Reconstruction of undersampled scanning ion conductance microscopy images through zero-shot learning-based...

Area of Science:

• Biomedical imaging
• Nanotechnology
• Computational microscopy

Background:

• Scanning Ion Conductance Microscopy (SICM) offers non-contact, nanoscale imaging crucial for biomedical applications.
• Current SICM imaging speeds are too slow for capturing dynamic biological processes.
• Existing accelerated imaging methods like compressed sensing (CS) and deep learning (DL) face challenges with reconstruction quality and data requirements.

Purpose of the Study:

• To develop a novel computational framework for accelerating SICM imaging.
• To enable high-fidelity image reconstruction from undersampled SICM data.
• To overcome the limitations of existing methods in terms of reconstruction quality and training data dependency.

Main Methods:

• Proposed a zero-shot super-resolution (SR) framework utilizing artificial neural networks.
• The framework exploits internal image statistics to train an image-specific SR network, eliminating the need for external datasets.
• Reconstruction of high-fidelity images from undersampled SICM measurements.

Main Results:

• The proposed zero-shot SR method demonstrated superior reconstruction accuracy compared to bicubic interpolation, CS, and baseline zero-shot SR (ZSSR) algorithms.
• Achieved comparable reconstruction quality to CS methods but with significantly fewer sampling points, enabling faster SICM imaging.
• Showcased robust performance across random initializations and effectiveness under noisy conditions.

Conclusions:

• The developed zero-shot SR framework provides a practical strategy for high-speed SICM imaging.
• This approach significantly enhances imaging speed without compromising image quality.
• Offers a promising pathway for accelerating imaging in other data-scarce scanning probe microscopy techniques.