Super‐Resolution Near‐Infrared Fluorescence Microscopy of Single‐Walled Carbon Nanotubes Using Deep Learning

Abstract

Single‐walled carbon nanotubes (SWCNTs) have unique optical and physical properties, with numerous biomedical imaging and sensing applications, owing to their near‐infrared (nIR) fluorescence which overlaps with the biological transparency window. However, their longer emission wavelengths compared to emitters in the visible range result in a lower resolution due to the diffraction limit. Moreover, the elongated high‐aspect‐ratio structure of SWCNTs poses an additional challenge on super‐resolution techniques that assume point emitters. Utilizing the advantages of deep learning and convolutional neural networks, along with the super‐resolution radial fluctuation (SRRF) algorithm for network training, a fast, parameter‐free, computational method is offered for enhancing the spatial resolution of nIR fluorescence images of SWCNTs. An average improvement of 22% in the resolution and 47% in signal‐to‐noise ratio (SNR) compared to the original images is shown, whereas SRRF leads to only 24% SNR improvement. The approach is demonstrated for a variety of SWCNT densities and length distributions, and a wide range of imaging conditions with challenging SNRs, including real‐time videos, without compromising the temporal resolution. The results open the path for accelerated and accessible super‐resolution of nIR fluorescent SWCNTs images, further advancing their applicability as nanoscale optical probes.

Keywords

• convolutional neural networks
• deep learning
• fluorescent nanoparticles
• near-infrared imaging
• single-walled carbon nanotubes
• super-resolution