Abstract:The second near-infrared (NIR-II) fluorescence imaging has attracted widespread attention from researchers due to its powerful advantage of assisting doctors in real-time tumor removal during surgery. Light has different penetration capabilities in different bands. Although NIR-IIb window (1,500-1,700 nm) imaging is clearer than NIR-IIa window (1,000-1,300 nm), there are no effective molecular probes for NIR-IIb clinical transformation. NIR-II images are rich in variety, including animals, tissues, etc., and the huge differences in structure also bring difficulties to image enhancement. In order to solve the above problems, we propose a universal NIR-II fluorescence image enhancement model based on arbitrary style transfer, and simultaneously achieve the quality enhancement of multiple types of NIR-IIa fluorescence images. The goal of this model design is to completely preserve the important information of the original image to prevent artifacts and meet the high demand for real-time imaging during NIR-II surgery. To this end, we design a fusion network based on the calculation method of square-square root-sign assignment, in which the content information of NIR-IIa is integrated into the style representation of NIR-IIb to increase the maintenance of the original content structure, so that the stylized feature map tends to show the content. In addition, we propose an attention matching loss, which extracts key structural information through the attention mechanism and maintains the structure of the enhanced image consistent with the original image. Extensive experiments demonstrate that our proposed model has a great advantage in maintaining the original content information, with the SSIM of 0.625 and the content loss of 0.279 compared to the original images. In addition, the quality of enhanced images is satisfactory, with the PSNR of 31.20 and the SSIM of 0.501 compared to the real images.

An accurate and universal approach for short-exposure-time microscopy image enhancement

Microscopic image detail enhancement based on local window and extremum

Deep Learning Enhanced Fluorescence Emission Difference Microscopy

Untrained Neural Network Enabling Fast and Universal Structured-Illumination Microscopy

Image Enhancement For Fluorescence Microscopy Based On Deep Learning With Prior Knowledge Of Aberration

Enhanced Stimulated Raman and Fluorescence Imaging by Single-Frame Trained BDN

Universal and High-Fidelity Resolution Extending for Fluorescence Microscopy Using a Single-Training Physics-Informed Sparse Neural Network

Deep Learning Enhanced Electrochemiluminescence Microscopy.

Fluorescence Microscopic Image Enhancement Method Based on Multi-saliency Guided Filtering Fusion

Enhancement of the Resolution for Low-Quality Fluorescence Microscopy Image by Novel Computational Method

Label-free microscopic cell images adaptive enhancement via weighted fusion of bright, dark, and weak structure features

Universal NIR-II fluorescence image enhancement via square and square root network

Bio-friendly long-term subcellular dynamic recording by self-supervised image enhancement microscopy

EEMEFN: Low-Light Image Enhancement Via Edge-Enhanced Multi-Exposure Fusion Network

Zero-shot learning enables instant denoising and super-resolution in optical fluorescence microscopy

Exposure Difference Network for Low-Light Image Enhancement

Fast and Lightweight Network for Single Frame Structured Illumination Microscopy Super-Resolution

Cross-system biological image quality enhancement based on the generative adversarial network as a foundation for establishing a multi-institute microscopy cooperative network

A Deep-Learning-based Workflow to Deal with the Defocusing Problem in High-Throughput Experiments

CIEGAN: A Deep Learning Tool for Cell Image Enhancement

Deep learning-based aberration compensation improves contrast and resolution in fluorescence microscopy