Open Access
Add to BibSonomy
Abstract
Spectroscopic image data has provided molecular discrimination for numerous fields including: remote sensing, food safety and biomedical imaging. Despite the various technologies for acquiring spectral data, there remains a trade-off when acquiring data. Typically, spectral imaging either requires long acquisition times to collect an image stack with high spectral specificity or acquisition times are shortened at the expense of fewer spectral bands or reduced spatial sampling. Hence, new spectral imaging microscope platforms are needed to help mitigate these limitations. Fluorescence excitation-scanning spectral imaging is one such new technology, which allows more of the emitted signal to be detected than comparable emission-scanning spectral imaging systems. Here, we have developed a new optical geometry that provides spectral illumination for use in excitation-scanning spectral imaging microscope systems. This was accomplished using a wavelength-specific LED array to acquire spectral image data. Feasibility of the LED-based spectral illuminator was evaluated through simulation and benchtop testing and assessment of imaging performance when integrated with a widefield fluorescence microscope. Ray tracing simulations (TracePro) were used to determine optimal optical component selection and geometry. Spectral imaging feasibility was evaluated using a series of 6-label fluorescent slides. The LED-based system response was compared to a previously tested thin-film tunable filter (TFTF)-based system. Spectral unmixing successfully discriminated all fluorescent components in spectral image data acquired from both the LED and TFTF systems. Therefore, the LED-based spectral illuminator provided spectral image data sets with comparable information content so as to allow identification of each fluorescent component. These results provide proof-of-principle demonstration of the ability to combine output from many discrete wavelength LED sources using a double-mirror (Cassegrain style) optical configuration that can be further modified to allow for high speed, video-rate spectral image acquisition. Real-time spectral fluorescence microscopy would allow monitoring of rapid cell signaling processes (i.e., Ca2+ and other second messenger signaling) and has potential to be translated to clinical imaging platforms.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Spectral imaging technologies have provided target detection and molecular analysis capabilities for a wide range of fields, including remote sensing [1–3], agriculture [4], food safety [5–8], historical document preservation [9,10], biological research [11,12], and clinical imaging [13–16]. In the field of fluorescence microscopy, spectral imaging approaches have provided molecular detection capabilities that allow more accurate quantification of stains and fluorescent proteins [17–21], separation of competing signals [22,23], and characterization and analysis of cell and tissue autofluorescence [12,24–26]. Spectral imaging capabilities have been incorporated into several commercial confocal microscope systems, as well as widefield fluorescence microscope modules.
Specific technologies used to enable spectral imaging include filter wheels, tunable filters, dispersive optics, Fourier transform optics, and others that have been summarized elsewhere [27,11,28–30]. Spectral imaging technologies can be divided into two general groups: spectrally and spatially dynamic. Spectrally dynamic optics include tunable filters and filter wheels that mechanically rotate or switch filters to achieve spectral filtering. For example, filter wheel–based hyperspectral imaging systems have been previously utilized for widefield fluorescence microscopy applications, albeit with limited temporal sampling rates [31,32], while we and others have previously demonstrated the utility of mechanically rotated thin-film tunable filters (TFTFs) for widefield fluorescence microscopy, again with limited temporal sampling capabilities [33,34]. Spatially dynamic optics include dispersive optics, acousto-optical tunable filters (AOTF), liquid crystal tunable filters (LCTF) [35] and interferometry-Fourier transform optics [36]. Commercial hyperspectral confocal microscope systems have been designed with spectral detectors that filter fluorescence emission using a grating and PMT array [37,38], while prism-based systems have been utilized for hyperspectral widefield microscope systems [39,40]. Alternatively, self-contained spectral cameras that utilize dispersive optics (such as from SPECIM and others) have also been used to acquire spectroscopic data, such as for laparoscopic applications [41]. Recent studies have demonstrated that HSI systems can be developed using both AOTFs [42] and LCTFs [43] and can achieve relatively fast image capture speeds, albeit with limitations due to polarization and other tuning and beam steering requirements. LCTF-based systems have also been evaluated for clinical endoscopy applications using reflected light [44]. Spectral imaging microscope systems have also been implemented using supercontinuum lasers to provide spectral illumination. Broadband supercontinuum laser sources require additional optics, such as AOTFs or gratings, in order to provide spectral discrimination. Poudel et al. (2018) provides an excellent review on the various supercontinuum configurations evaluated to date [45]. LED sources have also been utilized, both as high-power broadband, white light sources to replace arc lamps [46,47] and as multi-channel illumination sources, where four LED channels may be rapidly switched for multispectral fluorescence microscopy [48].
Interestingly, despite the many and diverse technologies used to acquire fluorescence microscope spectral image data, most approaches share common trade-offs [49,50]. Filter-based optics (such as filter wheels) for spectral imaging are typically mechanically limited and difficult to utilize for high-speed acquisition. To date, some of the fastest filter wheel modules can change between adjacent filter positions in approximately 20 ms, for a filter wheel with 6 positions [51]. It is possible that a spectral microscope system could be designed utilizing multiple filter wheel modules that are stacked, if each module contains a blank position that allows for selection of filters from adjacent modules. However, at best, spectral acquisition rates would still be limited by filter switch times and such a system would become complex if many wavelength bands were required, as multiple filter wheel modules would need to be stacked in serial. In addition, this approach could produce potential vibration artifact from the multiple moving components, and the system lifecycle that would be tied to mechanical wear. Fixed filters have been alternatively utilized in snapshot spectral imaging systems that can provide rapid acquisition speeds, but these typically require a compromise between the number of wavelength bands acquired (spectral sampling) and spatial sampling [38,39]. Dispersion-based systems may be able to achieve high-speed spectral imaging, but typically have a reduced photon budget due to the additional optics and slits employed, as well as fluorescence emission distributed onto many detector elements [27,45]. Current LED-based systems are limited by the number of LED channels available or mechanical movement required to cycle through multiple banks LEDs [48]. However, the trend in cellular signaling experiments is continuing to move towards complex, multi-function intracellular and intercellular dynamics for live cell imaging (i.e., the multiple utilities of Ca2+ signaling) [52]. Hence, there is a need for spectral imaging technologies that enable both high speed and high sensitivity in order to discriminate multiple components. Further driving this need is that many biomedical hyperspectral imaging studies are increasingly moving towards multi-label (5+) quantitative discrimination in performing spectroscopic analysis, promoting the need for increased spectral scan range with fast acquisition times [53–56]. These needs can be fulfilled when imaging bright specimens or specimens that are photostable and not rapidly changing. However, for rapid cell signaling experiments, especially those involving fluorescent protein–based labels, the specimen is often both dynamically changing and photosensitive. Hence, it is important to capture emitted photons as efficiently as possible, with both high sensitivity and speed.
We have previously demonstrated that scanning of the fluorescence excitation spectrum (excitation-scanning) may offer an alternative approach for spectral imaging fluorescence microscopy, while providing greatly increased (>10X) signal strength when compared to emission scanning when using identical illumination intensity [34,57]. Initial feasibility of this approach was demonstrated in a side-by-side comparison that utilized an array of thin-film tunable filters (TFTFs). While this implementation demonstrated feasibility for excitation-scanning spectral imaging, delays associated with mechanical rotation of TFTFs resulted in delays of 100-250 ms when switching between wavelength bands, resulting in acquisition times that were insufficient for rapid time-lapse cell signaling assays.
Here, we present initial results from an alternative technology for implementing excitation-scanning spectral imaging using an array of wavelength-dependent light emitting diodes (LEDs) that allow rapid wavelength switching. A novel hyperspectral optical geometry was developed that uses a pair of reflecting mirrors, similar to a Cassegrain telescope design, to combine optical output from the LEDs into a single liquid light guide for coupling into a fluorescence microscope. Bench-testing and imaging results indicate that this approach may be suitable for a range of experimental assays that can be performed on a widefield microscope base.
