NeuroSeg-III: efficient neuron segmentation in two-photon Ca²⁺ imaging data using self-supervised learning.

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Imaging

Background:

• Two-photon calcium (Ca2+) imaging is crucial for neuroscience.
• Neuron segmentation models require extensive manual annotation, hindering performance.
• Current methods face challenges with speed and accuracy in neuron segmentation.

Purpose of the Study:

• To develop an efficient and accurate neuron segmentation approach for two-photon Ca2+ imaging.
• To reduce the reliance on extensive manual annotation in neuroscience research.
• To introduce NeuroSeg-III, a self-supervised learning method for neuron segmentation.

Main Methods:

• Implemented a self-supervised pre-training network and a segmentation network.
• Utilized YOLOv8s, FasterNet, efficient multi-scale attention (EMA), and Bi-directional Feature Pyramid Network (BiFPN).
• Employed a self-supervised learning strategy for encoder pre-training, followed by fine-tuning with minimal annotated data.

Main Results:

• Achieved fast and precise neuron segmentation in imaging data.
• Demonstrated generalization across different Ca2+ indicators and imaging scales.
• Outperformed state-of-the-art benchmarks in speed and accuracy on a public dataset.

Conclusions:

• NeuroSeg-III offers an effective solution for neuron segmentation in two-photon Ca2+ imaging.
• The self-supervised approach significantly reduces annotation requirements.
• The method provides a computationally efficient and highly accurate tool for neuroscience research.