2.1.

Time-Resolved Fluorescence Measurements on Tissue

In vivo time-resolved fluorescence measurements have come into prominence only recently and a limited number of studies have been reported in the literature, which is summarized in Table 1 . For example, Pradhan ¹⁶ used fluorescence lifetime measurements to study breast tissue. They made use of laser pulses of $100\mathrm{fs}$ pulse width at $310\mathrm{nm}$ and a streak camera to obtain a 2-D map of the fluorescence lifetimes. It was shown that time-resolved fluorescence measurements at an emission wavelength around $340\mathrm{nm}$ can be used to distinguish malignant tumors from nonmalignant breast tissues.

Table 1

Experimental studies of time-resolved fluorescence spectroscopy on human tissues.

Fluorescence lifetime measurements have also been used to study tissue during endoscopy. Glanzmann ¹⁷ described a time-resolved fluorescence instrument based on pumped laser excitation sources and a streak camera coupled to a spectrograph as a detector. The instrument was used to measure the fluorescence decays of endogenous fluorophores and of $\delta$ -aminolevulinic acid (ALA)-induced protoporphyrin IX (PPIX) in an excised human bladder with carcinoma. The authors were able to correctly distinguish healthy tissue, which appeared abnormal in steady-state spectroscopy, from cancerous tissue. The instrument was then used to perform in situ characterization during endoscopy to obtain time-resolved spectra of tissue fluorescence of human bladder, bronchi, and esophagus.¹⁸ Further studies were conducted by the same group¹⁹ to test the viability of fluorescence lifetime as a new contrast parameter between normal and malignant tissue in the bronchi. However, they were unable to obtain a contrast in the fluorescence lifetime or the spectrum between normal and moderately dysplastic tissue in the spectral region between 510 and $650\mathrm{nm}$ . Mizeret ²⁰ also studied the feasibility of performing endoscopic fluorescence lifetime measurements using frequency modulation techniques in real time in a clinical setting. The instrument made use of an excitation source at $514.5\mathrm{nm}$ and a detection unit consisting of CCD units with image intensifier tubes. Fluorescence lifetime values for healthy bronchial tissue were reported. Pfefer ²¹ also performed in vivo fluorescence lifetime and spectral measurements on patients undergoing routine endoscopic surveillance for Barrett’s esophagus. A time-domain instrument consisting of a laser pulsing at $337\mathrm{nm}$ with a pulse width of $3\mathrm{ns}$ and an APD coupled to an oscilloscope was used in the experiments. The excitation wavelengths used were 337 and $400\mathrm{nm}$ and the emission was collected around $550\mathrm{nm}$ . Results showed that low-risk tissue and high-grade dysplasia could not be differentiated based on the observed data.

Furthermore, Schweitzer ^{22, 23} performed fluorescence lifetime studies on the human ocular fundus. They made use of TCSPC techniques with a pulsed laser of 300-ps pulse width at a wavelength of $457.9\mathrm{nm}$ and various emission wavelength ranges. They obtained a lifetime of $1.5\mathrm{ns}$ in the parapapillary region of the eye and a lifetime of $5\mathrm{ns}$ in the optical disc, which they attributed to collagen. They also showed that breathing 100% oxygen affects the fluorescence lifetimes and speculated that the lifetime of fluorophores present in the retinal pigment epithelium depends on the oxygen supply.

Fluorescence lifetime measurements have also been performed on tissue in vitro. For example, Maarek ²⁵ used time-domain in vitro fluorescence lifetime measurements and spectroscopy for diagnosis of atherosclerotic lesions. The instrument consisted of an excitation laser at $337\mathrm{nm}$ and a pulse width of $3\mathrm{ns}$ , a microchannel plate photomultiplier attached to a monochromator and connected to a digital oscilloscope. Fluorescence lifetimes were measured at emission wavelengths ranging from 370 to $510\mathrm{nm}$ . The authors were able to discriminate between stable and unstable lesions by studying the time-resolved spectra of components present in the arterial wall, namely, structural proteins such as collagen, lipoproteins, and cholesterol.²⁶

Moreover, Cubeddu ^{27, 28} used fluorescence lifetime imaging (FLIM) to study tumors in human skin. The instrumentation consisted of a dye laser pulsing at $405\mathrm{nm}$ at a pulse width of less than $1\mathrm{ns}$ and a gated imaging system. An exogeneous marker was used to induce selective accumulation of PPIX in proliferative tissue. Tumors were identified using lifetime imaging due to the longer fluorescence lifetime of PPIX $\left(18\mathrm{ns}\right)$ compared to that of healthy tissue $(\sim 10\mathrm{ns})$ . Siegel ²⁹ also used FLIM techniques, utilizing a pulsed laser at $415\mathrm{nm}$ and a pulse width less than $100\mathrm{fs}$ and a gated optical intensifier to image the autofluorescence from various biological tissue in vitro, such as animal tissue, knee joints, and human teeth. Using this instrument they were able to distinguish between tendon, bone, and surrounding tissue, which was not otherwise possible with white light and steady-state fluorescence measurements. Similarly, they showed that various regions in an extracted human tooth can be clearly distinguished using FLIM unlike reflectance and steady-state fluorescence measurements.