2. Methods
Optical ray trace modeling was used to simulate the theoretical performance of the LED-based spectral illuminator and to optimize the optical and geometric configuration. Prototyping and benchtop testing were then performed, based on the optimal design indicated by ray trace simulations. Finally, the spectral illuminator was integrated with a widefield fluorescence microscope and image performance was compared to our prior excitation-scanning spectral microscope system [33,34].
2.1 Ray trace simulations
Optical ray trace simulations were conducted using TracePro software (Lambda Research Corp.). A range of simulations were performed to select optical components and specify geometry in order to optimize transmission through the spectral illuminator [58]. An overall optical geometry was implemented that utilized two mirrors to direct light from all LEDs to a common entrance aperture of a liquid light guide. The design was similar to a Cassegrain telescope, but utilizing a concave mirror and a flat mirror (Fig. 1). To allow for many wavelengths of illumination, LEDs were arranged in a concentric pattern around the flat mirror and focused using a lens.
Fig. 1. Conceptual illustration for the spectral illuminator to combine optical output from multiple wavelength-specific LEDs located in a ring and reflected to a common location. (a) The drawing here only presents one LED for clear optical pathways. Renderings of the spectral illuminator optical components as simulated in TracePro optical ray trace software for (b) identical layout as the concept drawing and representative depiction of the optical components (mirrors-silver, lenses-blue and LED-red) used in simulations. (c) Displays an example of the ray trace results for a single LED and lens location (0.5% of rays shown for visualization purposes). (d) A rendering of all optical components able to fit this version of the simulated spectral illuminator.
Monte Carlo–based ray tracing was performed using 100,000 rays per simulation. This number was determined to be sufficient to achieve a repeatable result when multiple ray traces were conducted of the same model. LED illuminators were simulated by importing optical specifications from manufacturer data sheets (SMB1N series, Marubeni Co.). The radiance and spectral profile were extracted for each LED model and uploaded to the TracePro Surface Source Properties library. Lens and mirror models were imported from manufacturer-supplied Zemax files (Edmund Optics). A range of lens focal lengths (FL) and mirror diameters (D) were evaluated (Table 1) that included three different diameter mirrors and a single lens.
Table 1. Optical components used for ray trace simulations. Simulations were performed using three different diameter mirrors (columns 2-4), where the focal length (FL) and diameter (D) of each mirror is indicated in the column header. For each mirror, a range of focal length lenses were evaluated using either a single lens or dual lens optical path. All lenses were evaluated at a 15 mm diameter, with the exception of the 12 mm focal length lens, which was evaluated at a 12 mm diameter.
All simulations were implemented using custom macro scripts within TracePro to allow evaluation of a range of geometric locations and orientations for each component. In summary, the macro first defined the properties, location, and orientation of the two mirrors, next created a 5 mm detector at the center of the concave mirror to simulate the liquid light guide (LLG) entrance aperture, then defined LED source properties (525 nm), location, and orientation, and finally defined lens properties, orientation and location. A parametric sensitivity study was performed by changing the location of lenses and LEDs in the vertical and horizontal axes while performing a ray trace simulation for each location and measuring simulated optical power (irradiance) upon the LLG entrance aperture (Fig. 1.b,c). Lens positions were varied in 1 mm increments, while LED positions were varied in 10 mm increments, as defined in a nested loop macro (example simulation loops of the model are illustrated in Visualization 1). These data were used to determine optical component selection and placement in order to provide optimal output of the spectral illuminator.
2.2 Prototype fabrication
A prototype spectral illuminator was constructed using the optimal components and geometric spacing identified through optical ray trace modeling. The prototype consisted of 13 wavelength-specific LEDs (SMB1N Series, Roithner Lasertechnik GmbH, specifications in Table S1 of Supplemental 1), 44.5 mm FL x 12.7 mm diameter plano-convex lenses (49-860, Edmund Optics), and a 75 mm diameter mirror (CM750-075-F01, Thorlabs) customized to include a 7 mm center through hole. Optical components were mounted in a series of custom cage plates fabricated by Thorlabs. In addition, XY linear translation mounts were used to allow fine tuning of LEDs, focusing lenses, concave mirror, and flat mirror to test alignment sensitivity.
Custom printed circuit boards (PCB) were designed using Pad2Pad software (Pad2Pad, Inc.) for surface mount LEDs (LED board) and for routing signals from the computer interface boards to a series of LED current drivers (current driver board). Digital and analog controls were provided through NI 6363 and NI 6723 cards (National Instruments, Inc.), respectively. Digital signals were used to switch current drivers (RCD-24-1.20, RECOM) on and off rapidly, while analog signals were used as a reference voltage to regulate the current output of each LED driver, and hence LED power output.
2.3 Benchtop characterization
Benchtop testing consisted of determining optical power transmission and optimal spacing of components. Spectral irradiance was measured using a fiber-coupled spectrometer (QE65000, Ocean Optics) and integrating sphere (4P-GPS-030-SF, Labsphere) calibrated to a NIST-traceable light source (LS-1-CAL, Ocean Optics). All measurements were acquired by averaging 10 scans (resulting standard deviation of measurements is ± 0.002 mW). Irradiance was used to align the optical components and allow comparison with simulation data. To determine the optimal position for components, four LEDs were utilized that represented four extreme locations on the circular LED array (e.g., N, S, E, and W). An additional LED was added to ensure sampling of wavelengths throughout the target spectral range (365 nm, 395 nm, 430 nm, 450 nm and 525 nm). These five LEDs were illuminated simultaneously and the total power output was measured at the distal end (output) of the LLG (O7776558 series 300, Lumatec GmbH), while optical component positions were adjusted. The final position for each component was determined so as to achieve a maximal total power output while also ensuring that each wavelength-specific LED contributed approximately equal to the spectral power distribution. Upon selecting optimal component locations, the LLG was connected to a TE-2000 inverted fluorescence microscope (Nikon Instruments, Inc., Table 2) and the irradiance was assessed at the microscope stage. The optical power output, as a function of current driver setting, was characterized for each LED, thus allowing a power output look-up table to be constructed for each excitation wavelength. In addition, to allow a comparison of ray trace simulation results, a new set of simulation models were formed to replicate the prototype positioning and wavelength illumination. The revised ray trace simulations were performed at each optical component spacing as was evaluated for the prototype, mentioned above. Irradiance was measured at the interrogation surface of the model (i.e., the simulated entrance to the liquid light guide). Irradiance measurements of the prototype (prior to coupling with the fluorescence microscope) were compared to simulated irradiance of ray trace model for validation. In summary, the prototype optical irradiance was measured at the distal end of a coupled LLG and the simulation was measured at an interrogation plane that represented the entrance or proximal end of a LLG but did not account for acceptance angle effects of the LLG or transmission losses through the LLG.
Table 2. Two microscope platform configurations including the components and parameters used for imaging.
2.4 Spectral imaging feasibility tests
To assess the feasibility of LED-based excitation-scanning spectral imaging, the prototype system was compared to our previously-reported excitation-scanning spectral imaging system that utilizes an array of thin-film tunable filters, or TFTFs (VersaChrome, Semrock) mounted in a custom mechanical tuning device (VF-5, Sutter Instrument Co.) [33,34]. Both systems were implemented in succession using the same TE-2000 inverted microscope base (Fig. 2), equipped identically for each trial (Table 2). Image acquisition speed for each system was selected so as to fill as much of the dynamic range of the detector as possible at weak (low illumination power) spectral bands while avoiding oversaturation of strong (high illumination power) spectral bands.
Fig. 2. Light path schematics for the two spectral illumination pathways used in hyperspectral image acquisition: (a) TFTF-based illumination using mechanical tuning (both rotationally about the axis and perpendicular to the axis) implemented in a Sutter, VF-5 tuning system and (b) LED-based illumination using electronic switching of the custom wavelength-specific LED array. Both illumination sources were transmitted via liquid light guide (LLG) to a collimator on the back side of the widefield fluorescence microscope. The collimated beam was reflected off the dichroic long-pass (LP) filter to the sample. Fluorescence emission above the cutoff wavelength was transmitted through the dichroic LP filter and reflected to the camera detector for acquisition.
