The broader application of deep learning to spectral data remains a complex task due to the need for augmentation routines and architectures specific to spectral data. Here we present spectrai, an open-source Python/MATLAB deep learning package designed to facilitate the training of neural networks on spectral data and enable comparison between different methods. Spectrai provides numerous spectral data pre-processing and augmentation routines, as well as neural networks for spectral data including spectral denoising, spectral classification, spectral image segmentation, and spectral image super-resolution and transfer learning. We demonstrate application of spectrai to Raman spectroscopy and hyperspectral imaging across multiple biomedical domains.

In this study, we applied a unique fiberoptic Raman spectroscopy probe to access the post-treatment NPC treatment efficacy follow-up and accurate recurrent NPC diagnosis. Significant Raman feature differences are discovered among normal, NPC, and non-recurring post-treatment patients. By incorporating with partial-least-squares linear-discriminant-analysis (PLS-LDA), the in vivo fingerprint and high-wavenumber tissue Raman spectra provide high diagnostic accuracy for detecting recurrent NPC from both early post-treatment inflammation and long-term post-treatment fibrosis. We further investigate the major biochemicals associated with NPC tissue compared to normal nasopharyngeal tissue through quantitative modeling. In this work, we demonstrate that fiberoptic Raman spectroscopy is an effective diagnostic modality for real-time, label-free post-treatment surveying and recurrent tumor detection in NPC patients.

We develop a disposable sub-millimeter fiberoptic Raman needle probe for real-time in vivo tissue and biofluids Raman measurements. High-quality tissue Raman spectra in fingerprint and high-wavenumber regions can be collected within sub-second from different tissue types and biofluids using the needle Raman probe fabricated together with the structured background subtraction algorithm developed. We further validate the depth-resolved deep tissue Raman spectra measurement capability of the fiberoptic needle Raman probe by advancing the needle Raman probe into a mice brain model. We demonstrate that the sub-millimeter fiberoptic Raman probe developed can achieve real-time time collection of deep tissue and biofluids FP/HW Raman spectra with high signal to noise ratios, suggesting the potential of dual functioning of Raman optical biopsy and fine-needle aspiration biopsy for in vivo deep tissue and biofluids characterization in the human body.

Oropharyngeal squamous cell carcinoma (OPSCC) which refers to the cancer of the back of the throat, including the base of tongue and tonsils, has rapidly increased the past several decades and if undiagnosed, the tumors metastasize leading to many complications and decreased survival. In this study, Raman spectroscopy (RS) in combination with data classification algorithms was used to examine tonsil specimens (normal, benign, and malignant) to determine if RS could serve as a viable tool for real-time sensitive detection of OPSCC and ultimately other HPV-linked cancers.

Vibrational spectroscopies and especially infrared spectroscopy play an increasing in modern biodiagnostics. Applications range from non-invasive exhaled breath analysis to in-vivo assessment of cartilage damage rendering mid-infrared (MIR; 3-20 µm) photonics tools among the most flexible molecular sensing platforms nowadays available. With the emergence of quantum and interband cascade laser technology, the hybrid on-chip integration of entire MIR sensing device es is conceivable leading to IR-lab-on-chip systems. The inherent molecular selectivity of MIR signatures enables studying small (e.g., volatile organic compounds; VOCs) in the gas phase, as well as large biomacromolecules (e.g., proteins) in the liquid phase with unsurpassed detail in a label-free and non-destructive fashion in real-world complex mixtures if clinical relevance. Latest MIR photonic technology will be complemented by highlight applications demonstrating the utility of next-generation MIR photonics.

F-PTIR is described theoretically and demonstrated experimentally for f high spatial resolution chemical microscopy based on mid-infrared absorption spectroscopy. In F-PTIR, heat produced locally by infrared absorption results in reductions in fluorescence through temperature-dependent changes in quantum efficiency. The point-spread function for F-PTIR is identical to that of fluorescence, simplifying the interpretation of image contrast relative to alternative scattering/refraction based photothermal detection methods. Fluorescence labeling enables infrared spectroscopy just in the regions associated with the label, providing dual selectivity for local chemical environments. F-PTIR spectroscopy was used to identify phase-separated domains within pharmaceutical materials impacting bioavailability and formulations design.

Understanding metabolism is indispensable in unraveling the mechanistic basis of biological processes. However, in situ metabolic imaging tools are still lacking. Here we introduce a framework for mid-infrared (MIR) metabolic imaging by coupling the emerging high-information-throughput MIR microscopy with specifically designed IR-active vibrational probes. We present three categories of small vibrational tags including azide bond, 13C-edited carbonyl bond and deuterium-labeled probes to interrogate various metabolic activities in cells, small organisms and mice. Our technique is uniquely suited to metabolic imaging with high information throughput. In particular, we performed single-cell metabolic profiling and large-area metabolic imaging at tissue or organ level.

