Sparse Scanning Electron Microscopy Data Acquisition and Deep Neural Networks for Automated Segmentation in...

Area of Science:

• Microscopy
• Neuroscience
• Image Reconstruction

Background:

• Three-dimensional imaging and large field of view acquisition in electron microscopy face challenges with long acquisition times.
• Minimizing electron radiation dose is crucial for imaging radiosensitive biological samples.

Purpose of the Study:

• To demonstrate a workflow for random sparse scanning in scanning electron microscopy.
• To validate the use of compressive sensing and neural networks for image reconstruction from sparse data.
• To assess the feasibility of deep learning for neuron segmentation in reconstructed images.

Main Methods:

• Implemented a random sparse scanning acquisition protocol on a scanning electron microscope.
• Applied sparse image reconstruction techniques, including compressive sensing and neural networks.
• Utilized deep learning for automated segmentation of neuron structures from reconstructed images.

Main Results:

• Achieved a 2-3 fold reduction in average dwell time per pixel.
• Demonstrated successful neuron structure segmentation from sparsely acquired and reconstructed data.
• Confirmed the practical benefits of random sparse scanning for connectomics applications.

Conclusions:

• Random sparse scanning is a feasible and beneficial technique for scanning electron microscopy.
• This approach significantly reduces acquisition time and electron dose, crucial for large-scale biological imaging.
• The workflow accelerates connectomics research by enabling efficient neuron segmentation.