Figure 1. Representative data acquired by CI microscopy systems including: a transmission HD FT-IR system imaging a breast tissue core shown at the Amide I band; a transflection DFIR direct point scanning system also imaging breast tissue when tuned to Amide I; an IR-OH system imaging the PT effect with WF illumination on the Amide II band; and lastly, an AFM-IR system with CL controls imaging at Amide II the protein distribution of subcellular structures within a single cell.
Kevin Yeh, Seth Kenkel, and Rohit Bhargava[1,2,3]
Beckman Institute for Advanced Science and Technology
Departments of Bioengineering, Mechanical Science and Engineering,
Electrical and Computer Engineering, Chemical and Biomolecular Engineering, and Chemistry
 Cancer Center at Illinois, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States
Vibrational infrared (IR) spectroscopy on a spatially resolved microscopic scale, a mainstay of the emerging field of chemical imaging (CI), has seen unprecedented growth in recent years due to the diversity of new instrument designs and measurement techniques. CI expands on our knowledge of traditional microscopy by deriving its contrast intrinsically from the molecular composition of the sample. Across research ranging from medicine to material science, CI has the potential to enable many impactful contributions. For instance, by introducing automatic quantitative processes to pathology or analysis of microplastics via improvements, CI instrumentation are positioned to take advantage of recent revolutions in data science.
Until recently, there have been few realistic options to traditional FT-IR imaging. While modern FT-IR microscopes are an excellent research analytical tool, unsurpassed especially when prior knowledge of the sample is unavailable, its limitations are apparent for high-throughput routine inspection where we can safely assume a priori information, or for applications where sub-diffraction spatial resolution is necessary. These limitations largely were a result of the low light intensity from globar sources, simply a heated silicon-carbide rod, which was the only option practical for IR imaging. Advances toward higher definition (HD) FT-IR resulted from improved theoretical  understanding, but it was not until recent years with broad commercial availability of new quantum cascade lasers (QCL) that turnkey solutions provided a clear path forward. These high-intensity mid-IR sources were combined as multi-laser assemblies that together were tunable across the entire mid-IR fingerprint region. Their narrow-band emission, in conjunction with their discrete single wavelength broad tunability, allowed systems to measure only the precise spectral features of interest, thus enabling high-throughput, discrete frequency infrared (DFIR) spectroscopy.
For targeted applications, the number of discrete spectral bands required for analytics, especially with the recent developments in data science and machine learning (ML), can often be reduced to sufficiently few with limited loss of accuracy. It is often faster to acquire discrete bands sequentially, contrary to the advantage of simultaneous measurement of a wide spectral range as required by a Fourier transform. The most influential spectral features are chosen through careful design, often requiring significant domain expertise, as is the case with most traditional ML techniques. The process of feature extraction and metric definition is often based on prior scientific knowledge and there is a limit on the number of degrees of freedom that can realistically be defined. Thus, by being reliant on the developer, the performance of these older learning algorithms is capped. New deep learning networks require very little human intervention by comparison. Their performance scales by amount of data and computational power. Especially for biomedical applications, these advances in data science coupled with new CI technology for acquiring vast amounts of information rapidly with cheap computational power and storage, now have the potential to revolutionize diagnostic histopathology and translate CI toward practical clinical use.
The earliest DFIR imaging systems simply retrofitted FT-IR microscopes with a QCL [2, 3]. Over the next several years, several research groups and companies designed new microscopes specifically for integration with QCL sources that specifically take advantage of their many unique properties. The first type of DFIR microscope relies on the direct measurement of residual infrared light after absorption by the sample. These DFIR microscopes have demonstrated drastic improvements in resolution, signal-to-noise ratio (SNR), and imaging speed, thus for the first time, putting CI digital pathology within realistic reach of clinical translation. Systems designed around widefield (WF) laser illumination and utilizing the multi-channel advantage of arrayed detectors have shown exceptional imaging speeds . Meanwhile, scanning systems that tightly focus the coherent laser light into a single spot only a few microns wide  and sequentially mapping the sample with high-speed stages, have demonstrated full-slide imaging with unsurpassed spatial and spectral quality at IR diffraction limited resolution . The flexibility of such a platform has also led to expansion of features including simultaneous acquisition of multiple DFs  as well as the ability to discern molecular orientations through the measurement of linear dichroism .
Recent progress in photothermal (PT) detection provides new opportunities for high resolution CI not possible with conventional IR microscopy. Instead of recording attenuation of IR light, PT instruments encode molecular absorption indirectly by measuring the photo-induced, thermal mechanical response of the sample. Visible Microscopy (VM) and Atomic Force Microscopy (AFM) are the two most common detection methods, both offering resolution beyond the IR diffraction limit. Most notably, VM has shown potential for integrating CI into modern optical microscopy prevalent in clinical and research workflows for histopathology, an IR-optical hybrid (IR-OH) approach , achieving resolution at the sub-micron scale. Similarly, when coupled with AFM scanners, PT detection extends CI to the nanoscale, enabling the study of subcellular molecular chemical structures at high resolution for the first time.