Surface-enhanced Raman spectroscopy (SERS) is a powerful tool for vibrational spectroscopy, but is compromised by its low reproducibility, uniformity, biocompatibility, and durability. This is because it depends on “hot spots” for high signal enhancement. Here we report our experimental demonstration of a plasmon-free nanostructure composed of a two-dimensional array of porous carbon nanowires as a SERS substrate for highly sensitive, biocompatible, and reproducible SERS. Specifically, the substrate provides not only high signal enhancement, but also high reproducibility and fluorescence quenching capability. We experimentally demonstrated these excellent properties with various molecules such as rhodamine 6G (R6G), β-lactoglobulin, and glucose.

Raman optical activity (ROA) is a powerful tool for identifying the absolute conformational information and behavior of chiral molecules in aqueous solutions, but suffers from low sensitivity. Here we report our development of a silicon nanodisk array that tailors a chiral field to significantly increase the interaction between the excitation light and chiral molecules via exploiting a dark mode. Specifically, we used the array with pairs of chemical and biological enantiomers to show >100x enhanced chiral light-molecule interaction with negligible artifacts for ROA measurements. Our silicon nanodisk array opens a cost-effective way for conformational analysis of trace chiral molecules.

We present a shot noise limited, three-color SRS implementation to address two molecular vibrations simultaneously. The system allows fast, high quality stimulated Raman histology as well as background-free SRS imaging. It is ready to be tested in hospitals for its viability and image quality in comparison to classical rapid histology. It is based on a mode-locked fs-laser from which 2 narrow-band Stokes laser beams are extracted and subsequently modulated at 13 and 20MHz. The center part of the fs-laser is frequency doubled to pump a picosecond optical parametric oscillator, which can be tuned from 500 to 5000 cm−1.

We develop a high speed compressive Raman imaging technology using a programmable binary spectral filter and a single channel detector to perform fast Raman detection and concentration estimation of know species over millimeter field of views with a pixel dwell time of few tens of µs only. The technology is x100 times faster than commercial CCD based systems and x10 times faster than the EMCCD based systems. We demonstrate rapid imaging of pharmaceutical tablets and microplastics.

We use D2O probed stimulated Raman scattering (DO-SRS) and Multiphoton Fluorescence (MPF) microscopy to visualize metabolic changes in HeLa cells under excess AAA of phenylalanine or tryptophan. The cellular spatial distribution of de novo lipogenesis, protein synthesis, NADH, Flavin, unsaturated lipids, and cholesterol were all imaged and quantified in this experiment. Our studies reveal the increase in NADH to Flavin ratio by 10% and unsaturated lipids to saturated by 50% in cells treated with excess phenylalanine and tryptophan. Our study shows that DO-SRS can be used to as a high resolution imaging platform to study AAA regulated metabolic activities in cells.

Mastitis is a common disease in dairy cows and is considered to be one of the most intractable diseases in the world. We analyzed Raman spectra of milk samples from cows that have been treated for mastitis and have recovered, and those from cows that have no history of mastitis. The result shows that signal components derived from carotenoids and unsaturated fatty acids can be used as markers to predict mastitis history. Our new method based on simple and direct milk measurements will be a powerful tool to determine the prevalence and severity of mastitis in future field diagnosis applications.

Therapeutic drug monitoring (TDM) is required for an optimal treatment plan to control the dosage of high-risk drugs by monitoring their blood concentration. Currently, a combination of liquid chromatography and tandem mass spectrometry is used for TDM. However, this method requires expertise and skills in sample pretreatment and is not available in all hospitals. Raman spectroscopy allows us to quantify substances in biological samples with only simple pretreatment. Two methods have been reported for detecting low-concentration substances using Raman spectroscopy: surface-enhanced Raman spectroscopy and drop coating deposition Raman spectroscopy. However, it is difficult to quantify the concentration by these methods because the Raman spectra are measured in a dry state. Here, we present a new method to quantify low-concentration pharmaceutical analytes using droplet evaporation Raman spectroscopy. Methotrexate (MTX), one of the immunosuppressive drugs, is reported to cause adverse effects above 10 μM of its blood concentration 24 hours after administration. To quantify low-concentration MTX, we drop the solution onto a superhydrophobic substrate, and simultaneously measure the Raman spectra and the volume of the droplets before completely dried. In particular, we control the evaporation rate of the droplets in a humidified environment, allowing Raman measurements with sufficient exposure time. The initial concentration of the solution is determined from the measured drug concentration during evaporation and the concentration ratio obtained from the volume measurement. Using the new method, we can quantify the concentration of MTX at 50 μM, which is in the order of magnitude required for clinical use.