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

Scanning ion conductance microscopy (SICM) has emerged as an important technique in the biomedical field due to its non-contact, nanoscale imaging capabilities, but the slow imaging speed limits its ability to capture dynamic biological processes. Undersampling combined with computational reconstruction presents a promising paradigm to accelerate imaging. However, existing compressed sensing (CS) and deep learning methods either yield suboptimal reconstruction quality or require abundant training data that are difficult to acquire for SICM. To address these challenges, we propose a zero-shot super-resolution (SR) framework that uses artificial neural networks to reconstruct high-fidelity images from undersampled SICM measurements. By exploiting internal image statistics, the method extracts training samples solely from the input itself to train an image-specific SR network, eliminating the reliance on external datasets. Comparative experiments demonstrate that the proposed method achieves superior reconstruction accuracy compared to bicubic interpolation, CS, and the baseline zero-shot SR (ZSSR) algorithms, while exhibiting robust performance across random initialization. Furthermore, it attains comparable reconstruction quality with significantly fewer sampling points than CS methods, thereby enabling faster SICM imaging. Reconstruction experiments under noisy conditions further demonstrate the effectiveness of the proposed method in practical SICM applications. This work offers a practical strategy for high-speed SICM imaging and suggests a promising pathway for enhancing imaging speed in other data-scarce scanning probe microscopy techniques.