proper segmentation. Furthermore, they do not usually allow the user to correct the obtained segmentations manually.

By contrast, our GUI application first calculates a segmentation fully automatically. Afterward, the user can correct the reconstruction using several semi-automatic techniques, such as denoising, pixel relabeling, separation of overlapping structures, or union of disconnected structures. DeepSpineNet offers the possibility of obtaining precise segmentations minimizing the time spent by the user.

Screenshots

Videos

The videos shown below illustrate the main modules and functionalities available in the tool:

Binaries and Sources

All code and binaries are distributed under a Dual License model, depending on its usage. For its non-commercial use, it is released under an open-source license (MIT). Please contact Marcos Garcia-Lorenzo if you are interested in a commercial license. The code and binaries can be found at the following repository: DeepSpineNet repository.

Development Team

Isabel Vidaurre-Gallart: isabel.vidaurre |at| urjc.es
Nicusor Cosmin-Toader: cosmin.toader |at| urjc.es
Marcos Garcia-Lorenzo: marcos.garcia |at| urjc.es

Acknowledgments

The research leading to these results has received funding from the following entities: The European Union’s Horizon 2020 Research and Innovation Programme under grants agreement no. 785907 (Human Brain Project SGA2) and 945539 (Human Brain Project SGA3), the Spanish Government under grants: TIN2017-83132-C2-1-R (VIANA), PID2020-113013RB-C21(MEDAVI), FPU18/05304, and PRE2018-085403.