Single Molecule Localization Super-resolution Dataset for Deep Learning with Paired Low-resolution Images.

Area of Science:

• Biophysics
• Cell Biology
• Microscopy

Background:

• Deep learning has rapidly advanced super-resolution microscopy.
• Single molecule localization microscopy (SMLM) provides high-quality data but public datasets are scarce, limiting deep learning development.

Purpose of the Study:

• To introduce DL-SMLM, a novel biological image dataset for training super-resolution microscopy deep learning models.
• To provide paired low-resolution and super-resolution SMLM data for diverse subcellular structures.

Main Methods:

• The DL-SMLM dataset contains raw SMLM data and corresponding low-resolution images for six subcellular structures.
• Data segmentation allows generation of thousands of training pairs.
• Imaging system performance was validated using DNA origami samples.

Main Results:

• DL-SMLM includes 188 sets of raw SMLM data with 100 signal levels per low-resolution image.
• Demonstrated effectiveness of DL-SMLM by training super-resolution models.
• The dataset facilitates the development of deep learning-based super-resolution microscopy.

Conclusions:

• DL-SMLM effectively addresses the scarcity of public datasets for deep learning in super-resolution microscopy.
• The dataset supports the advancement of deep learning super-resolution techniques through accessible training data.