Test samples were prepared by Abberior, GmbH, and consisted of African green monkey kidney epithelial cells that were labeled using a custom 6-label scheme (Table 3), as well as corresponding single-label and unlabeled slides to serve as controls for building a spectral library. Cells were fixed and stained according to Wurm, et al. [59] and were embedded in Abberior Mount Solid Antifade.
Table 3. Cellular components labeled and the corresponding fluorescent label for six-labeled slides used for spectral imaging feasibility testing
Image data were acquired in the following order: 1) single-label control slides were imaged using the TFTF system, 2) the six-label slide was imaged and the XY microscope stage coordinates recorded, 3) the identical field of view, with identical XY coordinates, was imaged using the prototype LED-based system, and 4) single-label control slides were imaged using the LED-based system. Spectral images were acquired sequentially from shortest to longest wavelength in each system with only one wavelength illuminated at a time. This was achieved in the TFTF system by rotating and switching TFTF filters, while in the LED system by sequentially switching LED wavelengths on or off. This approach ensured that a side-by-side comparison of both systems was possible using the same field of view and identical microscope and camera settings with the exception of image acquisition times.
3. Results
Results from this study are organized in four subsections – optical ray trace simulations to assess early feasibility and to optimize component selection and geometry in silico, prototype development, benchtop irradiance testing to compare and validate physical and theoretical system performance, and feasibility image testing to assess potential for use in an integrated epifluorescence microscope system.
3.1 Ray trace simulations
Optical ray trace simulations were performed to evaluate the theoretical feasibility of the optical configuration prior to prototyping. Simulations consisted of modeling the components in the light path of the spectral illuminator, including the LED, lens, concave mirror, flat mirror and interrogation plane at the center of the concave mirror (representing the input aperture of the LLG). These tests were used to determine optimal lens and concave mirror parameters and location. Typical results consisted of irradiance output for the simulated single lens options at every position tested (example of how the data trends with the positioning of the LED and lens are shown in Visualization 1). The maximum optical transmission was achieved for each lens when the LED-lens spacing was equal to the lens FL. A sample-set of the data and summary graphs to present a synopsis of the entire one lens configuration simulation dataset is provided as Fig. S1 in Supplemental 1.
Transmission was measured as the integrated irradiance captured at the simulated LLG entrance aperture divided by the irradiance produced by the simulated LED. The values shown below (Table 4) summarize the model results by reporting the average maximum output for each lens configuration when a similar maximum value was achieved regardless of lens-LED spacing (this trend is depicted in the right column of Fig. S1 in Supplemental 1).
Table 4. Transmission maximums for each lens configuration using the three different diameter mirrors. The optimal combinations are highlighted by the boldened percent transmission
Simulation results indicated an inverse relationship between concave mirror FL and efficiency (optical transmission) where a 76.2 mm FL mirror averaged 11% transmission, a 114.3 mm FL mirror, 6.5% transmission and a 152.4 mm FL mirror averaged ∼4% transmission. However, it should be noted that simulated optical transmission was calculated solely on optical power incident on the interrogation plane, which was used to simulate the entrance aperture of the LLG, regardless of incident light angle. Hence, simulation results did not account for angle-dependent losses that could be encountered during optical coupling with the LLG. Among the six focal length lenses evaluated, the 45 mm lens provided the highest optical efficiency, with 11% optical transmission for the 76.2 mm FL mirror.
Upon reviewing all model simulation results, a lens configuration using the 76.2 mm focal length mirror and 45 mm focal length lens was selected for prototyping. This configuration was simulated to achieve 11% optical transmission to the entrance aperture of the LLG, which corresponds to approximately 100 mW of available optical power, depending on wavelength, that can be coupled into the LLG.
3.2 Prototype fabrication
A 3” cage system was utilized to encapsulate and align all optical components. Custom cage plates were machined to house the two mirrors, the lens array, and the LED array (Fig. 3.a,b). Due to limitations in packing density of lenses in the lens array, the prototype was designed to house 13 lenses, and hence 13 LEDs with corresponding peak wavelengths. A custom PCB was utilized to align surface mount LEDs with respective lenses. The lenses and LEDs were placed radially equidistant in a circular pattern of diameter of 63.5 mm. A second custom PCB was printed to connect individual analog and digital lines from the computer interface board to a respective current driver for each LED. Hence, separate analog power output and digital on/off control was provided for each LED. TTL triggering was utilized to synchronize LED wavelength switching with the camera acquisition. LED intensities (mW) were measured at reference voltage intervals of 0.5V from 4V to ∼2V (within 95% of the maximum forward current allowed) using the spectrometer detailed above. The voltage drop across the 1 Ω resistor was used to calculate the current supplied to the LED. The relationship between the reference voltage supplied, the current driver output, and the corresponding LED radiant flux was linear (see Fig. S2 of Supplemental 1). Cage plates were mounted on XY translational stages for precise alignment measurements when validating the optics positioning (Fig. 3.c).
Fig. 3. Fabrication of the dual mirror spectral illuminator prototype. (a) A CAD rendering of the custom cage plates, (b) assembly of the custom cage with LED circuitry and optical components. (c) Custom cage plates mounted to linear translation stages to assess alignment sensitivity during benchtop irradiance testing. (d) Example photograph of the spectral LED array with all wavelengths illuminated for demonstration purposes.
3.3 Benchtop characterization
The alignment process consisted of two steps: 1) alignment to produce the highest radiant power output for a selected wavelength LED as measured at the output of the LLG, and 2) alignment to compromise between the power output of all wavelengths so as to produce a spectrally balanced illuminator (i.e., similar optical power output across all wavelength bands). In some cases, a tradeoff was required between achieving maximal power of a single LED and achieving balanced spectral power output. Once positioned, results were compared to an updated ray trace model for validation. (Fig. 4)
Fig. 4. Prototype sensitivity response to optical component position and comparison to ray trace simulations. Independent variables were the concave mirror-lens spacing (represented by different colored data series in the graphs) and LED-lens spacing (represented by each data point in a series). (a) Irradiance measurements of the prototype LED-based illuminator using a spectrometer for integrated optical power. (b) Prototype alignment was adjusted via XY translational stages (see also Fig. 3.c). Positions were altered in 5 mm increments (circular data markers). A subset of data was sampled with a 1 mm increment (triangular data markers) so as to accurately identify the optimal position for components (expanded in red of panel (a)). (c) Simulated irradiance measurements of modeled illuminator using ray trace analysis of the same optical components as the prototype (panel (a)) for comparison. (d) Model positions were adjusted in silico to match the positions of prototype measurements. The flat mirror and lens spacing were held constant to replicate the stationary position of the prototype mirror and lens plate. (e) The spectral output of the prototype illuminator with 5 wavelengths illuminated simultaneously was used to visualize the dependence of spectral power output on position.
Results from the experimental sensitivity study of the prototype found that a 44.97 mm spacing between concave mirror and lens (focal length of the lenses) and a 33.5 mm spacing between LED and lens were optimal (Fig. 4.a). Smaller step sizes revealed that a minor change to a 45.97 mm spacing between concave mirror and lens provided a nominal improvement in overall power output (highlighted expanded view of Fig. 4.a). The total power output remained constant across the range of LED-lens spacings evaluated with fine tuning (linearity of the data in the expanded section of Fig. 4.a). Therefore, the optical component positions were determined by LED power distribution instead of integrated total power output (Fig. 4.e). A 3D plot was used to visualize the spectral power output of the system as a function of LED-lens spacing (Fig. 4.e). This information was used to determine the dependence of LED spectral power distribution upon LED-lens spacing and to select a LED-lens spacing that provided consistent power output across all LED wavelengths. The depth axis boundaries of the 3D graph (Fig. 4.e) are the same boundaries as the x-axis of the expanded graph in Fig. 4.a. While the 37.5 mm LED-lens spacing (the maximum z-axis value of Fig. 4.e) produced a single peak wavelength with the highest power output, the 28.5 mm LED-lens spacing produced an even distribution of power across all wavelength bands (the minimum z-axis value of Fig. 4.e). This can also be visualized in the compressed view of the 3D plot (the bottom of panel (e)). Hence, a LED-lens spacing of 28.5 mm was selected as a compromise between peak illumination power output and even spectral power distribution.
