1. Introduction

Autofluorescence methods monitor the fluorescence response of endogenous molecules to report structural, biochemical, and functional alterations owing to pathological transformations [1]. In malignant transformations, the autofluorescence fingerprints are primarily modulated by changes in cellular metabolism that affect the balance and the optical properties of metabolic cofactors nicotinamide adenine dinucleotide (phosphate) (NAD(P)H) and flavins, most notably flavin mononucleotide (FMN) and flavin adenine dinucleotide (FAD). These changes are reflected in emission wavelength and lifetime characteristics [2–6]. In addition, structural changes occurring in the extracellular matrix (ECM), owing to malignant transformation or scarring, can be monitored by the collagen autofluorescence signal [7,8]. Monitoring the autofluorescence signatures of these endogenous molecules is thus crucial to identify and characterize pathological transformations and, specifically, to discriminate between benign and malignant lesions.

In this context, multispectral autofluorescence lifetime imaging and spectroscopy have emerged as powerful and versatile techniques for rapid and label-free characterization of cells and tissue dynamics [9–11]. Simultaneous acquisition of multiple spectral and temporal characteristics offers the ability to harness multiparametric information from which tissue composition, structure and function can be inferred, paving the way for enhanced diagnostics and therapeutic interventions. The multidimensionality of the autofluorescence readout confers great versatility that is depicted by the wide range of clinical applications where this technology has demonstrated potential or found utility [12–15], including in clinical settings in vivo, e.g. [16–20].

Most clinical autofluorescence lifetime instrumentation use similar approaches with respect to the optical design, irrespective of the data acquisition method: a single excitation wavelength, typically in the near-ultraviolet (UV) range, excites autofluorescence that is subsequently collected and separated by wavelength, so that the signals from key endogenous fluorophores are decoupled and simultaneously maximized in each spectral band. UV excitation is preferentially employed because it offers efficient excitation of the most relevant and abundant endogenous fluorophores, including collagens, NAD(P)H, and flavins [21]. However, the broad overlapping emission of tissue fluorophores makes it quite challenging to simply isolate their signal in narrow spectral windows. Specifically, NAD(P)H and flavins (FAD/FMN) are efficiently excited in the 350-390 nm range, but their emission strongly overlaps in the 500-600 nm region [21]. Since the NAD(P)H signal is typically much stronger than that of flavins under UV excitation, autofluorescence in the 500-600 nm emanates predominantly from NAD(P)H, with a smaller contribution from flavins [22]. Yet, this band is notoriously associated with flavin autofluorescence, even with excitation below 400 nm. Isolation of NAD(P)H and flavins signals can be achieved by using a second excitation laser emitting in a spectral region where flavins are efficiently excited, but NAD(P)H is not, for example at ∼445 nm. Fluorescence lifetime measurements of NAD(P)H and flavins at multiple excitation wavelengths (e.g. 355/375 nm and 445 nm) in a single platform have been previously demonstrated, most commonly in pre-clinical and ex vivo settings, where the acquisition settings and time can be more carefully adjusted and tuned according to requirements [12,23,24]. In vivo application of dual excitation is less frequent. In one hand, this can be explained by the lack of suitable ∼445 nm laser sources for autofluorescence lifetime measurements using the pulse sampling technique, which has been the most successful in clinical translation e.g. [17,18]. On the other hand, most time-resolved studies using multi-wavelength excitation employ Time-Correlated Single Photon Counting (TCSPC) acquisition, which was until recently considered impractical for widespread clinical deployment, owing to its intrinsic sensitivity and inoperability under bright illumination settings. Thus, it has been seldomly employed in vivo, e.g. [16,25].

Indeed, recent work demonstrated the feasibility of fiber based TCSPC imaging in real time and its suitability for deployment in clinical settings exploiting stroboscopic illumination [26,27]. Here, we aimed to expand this work and report a system exploiting dual excitation and multispectral detection to realize fiber-based multiparametric TCSPC imaging with real time processing and display. The optical setup was designed to isolate and maximize the fluorescence emission of NAD(P)H and flavins, using excitation at 375 nm and 445 nm. We demonstrate the feasibility and clinical viability of this approach in relevant clinical specimens of colorectal cancer and, more specifically, we report one case that clearly pinpoints the need for separating out the autofluorescence from these fluorophores. To further expand the specificity of this setup, in tandem with the autofluorescence acquisition we also carry out reflectance spectral measurements making use of the stroboscopic white light source that is used to illuminate the specimen. Similar to autofluorescence, reflectance spectroscopy delivers information from endogenous molecules, measuring the characteristic reflectance spectrum produced as light penetrates the tissue and interacts with scatterers and absorbers. As a broadband light source is used for such measurements (from 400–800 nm), deeper layers of the tissue can be probed by this technique compared to autofluorescence measurements. The reflectance/absorbance spectrum of a sample includes scattering and absorption information that can be used to estimate biochemical (e.g. hemoglobin concentration [28], oxygenation status [29], and cytochrome-c content [30]) and morphological (scatter size and shape) characteristics of the tissue, thus complementing the structural and functional information provided by autofluorescence measurements. The complementarity and viability of this bimodal setup is demonstrated in relevant samples, including an in vivo measurement of human skin tissue.

