Volumetric denoising enables high-throughput volume electron microscopy and efficient downstream analysis

Area of Science:

• Microscopy
• Computational Biology
• Biophysics

Background:

• Volume electron microscopy (VEM) provides high-resolution 3D visualization of biological samples.
• Current VEM acquisition speeds and downstream analysis limitations hinder imaging throughput.
• High resolution in VEM can compensate for image noise in preserving 3D structural data.

Purpose of the Study:

• To identify optimal VEM acquisition conditions for increased imaging throughput.
• To evaluate the effectiveness of denoising methods versus axial interpolation for restoring VEM datasets.
• To develop and validate advanced 3D denoising techniques for VEM data.

Main Methods:

• Systematic search for optimal VEM acquisition parameters.
• Comparison of machine learning-based denoising methods, including a novel 3D context-based model.
• Evaluation of methods on simultaneously acquired VEM datasets using various tasks.
• Assessment of serial block-face cutting capabilities down to 20 nm.

Main Results:

• Sufficient spatial resolution effectively counteracts image noise for 3D structural preservation.
• 3D denoising outperforms axial interpolation in restoring volumetric datasets.
• The developed 3D denoising approach achieves superior faithful feature recovery compared to baseline methods.
• Fast imaging enabled by denoising facilitates robust serial block-face cutting.

Conclusions:

• An optimized VEM acquisition strategy can significantly increase imaging throughput.
• Volumetric denoising methods are crucial for enhancing the quality and utility of VEM data.
• This work provides practical guidelines for maximizing VEM throughput and data quality.