The optical power output measured in the prototype experimental sensitivity study was compared to simulation results (Fig. 4.a,c) and indicated similar overall trends in the power output of the system as a function of optical component position. Simulation results provided improved optical transmission when compared to the experimental prototype, likely due to the following reasons: 1) the simulation utilized a 5 mm diameter interrogation plane to measure power output whereas the free aperture of the LLG was only 3.5 mm diameter; 2) the simulation measured total power incident upon the interrogation plane whereas the experimental prototype measured power output after coupling through the LLG; 3) the experimental simulation measured power available at the interrogation plane regardless of angular dispersion whereas the LLG utilized in the prototype has a specified full angular acceptance of 72°. The differences in optimal spacing between the lens and concave mirror that were observed between the simulation (54 mm) and experimental prototype (44 mm) are likely accounted for by the nonidealities of the experimental prototype and the limitations of the simulation described above. While absolute power output values differed between experimental and simulation sensitivity studies, the overall trend of radiant power vs. optical component spacing was similar and allowed an optimal spacing to be identified.
Next, the spectral illuminator was implemented with a TE-2000 inverted epifluorescence microscope system and irradiance measurements were made at the microscope stage to assess the total system excitation power output. Within the spectral microscope system, there were multiple optical elements that contributed to optical transmission losses: lenses and mirrors of the dual mirror array illuminator, coupling losses at the entrance to the LLG, transmission losses through the LLG, the LLG microscope collimator, the dichroic mirror, and the lenses and apertures of the microscope objective. Hence, irradiance data acquired at the microscope stage represent the composite effects of all of these optical elements. Irradiance data were acquired to assess the maximum output of each LED at the stage as well as the power output vs. current supplied by the current drivers, which created a power output vs. reference voltage calibration for each LED. The maximum output ranged from 0.017–0.063 mW depending on LED (the TFTF setup ranged from 0.021–0.149 mW depending on the band-pass). To illuminate all 13 wavelengths at equal power output (i.e., flat spectral illumination), the highest power output available was 0.017 mW at the microscope stage. This power output was much less than predicted by the ray trace simulations or than indicated by benchtop testing of the spectral illuminator alone, and these additional transmission losses were attributed to effects of the LLG collimator and aperture stops within the microscope objective. Hence, for initial image feasibility testing (below), LEDs were operated at maximum power output to achieve sufficient signal strength within each spectral band, and post-acquisition spectral correction was performed to return the spectral image data to a flat spectral response.
3.4 Spectral imaging feasibility tests
Feasibility of the mirror-based spectral LED illuminator for performing excitation-scanning spectral imaging microscopy was evaluated through a side-by-side comparison with a previously-developed spectral illuminator based upon a 300 W Xe arc lamp and TFTF array mounted in a tilting filter wheel to allow for mechanical filter tuning. Both spectral illuminator systems were integrated with an inverted epifluorescence microscope platform and identical samples, from identical fields of view, were imaged on both systems using the same objective and camera. Raw spectral image data were visualized by summing all wavelength bands to view total fluorescence and through selection of three wavelength bands to view a RGB composite image. A spectral library was then formed by sampling spectra from single-labeled specimens and the library was then used with non-negatively constrained linear unmixing to visualize signals from each of the six labels, as well as autofluorescence. A false-colored merged composite image was then created from unmixed abundance images, and the root-mean-square (RMS) error associated with linear unmixing was also calculated. To assess potential effects of sequential wavelength acquisition on photobleaching, a sample FOV was repeatedly imaged and the spectral bands summed to calculate the total fluorescence intensity. A region of interest (ROI) was then selected and the average summed fluorescence signal extracted for each sequential spectral image set (Fig. S3 in Supplemental 1). Results indicate that a signal loss of ≤2% of summed fluorescence was present between sequential images for the TFTF system and ∼0% signal loss for the LED system (photobleaching loss was below the measurement capability of this approach). Hence, it is likely that the order of wavelength acquisition bands had negligible impact on the fluorescence signatures of the spectral images acquired using the settings described in this study.
A spectral library was constructed by imaging single label control samples for each fluorescent label and extracting the spectral signature of each label (Fig. S4 in Supplemental 1). In addition, an unstained slide was imaged to measure the autofluorescence (AF) signature. Imaging was conducted using the thin film tunable filter (TFTF)-based spectral light source and a LED-based spectral light source. Identical objectives, microscope configurations, and imaging parameters were used for the mixed label slide and single label slides. A background region of interest (ROI) was selected within each spectral image and the background spectrum extracted. The background spectrum and a correction factor (determined for each respective spectral light source) were used to subtract the background and correct each wavelength band to a flat spectral response [33,34]. A region of high signal strength was selected in the corrected images for each single label sample and autofluorescence, and the representative spectra were extracted. Spectral specificity was validated by linear unmixing of single label spectral images using a spectral library that consisted of the target fluorescent label and autofluorescence for both the TFTF-based system (Fig. S5 in Supplemental 1) and the LED-based system (Fig. S6 in Supplemental 1). The spectral library containing signatures from all fluorescent labels and autofluorescence was used to unmix spectral image data from the multi-label slides as described in the body of the manuscript. A comparison of the spectral signatures as measured by the two spectral microscope configurations is presented in Fig. S7 in Supplemental 1. The spectra were also compared to spectra listed on the manufacture website (Fig. S8 in Supplemental 1).
Results from excitation-scanning spectral imaging with the previously developed TFTF system demonstrated that distinct spectral signatures were able to be extracted from single-labeled specimens (Fig. S5 of Supplemental 1) and that analysis with linear unmixing allowed clear identification of signals from most labels – autofluorescence, mitochondria, dsDNA, vimentin, and Golgi labels were all easily identifiable (Fig. 5.f,i-l). The f-actin label was also identifiable, although appearing with uneven distribution across the field of view (Fig. 5.h), while the nuclear pore protein (NPP) had insufficient signal strength to be definitively identified as being localized to nuclear pore structures (Fig. 5.g). It is likely that the NPP signal was inherently low due to the size of the nuclear pore structures and, hence, could not be easily discriminated. Scaling the NPP intensity for improved visualization was found to only magnify the contribution of noise in both individual and merged images. The overlayed unmixed image (Fig. 5.c) allowed co-visualization of unmixed components. The Golgi label displayed a relatively strong signal, causing false-colored nuclei to appear purple in the overlayed image, due to merging with the dsDNA signal. However, the excitation spectrum identified from the single-label Golgi sample was weak, and likely mixed with cellular autofluorescence, which was also localized near the nuclei. Hence, it is possible that the Golgi signal resulting from linear unmixing may have been a mixture of autofluorescence and Golgi contributions and did not fully represent Golgi labeling in the cells.
Fig. 5. Spectral image data acquired from a 6-label slide using a TFTF-based system for excitation-scanning spectral imaging microscopy. Spectral image data were visualized as (a) a summed intensity and (b) a RGB false-colored image. Unmixed image data were also (c) false-colored and merged for visualization, along with (d) RMS error associated with the unmixing process. To perform unmixing, (e) a spectral library was constructed from single-label control specimens and used to estimate the relative abundance of each fluorescent label in the mixed sample: (f) autofluorescence, (g) NPP, (h) f-actin, (i) mitochondria, (j) dsDNA, (k) vimentin and (l) Golgi.