2. Materials and methods

2.1 Instrumentation

Optical and electronic schemes of the setup are illustrated in Fig. 1(A). The system consists of two ps-pulsed laser diodes that provide periodic excitation at 375 nm (BDS-SM-375, Becker and Hickl GmbH, Germany) and 445 nm (BDS-SM-445, Becker and Hickl GmbH). Both lasers are operated at 20 MHz and excite the sample alternately in 20 ms intervals using a pulse-group multiplexing scheme implemented through a GVD-120 board (GVD-120, Becker and Hickl GmbH), as illustrated in Fig. 1(B): when the 375 nm is on, the 445 nm laser is off, and vice-versa. Each group of pulses corresponds to a single fluorescence lifetime measurement, resulting in fluorescence lifetime acquisition rate of 50 Hz. Excitation light is delivered to the specimen via a single 300 µm excitation multimode fiber at the center of a custom optical fiber bundle (NA = 0.22, FiberTech Optica, Canada) that can be handheld by the operator and moved freely over the specimen during measurements. Autofluorescence signals emanating from the sample are collected by six collection fibers (200 µm core diameter) that are circularly arranged around the excitation fiber and passed through an ultrafast galvo-shutter (Model V-VM3, Optogama, Lithuania) that opens periodically and is synchronized with the white LEDs used for illuminating the sample. In this manner, the autofluorescence signal only reaches the detectors when the LEDs are turned off (see Fig. 1(B)). Autofluorescence is subsequently delivered to the detection system, consisting of a wavelength selection module that provides spectral resolution to the autofluorescence measurement and three photon counting hybrid detectors (HPM-100-40-CMOUNT, Becker and Hickl GmbH). The wavelength selection module comprises a set of dichroic mirrors and filters that determine the spectral collection range of each detector, as indicated in Fig. 1(A) and Table 1. These spectral ranges were selected to separate and maximize autofluorescence collection from endogenous fluorophores of interest, namely collagens, NAD(P)H, and flavins. The detectors are connected to a router (HRT-41, Becker and Hickl GmbH) that permits recording simultaneous autofluorescence intensity decays at the different spectral ranges using a single TCSPC acquisition card (SPC-130 EM, Becker and Hickl GmbH).

 

Fig. 1. (A) Schematic representation of the optical and electronic layouts of the instrument. Electronic trigger signals used for synchronization of all modules are represented by dashed lines. DM425, DM495, and DM593 indicate dichroic mirrors and corresponding cutoff wavelengths. F1-F3: band-pass filters (see Table 1 for spectral bandwidth). D1-D3: hybrid detectors. (B) Timing diagram of the real-time acquisition. Cameras, LEDs, and spectrometer are triggered simultaneously at 50 Hz. Lasers are multiplexed at 50 Hz for sequential excitation at different wavelengths. TCSPC measurements are carried out at 50 Hz (25 Hz for each excitation wavelength) when the shutter is open. (C) Representative autofluorescence intensity decays and absorbance spectrum obtained in a single measurement with our instrument, from in vivo human skin tissue. Integration times were 15 ms for autofluorescence measurements at each excitation wavelength and 3 ms for reflectance measurements.

Download Full Size | PDF

Table 1. Optical configuration of the autofluorescence lifetime setup indicating spectral range of each detection channel and most common endogenous fluorophores detected in each excitation-collection arrangement

View Table

At the sample end, our setup comprises a USB camera (FFY-U3-16S2C-C, FLIR, USA) that records the measurements at 50 Hz and a set of LEDs (MNWHL4, Thorlabs, USA). The LEDs provide adequate and uniform illumination of the sample during measurements without interfering with the autofluorescence acquisition. This is achieved through stroboscopic illumination at 50 Hz [26]: the autofluorescence acquisition is enabled only when the LEDs are off and disabled when the LEDs are on. The shutter, being closed while the LEDs are turned on, prevent direct bright light from entering the autofluorescence detection system during this period. The LEDs are turned on for 3 ms in every 20 ms period, which is sufficient to provide adequate illumination of the sample during the optical acquisition.

The white light emitted by the LEDs can also be a useful source of spectroscopic information through the measurement of the characteristic light spectra reflected by the sample. In this regard, reflectance spectral measurements are carried out during the “on” period of the LEDs (i.e. acquisition time of 3 ms), prior to every single TCSPC acquisition. Reflected white light spectra are collected by a single 200 µm optical fiber and passed through a 475 nm long-pass filter mounted on an inline filter holder (SILFH-3, Sarspec, Portugal) before being directed to a high-speed spectrometer (Ocean FX-VIS-NIR, Ocean Insight, USA) that records the characteristic reflectance spectra at 50 Hz. This optical configuration and acquisition window permitted collection of reflectance spectral measurements with high signal-to-noise ratio and dynamic range. Figure 1(C) shows a representative reflectance spectral measurement (in units of absorbance) of human skin in vivo.

