A deep learning-enabled toolkit for the 3D segmentation of ventricular cardiomyocytes

Area of Science:

• Cardiovascular Biology
• Bioimaging
• Computational Biology

Background:

• Accurate 3D segmentation of cardiomyocytes is crucial for understanding cardiac physiology and disease.
• Existing methods for 3D cardiomyocyte analysis are limited by experimental and analytical challenges, often relying on biased 2D inferences.
• A lack of open-source tools and datasets hinders high-resolution 3D cardiomyocyte segmentation.

Purpose of the Study:

• To present a deep learning-enabled toolkit for segmenting individual cardiomyocytes in 3D confocal microscopy volumes.
• To provide a comprehensive dataset of 73 annotated 3D volumes across seven species and diverse experimental conditions.
• To offer an image restoration workflow addressing imaging artifacts.

Main Methods:

• Development of a deep learning toolkit for 3D cardiomyocyte segmentation.
• Creation of a diverse dataset including 73 annotated 3D volumes (mouse, human, elephant) under various experimental conditions.
• Implementation of an image restoration workflow for artifact correction.

Main Results:

• The automatic segmentation workflow achieved an adapted Rand error of 0.063 (∼94% voxel-pair agreement).
• The semi-automatic workflow processed challenging datasets at a throughput of 3 cells/min.
• The toolkit and dataset are open-source with a graphical user interface.

Conclusions:

• The toolkit enables researchers to extract quantitative data on cardiomyocyte microstructure from 3D confocal image stacks.
• The open-source dataset and toolkit facilitate large-scale analyses and realistic 3D reconstructions of cardiomyocytes.
• This resource supports research into cardiac function at the cellular level across species and experimental models.