It should be noted for spectral image data acquired on both the TFTF system and the LED system, that it was not possible to find a region of interest (ROI) with strong f-actin that did not also have significant contributions from cellular autofluorescence, due to the highly autofluorescent nature of African Green Monkey kidney epithelial cells. Hence, the spectrum of the f-actin label, Star Green, was extracted from data provided by the manufacturer for the 38 wavelength bands used in the TFTF system and for the 13 wavelength bands used in the LED system. The localization of labeling in the unmixed data indicates that signals from f-actin do appear to be localized within actin filament structures, although there was non-uniform distribution of the unmixed f-actin signal, as described above.
Upon completion of imaging with the TFTF system, the same FOV was imaged using the LED array system for a side-by-side comparison (Fig. 7). Linear unmixing of spectral image data acquired with the LED-based system allowed clear identification of autofluorescence, dsDNA and vimentin (Fig. 6.f,j,k) and partial identification of NPP, f-actin, mitochondria and Golgi (Fig. 6.g,h,i,l). While the mitochondria and f-actin signals had degraded signal as compared to data from the TFTF-based system, unmixing still allowed identification of signal that was localized to the correct subcellular locations. Interestingly, the Golgi signal appeared better localized to perinuclear locations for the LED-based system than for the TFTF-based system. In comparing system response, it should be noted that the prototype LED-based system was configured with 13 wavelength bands, as compared to the 38 bands used for the TFTF-based system (wavelengths indicated in Table 2). In addition, radiant output power measurements of the LED system were acquired at the microscope stage, prior to imaging, indicating an average of ∼40% less power than that of the TFTF-based system.
Fig. 6. Spectral image data acquired from a 6-label slide using the prototype LED-based system for excitation-scanning spectral imaging microscopy. Spectral image data were visualized as (a) a summed intensity and (b) a RGB false-colored image. Unmixed image data were also (c) false-colored and merged for visualization, along with (d) RMS error associated with the unmixing process. To perform unmixing, (e) a spectral library was constructed from single-label control specimens and used to estimate the relative abundance of each fluorescent label in the mixed sample: (f) autofluorescence, (g) NPP, (h) f-actin, (i) mitochondria, (j) dsDNA, (k) vimentin and (l) Golgi.
Fig. 7. A side-by-side comparison excitation-scanning spectral image data acquired from the same field of view using: (a) a TFTF-based system when imaging with 38 wavelength bands, (b) a TFTF-based system when imaging with 13 wavelength bands (c) and the prototype LED-based system when imaging with 13 wavelength bands.
To further compare system performance of the TFTF-based system and the LED-based system, the number of wavelength bands utilized by the TFTF-based system was down-sampled so as to match the number and wavelength location of bands provided by the LED-based system. The down-sampled spectral image data were unmixed and analyzed using the same process as described above for the full spectral image data set, and resulting unmixed images were false-colored and merged for visual comparison (Fig. 7). In general, there was a strong agreement in signals identified between the three data sets – the TFTF-based system with all wavelength bands (Fig. 7.a), the TFTF-based system with reduced number of wavelength bands (Fig. 7.b) and the LED-based system (Fig. 7.c). The signal-to-noise ratio (SNR) ranges for the TFTF-based system with reduced number of wavelength bands and the LED-based system were equitable. The SNR of the TFTF-based system with the full wavelength range was higher and can be attributed to the higher irradiance output and the higher number of wavelengths scanned. Two noticeable differences between the TFTF-based system and the LED-based system were the level of noise (the LED-based system generated images with reduced SNR compared to that of the TFTF-based system) and that nuclei appear more blue in false-colored images from the LED-based system due to the absence of Golgi signal in the nuclei (more accurate identification of Golgi to perinuclear locations). The differences in SNR between the TFTF and LED systems are likely attributed to the following (in order of impact): 1) differences in illumination power and subsequent photon budget available at the detector, 2) the number of spectral bands scanned and 3) potentially minor differences in the bandwidth of the spectral bands between the two systems, which when combined with the excitation spectral properties of each fluorescent label may result in subtle variations in detection sensitivity for each label. Illumination power likely directly correlates with SNR, as a 40% decrease in average illumination is very comparable to the 40% average decrease in SNR, as measured across the different spectrally-unmixed channels. However, the LED-based system acquired spectral images that contained fewer (66% decreased) spectral bands than the TFTF-based system, and it is likely that the reduced number of spectral bands also affected the SNR of unmixed images. In fact, when the spectral image data from the TFTF-based system was subsampled to contain the same number of spectral bands and identical band location as the LED-based system, the unmixed images from both systems displayed a similar range in SNR values. Hence, it is likely that both decreased illumination power primarily affected the SNR of resultant unmixed images, while decreased spectral sampling had a secondary and compounding effect. Further data are supplied in Supplemental 1 for building and verifying the spectral library, including: Abberior stain spectral data (Fig. S4), analysis and verification of single-labeled control samples for both systems (Fig. S5 and Fig. S6) and comparison of fluorescent label spectra for each component between the two spectral light sources used for imaging (Fig. S7 and Fig. S8).
4. Discussion
Spectral imaging approaches have shown great utility in fluorescence microscopy, but often come at the price of increased acquisition time and/or decreased signal strength. In this manuscript, we have presented simulation, prototyping, benchtop testing, and feasibility imaging results for a prototype spectral illuminator that allows excitation-scanning spectral imaging to be performed on an epifluorescence microscope platform with rapid wavelength switch times. The novel spectral illuminator is based on an optical geometry that uses two mirrors to combine light from an array of wavelength-dependent LEDs in a manner similar to Cassegrain style telescopes. The prototype LED-based system was compared to a preexisting TFTF-based system, and results indicate that both systems provided the ability to discriminate multiple fluorescent label signatures in a highly-labeled sample, although the TFTF-based system provided higher illumination power at the sample stage and correspondingly higher signal-to-noise characteristics in unmixed images. The LED-based system did provide a streamlined optical design that acquired a spectral image stack at comparable acquisition rates of standard spectral imaging systems due to the electronic switching of band-passes instead of mechanical tuning. Both systems acquired spectral image data in a similar manner, through sequential illumination of excitation wavelength band while acquiring an image at each band. The TFTF system used an array of filters in a tiltable filter wheel, where each filter could be selected through rotation and then the angle of the filter relative to the incident light could be adjusted by tilting the wheel. The LED system used sequential electronic triggering of each LED, where a typical LED rise or fall time is on the order of 0.01 µs. The long-term goal of this work is to enable high-speed acquisition of spectral fluorescence microscopy images with an acquisition speed of 5-10 ms per wavelength band. Hence, the LED rise and fall time of ∼0.01 µs adds a negligible delay for high-speed spectral imaging. By contrast, the TFTF system requires between 50-200 ms to switch between wavelength bands, depending up on whether the wavelength band switch requires adjustment of the tilt of the filter wheel or rotation to an adjacent filter in the wheel. To ensure sufficient time for all wavelength switches, a 200 ms wait time was required after issuing each wavelength switch command before acquiring the corresponding wavelength band images. Hence, mechanical movement of the TFTF system becomes prohibitively rate limiting when using acquisition times of ∼10 ms per wavelength band.
Unmixed spectral image data acquired using both the TFTF system and LED system were comparable, albeit when operating the LED system at a reduced acquisition speed. When comparing the TFTF spectral image acquired with identical wavelength bands as the LED spectral image, unmixed images from both systems presented comparable SNR (Fig. 7.b,c). Acquisition speed of the LED system was slower (Table 2) to compensate for an average of 40% reduction in illumination power. Factors that contribute to the illumination power losses in the LED system include the incident angle of each LED light path upon the incident face of the LLG and the beam diameter at the LLG. In this initial prototype, the beam diameter is larger than the entrance aperture of the LLG, and light outside of the entrance aperture is discarded. By contrast, the TFTF system features a single beam that is orthogonally incident upon the LLG entrance aperture (as opposed to an off angle) and LLG coupling is more efficient. Hence, further refinement of the prototype LED-based system is needed to match the power output characteristics of the TFTF-based system and, ideally, to allow spectral data to be acquired at high speeds with negligible time delays introduced by the electronic wavelength switching of LEDs. These capabilities are important to support live, real-time, multi-label cellular studies.