Naturally, the complexity of this setup mandates careful synchronization between all modules. Particular attention must be given to the synchronization of the autofluorescence detection system with the white light illumination of the sample, to avoid overexposure of the detectors in the presence of direct bright light. In order to meet the timing constraints of our implementation and achieve robust synchronization between components, a dedicated synchronization unit was implemented in an Arduino Due (Arduino S.r.l., Italy). The master clock of the system was provided by the laser multiplexing trigger signal, generated by the GVD-120 unit, and adjusted to 50 Hz according to our requirements. The signal was used to trigger a sequence of events within the Arduino and synchronize the lasers with the camera, LEDs, shutter, TCSPC acquisition and spectrometer, strictly following the timing diagram of Fig. 1(B).

2.2 Image acquisition, processing, and visualization

The system was designed to operate as a single-point device, i.e. optical measurements through the fiber bundle do not carry any spatial information. Spatial information is provided by a USB camera that records the measurements (800 × 600 pixels) and allows tracking the position of the fiber as it is moved over the tissue. The camera is triggered simultaneously with the LEDs so that each frame is captured when the sample is under bright white illumination (see Fig. 1(B)). Unlike previous implementations of single-point fiber-based imaging devices employing additional light sources as guiding beams (e.g. [26,31]), we take advantage of the laser multiplexing scheme to determine the location of the measurements for each set of frames. The workflow is best illustrated in Fig. 2. Briefly, autofluorescence excitation alternates at 50 Hz between 375 nm and 445 nm. This is synchronized with the camera, which is also triggered at 50 Hz. Therefore, for every two consecutive frames of the camera, one frame is captured with 375 nm excitation and another is captured with 445 nm excitation. Since 445 nm light is visible to the camera and 375 nm is not, the only difference between two consecutive frames is the excitation spot, provided that the movement of the fiber is slow enough with respect to the camera frame rate of 50 Hz (see Fig. 2(A)). This is particularly evident in the blue channel of the RGB images (see Fig. 2(B)). Conveniently, the 445 nm spot is visible in the blue channel but not in the green channel of the RGB image. Direct subtraction of the two channels therefore eliminates most of the sample background, including specular reflections of the white light onto the sample that result in bright (white) spots in the RGB images (see Fig. 2(C)). At this point, the resulting images (processed over two consecutive frames) are subtracted to eliminate the remaining background and pinpoint the excitation spot. Intensity thresholding and centroid detection of the resulting binary mask are used to determine the x and y coordinates of the center of the excitation spot (Fig. 2(D)). Finally, the measurement region is defined using a fixed radius of 5 pixels, which corresponds to ∼2 mm of diameter (Fig. 2(E)). We chose to have a fixed rather than variable spot diameter to maintain the simplicity of the implementation and because it is just an approximation to the real case. Obviously, the spot size can vary significantly during one acquisition, depending on sample morphology, scattering and absorption, and excitation geometry. We believe that a spot diameter of 10 pixels is a good approximation for most measurements, when the fiber is approximately 5 mm over the tissue and with a relatively constant angle of incidence (∼60°).

 

Fig. 2. Workflow for real-time determination of the location of optical measurements. (A) Set of two consecutive frames, one captured with 445 nm excitation (visible), and another captured with 375 nm excitation (not visible). (B) Blue and green channels of each frame, where the excitation spot created by the 445 nm light is well visible. (C) Subtraction between the blue and green channels reduces the number of artefacts caused by specular reflections of the white light onto the tissue. (D) The excitation spot is segmented by subtraction of the two processed frames along followed by intensity thresholding. Centroid detection of the resulting binary mask permits determination of the center of the spot. (E) A circular region with 10 pixels of diameter defines the location of a single measurement.

Download Full Size | PDF

With this workflow, the measurement location is updated for each collection of two frames, at 25 Hz. Since every set of two consecutive autofluorescence measurements (at 375 nm and 445 nm excitation) is univocally linked to a single measurement location, autofluorescence lifetime maps can be generated at 25 Hz, thereby offering immediate feedback of the measurement. In regions where the measurement location overlaps, optical data are averaged out.

