PSSR

PSSR enhances resolution and signal-to-noise ratio of point-scanning imaging data from scanning electron microscopes and laser scanning confocal microscopes using deep learning-based supersampling.

Key Features:

• Deep Learning-Based Supersampling: Employs a deep learning framework to perform supersampling of undersampled point-scanning images to improve spatial resolution.
• Simulated Training Data ('crappifier'): Uses a 'crappifier' algorithm to degrade high-SNR, high-pixel-resolution ground-truth images into simulated low-SNR, low-resolution counterparts for model training.
• Multi-Frame Approach: Integrates information from adjacent frames for fluorescence time-lapse or other sequential data to improve spatiotemporal reconstruction accuracy.

Scientific Applications:

• Cellular and Tissue Imaging: Enhances detail in cellular and tissue images acquired by point-scanning modalities for downstream analysis.
• Time-Lapse Fluorescence Imaging: Improves spatiotemporal resolution in fluorescence time-lapse datasets by leveraging adjacent-frame information.
• Improved SNR and Reduced Acquisition Burden: Enables restoration of low-SNR, undersampled data, supporting faster acquisition and potential sample preservation.

Methodology:

Computational methods explicitly include deep learning-based supersampling trained on pairs of high-quality ground-truth and simulated low-quality images produced by a 'crappifier' degradation algorithm, with an optional multi-frame model that incorporates adjacent frames for prediction.