A high-throughput framework to detect synapses in electron microscopy images.

Area of Science:

• Neuroscience
• Computational Biology
• Biophysics

Background:

• Synaptic connections are crucial for learning and memory.
• Quantifying synaptic density and strength is vital for understanding brain plasticity and disease.
• Existing methods for synapse quantification are often low-throughput, error-prone, or lack resolution.

Purpose of the Study:

• To develop a high-throughput, automated method for quantifying synaptic changes.
• To overcome limitations of existing techniques in measuring synapse density and strength.
• To provide a scalable and accurate tool for neuroscience research.

Main Methods:

• Utilized a selective synapse staining technique with electron microscopy.
• Developed a machine-learning framework for automated synapse detection in images.
• Employed cross-validation with manual labeling and comparison with existing algorithms for validation.
• Incorporated a semi-supervised algorithm to handle data heterogeneity.

Main Results:

• Successfully detected thousands of synapses in mouse somatosensory cortex images.
• Demonstrated high accuracy through rigorous validation against manual and algorithmic benchmarks.
• Developed efficient and scalable algorithms for synapse detection.
• Made the developed algorithms freely available to the research community.

Conclusions:

• The developed machine learning framework enables high-throughput, accurate synapse quantification.
• This approach overcomes previous limitations in measuring synaptic plasticity and connectivity.
• The tool facilitates large-scale analysis of brain structure and function.