A general deep learning framework for neuron instance segmentation based on Efficient UNet and morphological...

Area of Science:

• Neuroscience
• Computational Biology
• Medical Image Analysis

Background:

• Deep learning excels in medical image analysis, particularly cell instance segmentation.
• Training deep learning models requires extensive, expert-annotated datasets, which are costly and time-consuming to create.

Purpose of the Study:

• To develop an automated framework for detecting and segmenting NeuN stained neuronal cells in histological images.
• To overcome the limitations of traditional segmentation methods by utilizing point annotations.

Main Methods:

• An end-to-end framework using point annotations and binary segmentation to synthesize pixel-level ground truth data.
• Training a U-Net-like neural network with an EfficientNet encoder using the synthetic annotations.
• Investigating and proposing an novel post-processing strategy involving ultimate erosion and dynamic reconstruction for instance segmentation.

Main Results:

• The proposed model demonstrates superior performance compared to recent methods in neuronal cell segmentation.
• The novel post-processing strategy effectively converts probability maps into segmented instances, outperforming classical techniques.
• The framework provides a robust and efficient solution for analyzing neurons in optical microscopic data.

Conclusions:

• The developed framework offers an efficient and accurate method for neuronal cell instance segmentation using minimal annotations.
• This approach facilitates large-scale analysis of neuronal populations, crucial for preclinical studies and neurodegenerative disease research.
• The automated system reduces the labor and expertise required for cell segmentation, accelerating biological discovery.