2.3 Sample collection and preparation

Ex vivo measurements of colon and rectal cancer. To demonstrate the suitability of our system to report endogenous contrast in biological tissues, measurements were carried out in sigmoid colon and rectal cancer surgical specimens. Immediately upon surgical resection, the samples were transported to the Pathology Service laboratory to be processed according to the standard procedures. Samples were carefully opened to expose the inner mucosa and the whole pieces were used for optical imaging. Both samples were thoroughly washed with running water, rinsed with Phosphate Buffer Saline (PBS) solution, and dried with absorbent paper before being positioned for optical measurements in a dedicated sample holder. Measurements were carried out within 30 minutes of resection and completed in under 3 minutes. Following imaging, the samples were returned to the Pathology Service for standard processing. Identification of malignant lesions was based on macroscopic assessment by experienced pathologists and confirmation by microscopic examination of the most representative histology slide with haematoxylin and eosin (H&E) staining. Specimens were obtained under the Champalimaud Foundation Biobank Informed Consent, clinical protocol 2021020203 approved by the Champalimaud Foundation Ethics Committee.

In vivo measurements of human skin. The suitability of our system to carry out autofluorescence and reflectance measurements in vivo was demonstrated by imaging human skin from the wrist of a voluntary. The volar wrist was positioned upwards, facing the camera, so that the medial area was centered in the field of view. Excitation light intensity was adjusted to maximize photon collection while ensuring that the in vivo autofluorescence measurement was carried out safely, i.e. without causing damage to the tissue. The average laser output power was kept below 5 µW and 35 µW for 375 nm and 445 nm excitation, respectively. If we consider the unlikely scenario where the fiber would be parked for 10 seconds and in direct contact with the skin, the dose delivered would be D₃₇₅ = 0.071 J/cm² and D₄₄₅ = 0.495 J/cm², which are below the Maximum Permissible Exposure (MPE) for the skin for 10 second of accidental exposure (MPE₃₇₅ = 0.996 J/cm², MPE₄₄₅ = 1.95 J/cm²). Realistically, because the fiber probe is constantly moved during measurements, the actual exposure time at any given location is less than 1 second. In addition, we try to maintain the fiber at ∼5 mm from the skin, which drastically reduces the dose delivered by our system.

2.4 Data processing

Autofluorescence data. Autofluorescence intensity decays for all five channels were processed immediately upon collection using the phasor approach [32]. Fluorescence decays with less than 50 photon counts were discarded from the analysis. Briefly, autofluorescence decay curves were transformed to the Fourier space according to Eqs. 1 and 2

The instrument response function (IRF) was measured for all channels (by removing band-pass filters) using excitation light scattered off a reflective surface. Day-to-day calibrations were realized independently for both excitation wavelengths using reference slides with measured fluorescence lifetimes of 0.9 ns and 3.6 ns, at 375 nm and 445 nm excitation, respectively. Fluorescence lifetime measurements were further validated using POPOP (τ = 1.36 ns in ethanol [33]) and Coumarin 6 (τ = 2.72 ns in ethanol [34]).

The relative autofluorescence intensity in each detection channel was calculated relative to the total autofluorescence signal collected for the corresponding excitation wavelength and normalized to the spectral bandwidth of the detector. Normalized optical redox ratio (RR) was calculated as indicated in Eq. 4, where F₂ and F₅ denote autofluorescence intensity in detection channels 2 and 5, respectively.

3. Results

3.1 Ex vivo measurement of rectal cancer

A freshly resected rectal cancer surgical specimen (see Fig. 3(A)) was measured to demonstrate the suitability of our system to report endogenous contrast from tissues in a relevant clinical application. The specimen was characterized by a large ulcerating lesion with fibrinous exudate and mucosal necrosis (delineated by the white dashed line) and a smaller ulcer in its near vicinity (red dashed line), which was not immediately visible at the naked eye and was identified through palpation at the time of the measurement. Tumor cells were later identified in both lesions by histopathology. In the large lesion, tumor was located superficially in the edges and underneath the exudate at its center. The smaller lesion was also confirmed to be an adenocarcinoma. The tissue surrounding the larger lesion was significantly firmer to touch than remote regions and bloodier at the surface. The specimen also comprised areas previously tattooed with Chinese ink, which is used to identify areas adjacent to suspicious lesions during endoscopy (see Fig. 3(J), white arrows). Autofluorescence and reflectance data for this specimen are presented in Fig. 3. The optical acquisition was completed in 2 minutes and 35 seconds and covered an estimated area of ∼20 cm² around the necrotic lesion, suspicious tumor, and remote tissue (see Fig. 3(B)), which was presumed and later confirmed to be normal (non-lesional) tissue. Tumor and necrotic area can be clearly identified by the autofluorescence lifetime data (panels C1-C5 of Fig. 3). In particular, the small lesion that is barely visible in the white light image (Fig. 3(A)) is clearly demarcated from surrounding tissues, despite its small size (less than 3 mm in diameter). This is particularly evident in detection channel 1 (panel C1), but still visible in the other detection channels. Interestingly, the edges of the larger lesion yield shorter lifetime than its center, in detection channels 2-4 (panels C2-C4), which is in line with the observations from histopathology. Both tumor and necrosis yield shorter autofluorescence lifetimes in comparison with normal tissue. In the immediate vicinity of the lesions, we measured a small increase in fluorescence lifetime that is more evident in channels 3 and 5 (panels C3 and C5, respectively), which may result from increased tissue stiffness and inflammatory response.

