Deep learning enables fast, gentle STED microscopy.

Area of Science:

• Biophysics
• Cell Biology
• Microscopy

Background:

• Stimulated Emission Depletion (STED) microscopy provides super-resolution imaging of subcellular structures.
• Photobleaching and photodamage limit imaging duration and sample integrity in STED microscopy.
• Current methods often require long pixel dwell times, exacerbating these issues.

Purpose of the Study:

• To develop a deep learning-based method for restoring STED microscopy images.
• To investigate if this method can mitigate photobleaching and photodamage.
• To enable efficient, long-term imaging of dynamic cellular processes.

Main Methods:

• A deep learning model was trained to restore STED images.
• The model was applied to 2D and 3D STED datasets with multiple targets.
• Pixel dwell time was significantly reduced (by one to two orders of magnitude) during image acquisition.

Main Results:

• Deep learning restoration effectively recovered image quality despite reduced pixel dwell times.
• The method mitigated photobleaching and photodamage, allowing for extended imaging.
• Robust restoration was achieved for noisy 2D and 3D STED images.
• The approach facilitated long-term imaging of mitochondrial dynamics.

Conclusions:

• Deep learning-based image restoration is a powerful tool for STED microscopy.
• This technique significantly reduces photobleaching and photodamage, expanding imaging capabilities.
• The method enables efficient, high-quality, long-term super-resolution imaging of cellular dynamics.