4.1 Model and prototype validation
A conceptual model and a subsequent series of ray trace simulations were performed to optimize optical power transmission by systematically adjusting geometric and optical parameters of the system. Upon optimization at a single wavelength, the model was expanded to simulate transmission of multiple wavelengths. However, the full parametric sensitivity study and optimization process was performed on just a single wavelength band before expanding to all wavelength bands, in order to streamline the simulation process. Hence, there may be subtle compromises in component spacing or alignment to allow approximately uniform optical transmission across all wavelength bands that were required to be made during the prototyping and benchtop testing phases and which could conceivably have been accounted for during the modeling phase, albeit at the expense of increased computational burden and more complex designs for optical alignment components to adjust geometry independently for each wavelength. A revised simulation that included 5 wavelength bands was used to compare wavelength-and LED placement-dependent effects between the simulation and benchtop measurements (Fig. 4). For imaging experiments, sample refocus was not performed for each wavelength band as it was assumed that the chromatically-corrected objective provided and equivalent focal length across the range of wavelengths utilized in this study.
In addition, the geometry of the initial prototype was designed to match the optimal optical power transmission characteristics provided by a particular combination of mirror and lens diameter and focal length. This included limiting the number of excitation wavelengths to 13, due to packing constraints in designing the lens array cage plate. In future prototypes, it is likely that alternative lens packing geometries may be implemented that will allow a greater number of wavelength bands to allow for increased wavelength sampling between 350 nm and 600 nm. One compromise to note was that shorter focal lengths of the concave mirror resulted in an incident angle of illumination that was greater than the acceptance angle (as determined by the numerical aperture) of the LLG. Ultimately, a shorter focal length (76.2 mm) mirror was implemented because the highest simulated transmission optical transmission could be improved by ensuring that the incident angle of light was within the LLG acceptance angle.
A design restriction maintained in both the design and prototype was the lenses and flat mirror were secured using the same cage plate (Fig. 3.b). This restriction was necessary as a separate cage plate for the flat mirror would have occluded the light path between the lens and curved mirror. However, this design restriction also served to limit the range of adjustment in the spacing between the flat mirror and the lenses (this distance can be minimally adjusted through the threaded lens tube that secures the flat mirror). Related, the constraint of mounting all lenses within a single cage plate, also limited the amount of adjustment that could be performed to mitigate effects from chromatic aberration.
Future modeling goals include the development of an illuminator geometry with 32 + wavelengths and that can allow increased geometric translation of individual optical components, while maintaining accurate alignment. In addition, the use of longer lens tubes will be simulated to assess the utility of enabling individual movement of lenses to compensate for chromatic effects.
4.2 Uniform spectral output
A benefit of the LED-based spectral illuminator is the control of the power output for individual wavelengths. This allows the ability to implement uniform excitation illumination (flat spectral output) across a range of wavelengths. For this initial study, the optical transmission through the liquid light guide and microscope body resulted in a low illumination power at the microscope stage. Based on benchtop testing of the integrated system with power measurements made at the microscope stage, we determined that a uniform spectral output of 0.017 mW could be achieved, or that, alternatively, LEDs could be operated at maximum output ranging between 0.017–0.063 mW, depending on wavelength band. The proof-of-principle imaging results described here were performed using the maximum power output setting per wavelength band to maximize signal-to-noise characteristics of the acquired images. Post-acquisition correction was used to restore spectral image data back to a uniform spectral response (this same process used to correct spectral data acquired from the TFTF-based system) [34]. However, in future prototypes, if the optical power output available at the stage is increased, it may be advantageous to operate the system in a uniform spectral output mode and to remove the need for post-acquisition spectral correction while also increasing the useable lifetime of the LEDs as they would not be operated at 95% of maximal rated output.
4.3 Future system design objectives
The prototype LED-based system described in this research presented with reduced illumination throughput due to light losses when coupling to the LLG, likely a result of beam diameter and incident angle mismatch. This loss resulted in a lower photon budget when acquiring fluorescence images. In order to achieve a moderate SNR, the acquisition speed was reduced when compared to the TFTF-based system. Future work will focus on reducing the beam diameter for each LED. If necessary, the concave mirror and hence incident angle on the LLG may be adjusted. Based upon the LLG manufacturer specifications, a 3 mm beam diameter and ≤36° incident angle are needed to achieve optimal coupling. Meeting these parameters will provide increased illumination power, and a corresponding increase in available acquisition speed. An additional future objective is to modify the geometry of the system so as to allow for 32 excitation wavelength bands.
Of note, the LED-based spectral illuminator module described here features a streamlined and compact optical design. This design provides benefits in terms of production scalability and translatability to other imaging platforms. We anticipate that the spectral illumination source could be commercialized in a manner that is simple and moderately priced when manufactured at large-scale. In specific, the streamlined and modular nature of the light path would allow for hardware and cage system components to be manufactured using 3D printing additive techniques to minimize production costs. Similarly, optical components would have pre-defined placement within the assembly that would require minimal alignment or adjustment. In addition, the spectral illuminator design could be modified to allow more LEDs or wavelengths by increasing the mirror diameter and/or allowing for concentric rings of LEDs. It may also be possible to further increase the number of wavelength bands of the spectral illuminator by using band-pass filters, placed immediately after LEDs, to further select specific wavelength bands for illumination and when using wide bandwidth LEDs, to select more than one wavelength band from a specific model of LED. The modular spectral illumination source could also be translated to other platforms, such as high throughput imaging systems, microscope slide scanners, or endoscopes. Hence, the streamlined design presented here has potential for high-speed spectral imaging for fluorescence microscopy, as well as potentially a range of alternative imaging platforms.
Funding
Alabama Space Grant Consortium (NNH19ZHA001C); National Science Foundation (1725937); National Center for Advancing Translational Sciences (UL1TR001417); National Heart, Lung, and Blood Institute (P01HL066299); National Heart, Lung, and Blood Institute (R01HL137030); Economic Development Partnership of Alabama.
Acknowledgements
Optical design support was provided by Lambda Research Corp., TracePro Expert non-profit thesis version. Optical design support and component fabrication were contracted through ThorLabs, Inc.
Disclosures
Drs. Leavesley and Rich disclose financial interest in a start-up company, SpectraCyte LLC, founded to commercialize spectral imaging technologies.
Data availability
Data underlying the results in this paper are available in Ref. [60]
Supplemental document
See Supplement 1 for supporting content.
References
1. D. A. Landgrebe, “Multispectral land sensing: where from, where to?” IEEE Trans. Geosci. Remote Sensing 43(3), 414–421 (2005). [CrossRef]
2. R. A. Schowengerdt, Remote Sensing: Models and Methods for Image Processing (Elsevier, 2006).
3. J. R. Schott, Remote Sensing: The Image Chain Approach (Oxford University Press on Demand, 2007).
4. L. Wang, J. Jin, Z. Song, J. Wang, L. Zhang, T. U. Rehman, D. Ma, N. R. Carpenter, and M. R. Tuinstra, “LeafSpec: An accurate and portable hyperspectral corn leaf imager,” Comput. Electron. Agric. 169, 105209 (2020). [CrossRef]
5. R. Lu and Y.-R. Chen, “Hyperspectral imaging for safety inspection of food and agricultural products,” in (International Society for Optics and Photonics, 1999), pp. 121–133.