 

Fig. 3. Ex vivo autofluorescence lifetime and reflectance measurement of rectal cancer. (A) White light image of surgical specimen. Area limited by dashed line corresponds to analyzed region. (B) Magnified white light image of the specimen, delineating ulcerating lesions (white and red dashed lines). Black line indicates measurement boundaries. Arrows indicate areas of interest: tumor (red), normal (green), perilesional (orange), and necrosis (cyan). (C1-C5) Autofluorescence lifetime and (D1-D5) normalized autofluorescence intensity maps for each detection channel. (E) Normalized optical redox ratio. (F) Normalized absorbance spectra at locations of interest, as indicated in panel B. (G) Absorbance map at 630 nm normalized to the absorbance at 540 nm. (H-J) Integrated absorbance over three spectral ranges of interest. Scale bar = 10 mm.

Download Full Size | PDF

With respect to the autofluorescence intensity data (panels D1-D5), differences between regions are more subtle, and mostly confined to detection channels 2 and 3 (D2 and D3, respectively). These data indicate a shift of the autofluorescence emission towards longer wavelengths in the tumor with respect to normal tissue, which can be interpreted as a shift towards oxidative environment, considering that flavins are contributing slightly to channel 3 autofluorescence but not to channel 2. This is further evidenced in the optical redox ratio map (Fig. 3(E)), which indicates a relative decrease of NAD(P)H autofluorescence intensity in the tumor and perilesional areas.

In general, tissue absorbance is dominated by hemoglobin absorbance. The absorbance spectra obtained from regions of interest in the tissue peak at around ∼576 nm (Fig. 3(F)), which coincides with an absorption peak of oxyhemoglobin [35]. The ∼540 nm absorbance peak of oxyhemoglobin is also well visible as a shoulder in the absorbance curve of perilesional tissue (Fig. 3(F), orange curves) but is much less pronounced in normal tissue and tumor. This is probably due to the increased blood content in the perilesional region, as is visible in the white light image (Fig. 3(A)). Interestingly, the characteristic double peak profile of oxyhemoglobin with an absorption valley at 560 nm is not visible in any of the curves. Our data indicate higher tissue absorbance in the 550-560 nm spectral range with respect to the absorbance measured at 540 nm, which could be attributed to a significant contribution of deoxyhemoglobin, with its characteristic peak at ∼555 nm [35]. This is expected since the tissue was deprived of oxygen for over 30 minutes prior to the measurement. Thus, the measured absorbance spectra are a result of competing contributions from oxygenated and deoxygenated hemoglobin, which is most visible in the 500-600 nm spectral range.

Differences in absorbance are most pronounced beyond ∼580 nm. Where differences are maximized (∼630 nm), the absorbance map normalized to the absorbance at 540 nm clearly highlights a region of increased blood content around the tumor and necrotic lesion (Fig. 3(G), in green). Interestingly, normal tissue and tumor have similar absorbance throughout the measured wavelength range, and thus cannot be distinguished in the absorbance maps (panels H-J).

3.2 Ex vivo measurement of sigmoid colon cancer

Autofluorescence lifetime and reflectance data obtained from a sigmoid colon cancer specimen are presented in Fig. 4. The tumor presented a 1 cm long ulcerated lesion that was clearly visible at the naked eye (see Fig. 4(A), red square). Unlike the previous measurement where we mapped the entire region surrounding the tumor, in this acquisition we opted to map the tumor and remote normal tissue separately, as shown in Fig. 4. The total acquisition time was 59 seconds. Interestingly, we found no differences between tumor and normal tissues in autofluorescence data collected with 375 nm excitation (channels 1-3). This is clear from the autofluorescence lifetime maps (panels B to D) and in the corresponding average lifetimes measured within the regions of interest, see Fig. 4(H). However, there are clear differences between tumor and normal tissue in the autofluorescence data collected with 445 nm excitation (channels 4 and 5). In channel 4, we measured a longer autofluorescence lifetime in tumor (2.96 ± 0.12 ns) compared to normal tissue (2.58 ± 0.08 ns). The opposite trend was observed in channel 5, with normal tissue yielding a longer lifetime (1.96 ± 0.08 ns) compared to tumor (1.45 ± 0.08 ns). Differences between normal and tumor tissues are also observed in the redox ratio map (Fig. 4(G)). Interestingly, data indicate a shift towards glycolytic environment (relative increase of NAD(P)H autofluorescence), which is opposite from observations in Fig. 3.