Despite the apparent benefits in resolution, understanding of the processes involved in image formation is limited, resulting in uncertainty in the recorded data. Consequently, these techniques have been reserved for experts whose experience is vital for designing experiments with tightly controlled conditions to mitigate artifacts. Although this approach has seen some success, the applications are obviously limited. Improving PT instruments guided by understanding of image formation offers an alternative to this approach. For example, theory-driven design has led to recent advancements in AFM-based IR (AFM-IR) spectrosocpic imaging instruments. To first approximation, the deflection signal of conventional AFM-IR is proportional to the IR absorption near the probe tip; however, the AFM cantilever introduces additional signal correlated to local mechanical properties of the sample. Guided by rigorous analytical modeling of the cantilever, better AFM-IR instruments have been designed incorporating additional measurements  and advanced controls  to not only correct this effect but also improve the noise of the recorded signal by a factor of five. Each innovation enables new possibilities for CI at the nanoscale such as accurate compositional mapping of biological samples ; however more theoretical progress is required to further the limits of detection. Although PT detection is useful for measuring nanoscale molecular information, understanding the image formation is essential for navigating the complex physics governing the recorded signal.
With the extent of progress across theoretical understanding, instrument design, and data analytics, CI is heading to better data for a multitude of new applications, from the microscale to the nanoscale, that have not been previously feasible. Representative data from state-of-the-art CI microscopy systems, as shown in Figure 1, demonstrate a new scope of biomedical experiments addressable by these modern capabilities. Nevertheless, each of these systems offer unique potential. Systems based on measurement of IR light will be faster simply due to the lack of measurement overhead where each essentially instantaneous detector measurement, sometimes from a single laser shot, can directly map to a single pixel on the image. However, due to the longer wavelength of IR light, these systems are also IR diffraction limited at best and may not sufficiently resolve fine morphologic features. On the other hand, the techniques that rely on a photo-induced sample response can offer substantially higher resolution, in many cases, depending on the resolution limit of the probe instead of the IR beam. Since substantially higher laser power is required to generate a measurable deformation, this restricts possible area of IR illumination typically requiring focused single-point excitation, while also often relying on lock-in amplification or interferometry. Therefore, these techniques are much slower due to the quantity of raw measurements required and its subsequent computation in order to generate each data point. It is important to understand the properties and trade-offs of these techniques and to choose appropriately based on the intended application. The diversity of capabilities provided by this new generation of CI microscopy systems presents scientists with tools appropriate for extracting new information from samples for a broad range of investigations at all length scales.
(1) Reddy, R. R. K. R.; Walsh, M. M. J. M.; Schulmerich, M. V; Carney, P. S.; Bhargava, R. High-Definition Infrared Spectroscopic Imaging. Appl. Spectrosc. 2013, 67 (1), 93–105.
(2) Kole, M. R.; Reddy, R. K.; Schulmerich, M. V; Gelber, M. K.; Bhargava, R. Discrete Frequency Infrared Microspectroscopy and Imaging with a Tunable Quantum Cascade Laser. Anal. Chem. 2012, 84 (23), 10366–10372.
(3) Yeh, K.; Schulmerich, M.; Bhargava, R. Mid-Infrared Microspectroscopic Imaging with a Quantum Cascade Laser. In SPIE; Druy, M. A., Crocombe, R. A., Eds.; 2013; Vol. 8726, p 87260E.
(4) Yeh, K.; Kenkel, S.; Liu, J.-N. J.-N.; Bhargava, R. Fast Infrared Chemical Imaging with a Quantum Cascade Laser. Anal. Chem. 2015, 87 (1), 485–493.
(5) Yeh, K.; Bhargava, R. Discrete Frequency Infrared Imaging Using Quantum Cascade Lasers for Biological Tissue Analysis. In SPIE; Mahadevan-Jansen, A., Petrich, W., Eds.; 2016; Vol. 9074, p 970406.
(6) Mittal, S.; Yeh, K.; Suzanne Leslie, L.; Kenkel, S.; Kajdacsy-Balla, A.; Bhargava, R.; Leslie, L. S.; Kenkel, S.; Kajdacsy-Balla, A.; Bhargava, R. Simultaneous Cancer and Tumor Microenvironment Subtyping Using Confocal Infrared Microscopy for All-Digital Molecular Histopathology. Proc. Natl. Acad. Sci. 2018, 115 (25), E5651–E5660.
(7) Yeh, K.; Lee, D.; Bhargava, R. Multicolor Discrete Frequency Infrared Spectroscopic Imaging. Anal. Chem. 2019, 91 (3), 2177–2185.
(8) Phal, Y.; Yeh, K. L.; Bhargava, R. Polarimetric Infrared Spectroscopic Imaging Using Quantum Cascade Lasers. In Advanced Chemical Microscopy for Life Science and Translational Medicine; Simpson, G. J., Cheng, J.-X., Min, W., Eds.; SPIE, 2020; Vol. 1125210, p 34.
(9) Schnell, M.; Mittal, S.; Falahkheirkhah, K.; Mittal, A.; Yeh, K.; Kenkel, S.; Kajdacsy-Balla, A.; Carney, P. S.; Bhargava, R. All-Digital Histopathology by Infrared-Optical Hybrid Microscopy. Proc. Natl. Acad. Sci. 2020, 117 (7), 3388–3396.
(10) Kenkel, S.; Mittal, A.; Mittal, S.; Bhargava, R. Probe–Sample Interaction-Independent Atomic Force Microscopy–Infrared Spectroscopy: Toward Robust Nanoscale Compositional Mapping. Anal. Chem. 2018, 90 (15), 8845–8855.
(11) Kenkel, S.; Mittal, S.; Bhargava, R. Closed-Loop Atomic Force Microscopy-Infrared Spectroscopic Imaging for Nanoscale Molecular Characterization. Nat. Commun. 2020, 11 (1).
(12) Kenkel, S.; Bhargava, R. Nanoscale Imaging of Biological Samples with Responsivity Corrected Atomic Force Microscopy-Infrared (AFM-IR) Spectroscopy. 2019, No. May, 51.