6. G. ElMasry and J. P. Wold, “High-Speed Assessment of Fat and Water Content Distribution in Fish Fillets Using Online Imaging Spectroscopy,” J. Agric. Food Chem. 56(17), 7672–7677 (2008). [CrossRef]
7. D.-W. Sun, Hyperspectral Imaging for Food Quality Analysis and Control (Elsevier, 2010).
8. G. ElMasry, D.-W. Sun, and P. Allen, “Near-infrared hyperspectral imaging for predicting colour, pH and tenderness of fresh beef,” J. Food Eng. 110(1), 127–140 (2012). [CrossRef]
9. D. Goltz, M. Attas, G. Young, E. Cloutis, and M. Bedynski, “Assessing stains on historical documents using hyperspectral imaging,” Journal of Cultural Heritage 11(1), 19–26 (2010). [CrossRef]
10. S. J. Kim, S. Zhuo, F. Deng, C.-W. Fu, and M. Brown, “Interactive visualization of hyperspectral images of historical documents,” IEEE Trans. Visual. Comput. Graphics 16(6), 1441–1448 (2010). [CrossRef]
11. Q. Li, X. He, Y. Wang, H. Liu, D. Xu, and F. Guo, “Review of spectral imaging technology in biomedical engineering: achievements and challenges,” J. Biomed. Opt. 18(10), 100901 (2013). [CrossRef]
12. P. F. Favreau, J. A. Deal, B. Harris, D. S. Weber, T. C. Rich, and S. J. Leavesley, “Label-free spectroscopic tissue characterization using fluorescence excitation-scanning spectral imaging,” J. Biophotonics 13, e201900183 (2020). [CrossRef]
13. B. Fei, “Hyperspectral imaging in medical applications,” in Data Handling in Science and Technology (Elsevier, 2020), Vol. 32, pp. 523–565.
14. G. Lu and B. Fei, “Medical hyperspectral imaging: a review,” J. Biomed. Opt. 19(1), 010901 (2014). [CrossRef]
15. C. M. Browning, J. Deal, S. Mayes, A. Arshad, T. C. Rich, and S. J. Leavesley, “Excitation-scanning hyperspectral video endoscopy: enhancing the light at the end of the tunnel,” Biomed. Opt. Express 12(1), 247 (2021). [CrossRef]
16. M. Ebner, E. Nabavi, J. Shapey, Y. Xie, F. Liebmann, J. M. Spirig, A. Hoch, M. Farshad, S. R. Saeed, R. Bradford, I. Yardley, S. Ourselin, A. D. Edwards, P. Führnstahl, and T. Vercauteren, “Intraoperative hyperspectral label-free imaging: from system design to first-in-patient translation,” J. Phys. D: Appl. Phys. 54(29), 294003 (2021). [CrossRef]
17. F. Cutrale, V. Trivedi, L. A. Trinh, C.-L. Chiu, J. M. Choi, M. S. Artiga, and S. E. Fraser, “Hyperspectral phasor analysis enables multiplexed 5D in vivo imaging,” Nat. Methods 14(2), 149–152 (2017). [CrossRef]
18. N. S. Annamdevula, R. Sweat, J. R. Griswold, K. Trinh, C. Hoffman, S. West, J. Deal, A. L. Britain, K. Jalink, T. C. Rich, and S. J. Leavesley, “Spectral imaging of FRET-based sensors reveals sustained cAMP gradients in three spatial dimensions,” Cytometry, Part A 93(10), 1029–1038 (2018). [CrossRef]
19. S. J. Leavesley, A. L. Britain, L. K. Cichon, V. O. Nikolaev, and T. C. Rich, “Assessing FRET using spectral techniques,” Cytometry 83, 10 (2013). [CrossRef]
20. S. Levy, C. D. Wilms, E. Brumer, J. Kahn, L. Pnueli, Y. Arava, J. Eilers, and D. Gitler, “SpRET: Highly Sensitive and Reliable Spectral Measurement of Absolute FRET Efficiency,” Microsc. Microanal. 17(2), 176–190 (2011). [CrossRef]
21. T. Zimmermann, J. Rietdorf, and R. Pepperkok, “FEBS Lett.,” 546, 87–92 (2003).
22. S. J. Leavesley, N. Annamdevula, J. Boni, S. Stocker, K. Grant, B. Troyanovsky, T. C. Rich, and D. F. Alvarez, “HyperSpectral imaging microscopy for identification and quantitative analysis of fluorescently-labeled cells in highly autofluorescent tissue,” J. Biophotonics 5(1), 67–84 (2012). [CrossRef]
23. J. R. Mansfield, K. W. Gossage, C. C. Hoyt, and R. M. Levenson, “Autofluorescence removal, multiplexing, and automated analysis methods for in-vivo fluorescence imaging,” J. Biomed. Opt. 10(4), 041207 (2005). [CrossRef]
24. L. Giannoni, F. Lange, M. Sajic, K. J. Smith, and I. Tachtsidis, “A Hyperspectral Imaging System for Mapping Haemoglobin and Cytochrome-c-Oxidase Concentration Changes in the Exposed Cerebral Cortex,” IEEE J. Sel. Top. Quantum Electron. 27(4), 1–11 (2021). [CrossRef]
25. K. Okubo, Y. Kitagawa, N. Hosokawa, M. Umezawa, M. Kamimura, T. Kamiya, N. Ohtani, and K. Soga, “Visualization of quantitative lipid distribution in mouse liver through near-infrared hyperspectral imaging,” Biomed. Opt. Express 12(2), 823 (2021). [CrossRef]
26. E. Felli, M. Al-Taher, T. Collins, A. Baiocchini, E. Felli, M. Barberio, G. M. Ettorre, D. Mutter, V. Lindner, A. Hostettler, S. Gioux, C. Schuster, J. Marescaux, and M. Diana, “Hyperspectral evaluation of hepatic oxygenation in a model of total vs. arterial liver ischaemia,” Sci. Rep. 10(1), 15441 (2020). [CrossRef]
27. Y. Garini, I. T. Young, and G. McNamara, “Spectral imaging: principles and applications,” Cytometry, Part A 69A(8), 735–747 (2006). [CrossRef]
28. M. J. Khan, H. S. Khan, A. Yousaf, K. Khurshid, and A. Abbas, “Modern Trends in Hyperspectral Image Analysis: A Review,” IEEE Access 6, 14118–14129 (2018). [CrossRef]
29. S. Ortega, H. Fabelo, D. K. Iakovidis, A. Koulaouzidis, and G. M. Callico, “Use of hyperspectral/multispectral imaging in gastroenterology. Shedding some–different–light into the dark,” J. Clin. Med. 8(1), 36 (2019). [CrossRef]
30. M. Halicek, H. Fabelo, S. Ortega, G. M. Callico, and B. Fei, “In-vivo and ex-vivo tissue analysis through hyperspectral imaging techniques: Revealing the invisible features of cancer,” Cancers 11(6), 756 (2019). [CrossRef]
31. T. E. Renkoski, U. Utzinger, and K. D. Hatch, “Wide-field spectral imaging of human ovary autofluorescence and oncologic diagnosis via previously collected probe data,” J. Biomed. Opt. 17(3), 036003 (2012). [CrossRef]
32. K. S. Park, D. U. Kim, J. Lee, G. H. Kim, and K. S. Chang, “Simultaneous multicolor imaging of wide-field epi-fluorescence microscopy with four-bucket detection,” Biomed. Opt. Express 7(6), 2285 (2016). [CrossRef]
33. P. Favreau, C. Hernandez, A. S. Lindsey, D. F. Alvarez, T. Rich, P. Prabhat, and S. J. Leavesley, “Thin-film tunable filters for hyperspectral fluorescence microscopy,” J. Biomed. Opt. 19(1), 011017 (2013). [CrossRef]
34. P. F. Favreau, C. Hernandez, T. Heaster, D. F. Alvarez, T. C. Rich, P. Prabhat, and S. J. Leavesley, “Excitation-scanning hyperspectral imaging microscope,” J. Biomed. Opt. 19(4), 046010 (2014). [CrossRef]