 

Fig. 4. Ex vivo autofluorescence lifetime and reflectance measurement of a sigmoid colon adenocarcinoma. Endogenous contrast between normal tissue and tumor is more evident in channels 4 and 5 (excitation at 445 nm) compared to the detection channels at 375 nm excitation. (A) White light image of the specimen. (B-F) Autofluorescence lifetime maps in detection channels 1-5, respectively. (G) Optical redox ratio map. (H) Average autofluorescence lifetimes and (I) absorbance spectra measured from normal and tumor regions, as delineated in panel A [ROI_normal = 484 (22 × 22) pixels, ROI_tumor = 324 (18 × 18) pixels]. In panel I, solid lines represent the average spectrum and dotted lines the corresponding standard deviation.

Download Full Size | PDF

With respect to the absorbance spectra (see Fig. 4(I), average spectra obtained from ROIs depicted in panel A), data indicate increased blood content in the tumor, probably as a result of the ulcer, which is evidenced by the higher relative absorbance in the 500-600 nm with respect to longer wavelengths (Fig. 4(I), red curve), in comparison with the absorbance measured in normal tissue (blue curve). As expected, and in line with the previous measurement, the absorbance spectra of the tissue are dominated by hemoglobin absorbance, with the characteristic peak at ∼576 nm from its oxygenated form and a shoulder at ∼555 nm, most likely from deoxyhemoglobin contribution.

3.3 In vivo measurement of human skin

Human skin from the wrist was imaged in vivo to demonstrate the viability of our setup in a clinically relevant scenario and to further highlight the complementarity of autofluorescence lifetime and reflectance measurements in a living tissue. Imaged skin was smooth (no wrinkles or visible lesions) and had homogenous pigmentation. In vivo autofluorescence lifetime and reflectance maps of skin are shown in Fig. 5. An area of approximately 16 cm² was mapped in under 45 seconds. We obtained relatively homogenous autofluorescence lifetime maps in all detection channels (see Fig. 5(A-E)), with average lifetimes of 2.87 ± 0.09 ns, 3.28 ± 0.07 ns, 3.39 ± 0.06 ns, 4.04 ± 0.09 ns, and 3.61 ± 0.09 ns, for channels 1 to 5, respectively, measured in the region of interest delineated in panel A. Despite the relative homogeneity, we observed a slight decrease of 100-300 ps in the measured fluorescence lifetimes across the map (from right to left), independent of variations in autofluorescence intensity, which we tentatively attribute to gradual changes in tissue structure across the measured region. This observation is accompanied by a general increase in tissue absorbance (from right to left), particularly at shorter wavelengths, as shown in Fig. 5(K-M).

 

Fig. 5. In vivo measurement of human skin tissue showing (A-E) autofluorescence lifetime and (F-J) absorbance maps for different spectral ranges. Panel F also shows normalized absorbance curves of skin in regions indicated by the pink and yellow arrows, the latter pointing to the location of the blood vessel underneath the epidermis. Panels (K-O) show absorbance profiles for each reflectance map (F-J, respectively) along the dashed line depicted in panel J. The autofluorescence lifetime maps show relatively homogenous signatures throughout the measured region. Blood vessels underneath the epidermis (indicated by arrows in panels J and O) are clearly visible in the orange and red absorbance maps (panels I and J, respectively). Scale bars = 10 mm.

Download Full Size | PDF

The absorption spectrum of human skin is modulated by the contributions of many chromophores, including melanin, collagen, or hemoglobin, and their relative proportions in the tissue [36]. In particular, melanin is found in the epidermis and absorbs strongly in the short wavelength range of the visible spectrum [37,38]. Incident photons are thus preferentially absorbed by the outermost layers and have poor penetration beyond the epidermis [36]. Indeed, our spectral data indicate increased absorbance in the short wavelength range of the spectrum (see yellow and purple curves in subpanel of Fig. 5(F)). Data are also suggestive of the significant contribution of other chromophores beyond hemoglobin, since its characteristic spectral shape in the 540-580 nm band is not visible. Absorbance in this spectral region is relatively uniform across the measured region, as demonstrated in the absorbance maps (Fig. 5(F) and 5(G)). Conversely, long wavelength photons can harness information from deeper layers, including vascular morphology. Our data clearly delineates the median blood vessels of the wrist from surrounding tissues (see Fig. 5(I) and 5(J) and accompanying line profiles N and O).

