Deep learning enables fast, gentle STED microscopy.

Area of Science:

• Microscopy and imaging science
• Cell biology
• Artificial intelligence in science

Background:

• Stimulated Emission Depletion (STED) microscopy achieves super-resolution imaging of subcellular structures.
• Photobleaching and photodamage limit imaging duration and quality in STED microscopy due to prolonged pixel dwell times.
• Restoring noisy STED images often requires compromises between noise reduction and structural integrity.

Approach:

• Developed a deep learning-based denoising method for STED microscopy images.
• The approach significantly reduces pixel dwell time, mitigating photobleaching and photodamage.
• Validated the method on both 2D and 3D STED datasets with multiple targets.

Key Points:

• Deep learning denoising allows for a reduction in pixel dwell time by one to two orders of magnitude.
• This reduction effectively mitigates photobleaching and photodamage, preserving sample integrity.
• The method provides efficient and robust restoration of noisy STED images.
• Enables high-quality, long-term imaging of dynamic biological processes, such as mitochondrial dynamics.

Conclusions:

• Deep learning denoising is a powerful tool for enhancing STED microscopy.
• The technique overcomes key limitations associated with phototoxicity and imaging time.
• Facilitates advanced live-cell imaging studies, particularly for dynamic subcellular structures.