Deep-learning-assisted Fourier transform imaging spectroscopy for hyperspectral fluorescence imaging

Abstract

Hyperspectral fluorescence imaging is widely used when multiple fluorescent probes with close emission peaks are required. In particular, Fourier transform imaging spectroscopy (FTIS) provides unrivaled spectral resolution; however, the imaging throughput is very low due to the amount of interferogram sampling required. In this work, we apply deep learning to FTIS and show that the interferogram sampling can be drastically reduced by an order of magnitude without noticeable degradation in the image quality. For the demonstration, we use bovine pulmonary artery endothelial cells stained with three fluorescent dyes and 10 types of fluorescent beads with close emission peaks. Further, we show that the deep learning approach is more robust to the translation stage error and environmental vibrations. Thereby, the He-Ne correction, which is typically required for FTIS, can be bypassed, thus reducing the cost, size, and complexity of the FTIS system. Finally, we construct neural network models using Hyperband, an automatic hyperparameter selection algorithm, and compare the performance with our manually-optimized model.

Refractive index dispersion measurement in the short-wave infrared range using synthetic phase microscopy

“…With large penetration depth, SWIR imaging becomes popular for biomedical imaging. [38][39][40] The refractive index dispersion of many mounting media is not yet available in the SWIR range, to the best of our knowledge. Here we have measured the refractive index dispersion of three microscopy mounting media (glycerol, FluorSave, and Eukitt) in the 1100-1650 nm wavelength range, using the microspheres of known refractive index dispersion.…”

Photoluminescence Probes in Data-Enabled Sensing

“…In recent work, ANNs and convolutional neural networks for deep learning are being used for pattern recognition (135)(136)(137). In fact, artificial intelligence technologies are not just being used to detect specific features, but they are being used to correct (138,139) hyperspectral images (140) for measurement limitations. Hagen et al (138) used their database of low-and high-exposure images of bovine pulmonary artery endothelial cells stained with probes for mitochondria and actin and liver cancer cells stained using a membrane probe to compare the performance of a content aware (CARE) neural network that compared the noisy and high-quality images to those of a neural network and standard block matching method that only process the noisy images.…”