4. Discussion

Autofluorescence lifetime and reflectance spectroscopy are label-free modalities that have long been exploited for characterization of biochemical alterations in tissues in many applications. Reflectance spectroscopy measures the spectral response of the tissue produced by the interaction of light with scatterers and absorbers, such as hemoglobin or melanin, offering non-invasive probing of vascular morphology, oxygenation, or pigmentation. On the other hand, autofluorescence lifetime measurements of key endogenous fluorophores can report on tissue structural and metabolic alterations. The two techniques can thus offer complementary information and, together, their implementation in clinical settings could be of great value for many applications. Indeed, several studies reported the application of combined time-resolved spectrofluorimetry and diffuse reflectance spectroscopy in clinical context, including for tumor identification and margin assessment [16,39–42]. Here, we presented a bimodal setup that combines autofluorescence and reflectance measurements to realize multidimensional fiber-based optical imaging of tissues in a single-point device. However, unlike traditional diffuse reflectance systems where illumination and collection fibers are within the same probe, in our setup, broadband illumination is provided by LEDs positioned at approximately 40 cm from the sample. Hence, the collected reflectance signals originate from both specular and diffuse reflections, particularly from the most superficial layers. This contrasts with traditional diffuse reflectance setups, in which the probed depth can be “tuned” by adjusting the separation between illumination and collection fibers [43]. In this context, the reflectance information collected by our setup is equivalent to that obtained by a hyperspectral camera, only at a fraction of the cost and with higher spectral resolution.

With respect to the autofluorescence lifetime setup, the work presented here improves previous implementations of fast fiber based TCSPC imaging [26,27]. A key development was the integration of a second laser source (445 nm) to realize quasi-simultaneous dual excitation of tissue autofluorescence in an imaging platform. The 445 nm laser source adds dimensionality and specificity to the autofluorescence measurement by optimizing fluorescence excitation and collection from key endogenous fluorophores, namely FMN and FAD. This is a significant improvement to similar autofluorescence lifetime imaging systems that employ a single excitation wavelength, typically in the (near-)UV region, as dual excitation at 375 nm and 445 nm permits complete decoupling of the NAD(P)H and flavins autofluorescence signals (channels 2 and 5 of our system, respectively). The exploitation of TCSPC measurements at multiple excitation and collection wavelengths in a single platform has been previously demonstrated, e.g. [12,16,24,25]. However, to the best of our knowledge, this is the first time where such configuration is implemented in a clinically viable TCSPC imaging system. We note that a similar device was recently introduced by Serafino and Jo [44] yet making use of the frequency domain method for fluorescence lifetime acquisition. The benefit of using a second excitation wavelength that complements excitation at 375 nm was best illustrated in the ex vivo measurement of sigmoid colon cancer (see Fig. 4). Here, endogenous contrast between tumor and normal tissue is reported in the autofluorescence data excited with 445 nm excitation, but not with 375 nm. Clearly, an instrument employing single excitation at 375 nm would not offer the most accurate characterization of this lesion. 375 nm light preferentially excites collagens, NAD(P)H, and, to a lesser extent flavins. At this excitation wavelength, NAD(P)H emission strongly overlaps with that of flavins, particularly in the 500-600 nm band, and therefore it is not possible to completely decouple the autofluorescence signal from these molecules [21]. This is evident in our ex vivo data, where we measured similar autofluorescence lifetimes in channels 2 and 3 (see Figs. 3 and 4), suggesting that the same fluorophores are contributing to the signal. For this reason, excitation at 445 nm adds specificity to the measurement as flavins are most efficiently excited at this wavelength, while NAD(P)H is not. In practice, our data demonstrate that flavin autofluorescence may report contrast where NAD(P)H does not, highlighting the benefits of a dual excitation autofluorescence lifetime system designed to specifically target these key endogenous fluorophores. We note that this autofluorescence pattern is not exclusive to this sample and was observed in other colorectal cancer specimens. This topic will be subject of further discussion in a follow-up publication.

The 445 nm excitation spot was also used as an optical marker on the sample to identify the approximate location of the measurement at any given time point during the acquisition. The use of a guiding beam is not novel [14,26,31], including at 445 nm. The novelty here is that the 445 nm light is also (and primarily) used to obtain time-resolved spectroscopic information from the sample. Necessarily, the addition of the second laser source comes at the cost of speed: autofluorescence lifetime maps for each excitation wavelength are generated at 25 Hz instead of 50 Hz as in the original design [26]. In practice, this means that consecutive measurements are more spaced out, resulting in less smooth optical maps and lower resolution [31]. To some extent, this can be compensated by reducing the fiber scanning speed to increase sampling in areas of interest and improve delineation of structures, e.g. tumoral margins. On the downside, reducing scanning speed for the whole sample may also lead to long acquisition times that may not be compatible with clinical implementation, even in ex vivo settings. There is thus a trade-off between fiber scanning speed and time required to map the entire specimen [26,31,45]. The ideal scanning speed therefore depends on multiple factors, including size of the specimen, size of the lesion, type of lesion, and time available for measurements. We also note that while the decrease in measurement rate is significant, 25 Hz is within the video rate range and feedback of the measurements is still transmitted in real time. Moreover, this scheme allowed us to maintain the autofluorescence integration time at 15 ms, which is long enough to acquire autofluorescence decays with sufficient signal-to-noise ratio, as previously demonstrated [26].

