A Deep Learning Approach for Neuronal Cell Body Segmentation in Neurons Expressing GCaMP Using a Swin Transformer

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Neuronal cell body analysis is vital for understanding brain function and disease.
• Current manual segmentation methods are time-consuming, subjective, and prone to errors.
• Distinguishing neuronal cell bodies from axons and dendrites is challenging.

Purpose of the Study:

• To develop an automated, accurate, and efficient deep learning method for neuronal cell body detection and segmentation.
• To overcome the limitations of manual annotation in neuroscience research.
• To provide a tool for studying neuronal soma size changes in various conditions.

Main Methods:

• Development of a deep learning model utilizing a state-of-the-art shifted windows (Swin) transformer.
• Application of the Swin transformer for 2D detection and segmentation of neuronal somas.
• Testing the algorithm on mouse acute brain slices using multiphoton microscopy under varying fluorescence conditions.

Main Results:

• The Swin transformer algorithm achieved high performance with a mean Dice score of 0.91, precision of 0.83, and recall of 0.86.
• The Swin transformer outperformed two convolutional neural networks in detecting neuronal cell boundaries, especially for GCamP6s expressing neurons.
• The algorithm demonstrated robustness across different experimental conditions, including low and high signal fluorescence.

Conclusions:

• The developed Swin transformer-based algorithm provides a fast, reproducible, and unbiased approach for neuronal cell body segmentation.
• This deep learning tool can significantly assist researchers in accurately analyzing fluorescently labeled neuronal somas in brain slices.
• The flexible algorithm facilitates the study of neuronal soma size dynamics in both physiological and pathological states.