35. A. St-Georges-Robillard, M. Masse, M. Cahuzac, M. Strupler, B. Patra, A. M. Orimoto, J. Kendall-Dupont, B. Péant, A.-M. Mes-Masson, F. Leblond, and T. Gervais, “Fluorescence hyperspectral imaging for live monitoring of multiple spheroids in microfluidic chips,” Analyst 143(16), 3829–3840 (2018). [CrossRef]
36. H. Choi, D. Wadduwage, P. T. Matsudaira, and P. T. C. So, “Depth resolved hyperspectral imaging spectrometer based on structured light illumination and Fourier transform interferometry,” Biomed. Opt. Express 5(10), 3494 (2014). [CrossRef]
37. M. B. Sinclair, D. M. Haaland, J. A. Timlin, and H. D. Jones, “Hyperspectral confocal microscope,” Appl. Opt. 45(24), 6283–6291 (2006). [CrossRef]
38. W. F. Vermaas, J. A. Timlin, H. D. Jones, M. B. Sinclair, L. T. Nieman, S. W. Hamad, D. K. Melgaard, and D. M. Haaland, “In vivo hyperspectral confocal fluorescence imaging to determine pigment localization and distribution in cyanobacterial cells,” Proc. Natl. Acad. Sci. 105(10), 4050–4055 (2008). [CrossRef]
39. Y. Zhang, C. Haskins, M. Lopez-Cruzan, J. Zhang, V. E. Centonze, and B. Herman, “Detection of Mitochondrial Caspase Activity in Real Time In Situ in Live Cells,” Microsc. Microanal. 10(4), 442–448 (2004). [CrossRef]
40. R. M. Zucker, P. Rigby, I. Clements, W. Salmon, and M. Chua, “Reliability of confocal microscopy spectral imaging systems: Use of multispectral beads,” Cytometry 71A, 174–189 (2007). [CrossRef]
41. E. J. Baltussen, E. N. Kok, S. G. B. de Koning, J. Sanders, A. G. Aalbers, N. F. Kok, G. L. Beets, C. C. Flohil, S. C. Bruin, and K. F. Kuhlmann, “Hyperspectral imaging for tissue classification, a way toward smart laparoscopic colorectal surgery,” J. Biomed. Opt. 24(01), 1 (2019). [CrossRef]
42. K. Chen, R. Yan, L. Xiang, and K. Xu, “Excitation spectral microscopy for highly multiplexed fluorescence imaging and quantitative biosensing,” Light Sci Appl 10(1), 97 (2021). [CrossRef]
43. S. Lohumi, B.-K. Cho, and S. Hong, “LCTF-based multispectral fluorescence imaging: System development and potential for real-time foreign object detection in fresh-cut vegetable processing,” Computers and Electronics in Agriculture 180, 105912 (2021). [CrossRef]
44. P. Bartczak, M. Iso-Mustajarvi, H. Vrzakova, R. Bednarik, M. Fraunberg, and A.-P. Elomaa, “A Portable System for On-Site Medical Spectral Imaging: Pre-Clinical Development and Early Evaluation,” in 2018 IEEE 31st International Symposium on Computer-Based Medical Systems (CBMS) (2018), pp. 256–261.
45. C. Poudel and C. F. Kaminski, “Supercontinuum radiation in fluorescence microscopy and biomedical imaging applications,” J. Opt. Soc. Am. B 36(2), A139–A153 (2019). [CrossRef]
46. J. Luo, H. Zhang, E. Forsberg, S. Hou, S. Li, Z. Xu, X. Chen, X. Sun, and S. He, “Confocal hyperspectral microscopic imager for the detection and classification of individual microalgae,” Opt. Express 29(23), 37281 (2021). [CrossRef]
47. S. Zhang, J. Y. Xin Cheng, J. J. Chua, and M. Olivo, Characterization of Quantum Dots with Hyperspectral Fluorescence Microscopy for Multiplexed Optical Imaging of Biomolecules (Bioengineering, 2022).
48. “Fluorescence Microscopy | LED Illuminators,” https://www.coolled.com/.
49. U. Bradter, J. O’Connell, W. E. Kunin, C. W. H. Boffey, R. J. Ellis, and T. G. Benton, “Classifying grass-dominated habitats from remotely sensed data: The influence of spectral resolution, acquisition time and the vegetation classification system on accuracy and thematic resolution,” Sci. Total Environ. 711, 134584 (2020). [CrossRef]
50. N. S. Annamdevula, B. Sweat, P. Favreau, A. S. Lindsey, D. F. Alvarez, T. C. Rich, and S. J. Leavesley, “An approach for characterizing and comparing hyperspectral microscopy systems,” Sensors 13(7), 9267–9293 (2013). [CrossRef]
51. H. Shao, S. Li, S. C. Watkins, and A. Wells, “α-Actinin-4 Is Required for Amoeboid-type Invasiveness of Melanoma Cells,” J. Biol. Chem. 289(47), 32717–32728 (2014). [CrossRef]
52. P. Thakore, H. A. Pritchard, C. S. Griffin, E. Yamasaki, B. T. Drumm, C. Lane, K. M. Sanders, Y. Feng Earley, and S. Earley, “TRPML1 channels initiate Ca2 + sparks in vascular smooth muscle cells,” Sci. Signal. 13(637), eaba1015 (2020). [CrossRef]
53. V. L. Sutherland, J. A. Timlin, L. T. Nieman, J. F. Guzowski, M. K. Chawla, P. F. Worley, B. Roysam, B. L. McNaughton, M. B. Sinclair, and C. A. Barnes, “Advanced imaging of multiple mRNAs in brain tissue using a custom hyperspectral imager and multivariate curve resolution,” J. Neurosci. Methods 160(1), 144–148 (2007). [CrossRef]
54. A. M. Valm, S. Cohen, W. R. Legant, J. Melunis, U. Hershberg, E. Wait, A. R. Cohen, M. W. Davidson, E. Betzig, and J. Lippincott-Schwartz, “Applying systems-level spectral imaging and analysis to reveal the organelle interactome,” Nature 546(7656), 162–167 (2017). [CrossRef]
55. J. Deal, S. Mayes, C. Browning, S. Hill, P. Rider, C. Boudreaux, T. C. Rich, and S. J. Leavesley, “Identifying molecular contributors to autofluorescence of neoplastic and normal colon sections using excitation-scanning hyperspectral imaging,” J. Biomed. Opt. 24(02), 1 (2018). [CrossRef]
56. A. J. Bares, M. A. Mejooli, M. A. Pender, S. A. Leddon, S. Tilley, K. Lin, J. Dong, M. Kim, D. J. Fowell, N. Nishimura, and C. B. Schaffer, “Hyperspectral multiphoton microscopy for in vivo visualization of multiple, spectrally overlapped fluorescent labels,” Optica 7(11), 1587 (2020). [CrossRef]
57. S. J. Leavesley, B. Sweat, C. Abbott, P. Favreau, and T. C. Rich, “A theoretical-experimental methodology for assessing the sensitivity of biomedical spectral imaging platforms, assays, and analysis methods,” J. Biophotonics 11(1), e201600227 (2018). [CrossRef]
58. S. G. Mayes, S. A. Mayes, C. Browning, M. Parker, T. C. Rich, and S. J. Leavesley, “A spherical mirror-based illumination system for fluorescence excitation-scanning hyperspectral imaging,” in (International Society for Optics and Photonics, 2019), Vol. 10881, p. 108810N.
59. C. A. Wurm, D. Neumann, R. Schmidt, A. Egner, and S. Jakobs, “Sample Preparation for STED Microscopy,” in Live Cell Imaging, D. B. Papkovsky, ed., Methods in Molecular Biology (Humana Press, 2010), Vol. 591, pp. 185–199.
60. “BioImaging and BioSystems | University of South Alabama,” https://www.southalabama.edu/centers/bioimaging/.