Autofluorescence lifetime and reflectance data obtained from ex vivo surgical specimens demonstrate the viability of our setup and its suitability to report endogenous contrast. In the rectal cancer specimen, autofluorescence data clearly delineates the tumor and necrotic lesion from the surrounding normal tissue. Differences are evident in all detection channels, particularly in the lifetime parameters (Fig. 3, C1-5). Interestingly, based on this sample alone, tissue autofluorescence appears to be relatively unaffected by Chinese ink, although further measurements are necessary to confirm this hypothesis. A key highlight of these measurements is the delineation of a small region (∼3 mm in diameter) of adenocarcinoma that was easily missed at the naked eye and was only identified through palpation (Fig. 3(B), red arrow). This result clearly demonstrates the benefits of employing a label-free optical method for quick assessment of suspicious tissues. In addition to the contrast provided by autofluorescence data, reflectance measurements also highlight the area around the necrosis and the tumor, owing to the increased contribution of hemoglobin to the measured absorbance spectrum (see Fig. 3(F) and 3(G)). In general, absorbance spectra of the tissue are dominated by hemoglobin absorbance, with competing contributions from its oxygenated and deoxygenated forms. This is expected since at the time of measurements tissue had been deprived of oxygen for some time: while optical measurements were realized within 30 minutes of resection, vascular ligation and consequent restriction of blood supply occurred at an earlier stage of the surgery.

In vivo measurements of human skin served to demonstrate the feasibility of TCSPC-based measurements and the operation of our setup in a scenario that more closely mimics the real-world clinical environment, particularly with respect to the lighting conditions. The measured autofluorescence lifetimes are in general agreement with previous observations of normal human skin tissue [16,25]. The field of view was brightly illuminated throughout the acquisition by the set of LEDs mounted in the dedicated measurement platform, and according to the stroboscopic method that was previously described [26]. Thus, unlike other optical platforms, our system does not require the room to be in complete darkness during measurements. Indeed, the setup comprising the camera and the ensemble of LEDs could be replicated and scaled up to provide dedicated lighting in a clinical setting (for example in an ambulatory surgical room or medical office) and, in this way, take advantage of the sensitivity and specificity offered by the TCSPC method. We believe that our setup could find wide applicability in these settings, for fast and non-invasive assessment of lesions, particularly in skin applications.

The complementarity of autofluorescence and reflectance measurements was best demonstrated in vivo, as the endogenous signals from well-oxygenated blood are more prominent, thus highlighting the vascular structures underneath the epidermis (see Fig. 5). While the penetration depth of 375 nm and 445 nm light is typically limited to less than 400-500 µm in biological tissues [36,46,47], it is unclear whether data collected within this region could offer insights of metabolic or structural changes occurring in deeper layers of the tissue. To some extent, white light reflectance measurements can probe the tissue beyond autofluorescence, thus harnessing label-free information from deeper regions. However, it is still unclear how deep we can probe the tissue using both techniques and this illumination-collection configuration. Penetration depth of our measurements will be assessed in future studies, exploring realistic tissue phantoms and ex vivo specimens.

4.1 Limitations

One of the major limitations of this work is the relatively high cost of the instrumentation, which is still a major barrier to clinical deployment. The component cost of this instrument considering only the major elements (i.e. two lasers, three hybrid detectors, TCSPC card, shutter) can ascend to €70,000. To this we still need to add the cost of other parts such as optical components (filters, mirrors, mounts, etc.), fiber optic probes, camera, LEDs, small electronic parts and drivers, or computer, which can bring the total cost to over €100,000. This figure could be significantly reduced through mass production. Other ways of mitigating the costs include tailoring the system to specific clinical applications, which would reduce the number components (e.g. one or two detectors may be sufficient in some applications) or replacing high-end instrumentation (lasers, detectors, and TCSPC acquisition card) by cheaper alternatives. Either way, these changes would likely impact the sensitivity and specificity of the autofluorescence detection, and potentially limit the clinical utility of the instrument.

4.2 Future work

Future work will focus on the routine implementation of this setup in clinical settings. In part, this will entail validation of optical measurements in relevant clinical applications. To this end, work is already underway aiming to demonstrate the clinical utility of autofluorescence and reflectance measurements in the characterization of lesions of the digestive tract. In parallel, we will focus our efforts on the integration of this system in the clinical workflow. In one hand, we envisage setting up this system in a dedicated room for quick assessment of skin lesions in ambulatory settings. On the other hand, we will work on the endoscopic integration of our system. This is paramount and will open the door to new fields where combined autofluorescence lifetime and reflectance measurements can find utility.

5. Conclusions

We reported a bimodal setup combining dual excitation multispectral time-resolved autofluorescence and reflectance spectroscopy for real time multiparametric optical mapping of tissues. We demonstrated the feasibility, practicality, and viability of this system in clinically relevant applications. The two techniques provide complementary label-free information that could aid clinical decision making. In the short term, we envisage deployment of this setup in ambulatory settings, offering quantitative and objective assessment of lesion in open procedures or endoscopically.