Other studies have performed time-resolved fluorescence measurements on human skin. Konig, ³⁰ Konid and Riemann,³¹ and Riemann ³² used 3-D autofluorescence lifetime imaging and investigated various tissues of patients with psoriasis, nevi, dermatitis, basalioma, and melanoma in vitro as well as the skin of healthy subjects in vivo. Their system was able to achieve a subcellular spatial resolution and a temporal resolution of $250\mathrm{ps}$ . The authors analyzed the fluorescence decays from images obtained at various depths of the skin and reported lifetimes of around $1.85\mathrm{ns}$ at the stratum corneum and a lifetime of about $2.4\mathrm{ns}$ at a depth of $50\mu \mathrm{m}$ within the stratum spinosum. The same group also reported using multiphoton imaging of skin at an excitation wavelength of $740\mathrm{nm}$ with a 150-ps resolution.³³ A biexponential fit was used to recover a fast lifetime of $0.4\mathrm{ns}$ and a slow lifetime of $2.4\mathrm{ns}$ at the stratum spinosum in the epidermis. Additionally, Pitts and Mycek² described a time-domain instrument with both spectral and temporal resolution to measure fluorescence lifetimes and emission spectrum in vivo and in vitro. The instrument was based on a pumped dye laser with a pulse width of $4\mathrm{ns}$ as an excitation source and an APD along with a 1-GHz oscilloscope as a detector. Fluorescence lifetime measurements were performed on human skin in vivo at an excitation wavelength of $337\mathrm{nm}$ and an emission wavelength of $460\mathrm{nm}$ . The time-resolved data were fit to a double-exponential decay, yielding lifetimes of 0.938 and $5.3\mathrm{ns}$ . The longer lifetime was attributed to the emission from collagen. The instrument was also used to measure fluorescence lifetimes of biomolecules present in human cells.

2.2.

Optical Noninvasive Screening of Type 2 Diabetes

A few groups have applied spectroscopic studies to the noninvasive screening of type 2 diabetes. For example, Meerwaldt ³⁴ observed steady-state autofluorescence measurements of skin on the arm and lower leg of both diabetic and healthy subjects. Diabetic and healthy subjects were age- and sex-matched. Using excitation spectrum between 300 and $420\mathrm{nm}$ and emission spectrum between 420 and $600\mathrm{nm}$ , they found a positive correlation between skin autofluorescence intensity with age, diabetes duration, and mean HbA1c levels from the previous year. The fluorescence intensity also correlated well with the concentrations of AGEs in tissue biopsies, such as pentosidine and carboxymethyllysine. This demonstrated the capability of steady-state fluorescence measurements to distinguish diabetics from healthy subjects. The study’s primary focus was monitoring disease progression and evaluating the patient’s risk for diabetic complications. In a later study, the same group showed that autofluorescence also correlated with severity of diabetic neuropathy³⁵ and vascular damage in type 2 diabetic subjects.³⁶ However, the effect of skin color was not addressed in these studies as nearly all subjects recruited were Caucasian. In fact, the researchers reported that data from dark-skinned subjects had to be discarded, because of nearly complete absorption of excitation light by the skin.³⁶ Otherwise, the steady-state measurements must be adjusted to account for skin complexion.

Brown ³⁷ clinically evaluated the use of a near-IR (NIR) spectroscopic device for its ability to correctly identify diabetic and nondiabetic subjects. The NIR wavelength region under observation in the study was from 1250 to $2500\mathrm{nm}$ . The diffusely reflected radiation was collected and the tissue absorbance was retrieved. Data from both diabetic and healthy subjects were collected and the device was found to have a sensitivity and specificity of 77.7 and 70%, respectively. Thus, this system’s sensitivity and specificity are comparable to the fasting plasma glucose (FPG) test. Although a definite chemical or physiological explanation for the difference between diabetic and nondiabetic patients is yet to be elucidated, they speculate that structural changes in skin due to chronic hyperglycemia contribute to the differentiation between diabetic and nondiabetic subjects. However, the same group’s previous work³⁸ suggested that NIR spectroscopy does not detect AGEs directly.

Recently, Maynard ³⁹ reported using a spectroscopic device for type 2 diabetes screening utilizing steady-state fluorescence as well as diffuse reflectance measurements. The fiber optic probe illuminated skin from the forearm of subjects using LED lights centered at 375, 405, 420, 435, and $460\mathrm{nm}$ . A spectrometer and CCD array then collected the light reflected and/or emitted from the skin and detected it. The spectra were corrected for skin pigmentation, hemoglobin content, and light scattering as well as age to obtain an “intrinsic” dermal fluorescence spectra. These data were then used to determine a likelihood that a subject belonged to the abnormal class, or that they were either prediabetic or diabetic. This method of screening was compared to the FPG test and A1C testing. Assuming a clinically relevant specificity of 77.4%, their method showed a sensitivity advantage of 28.8 and 17.1% versus FPG and A1C testing, respectively. The group also reported that the sensitivity of their device was not significantly different between subjects of lighter skin and those with of darker skin. Thus, their study demonstrates the promise of steady-state fluorescence for type 2 diabetes screening if factors such as skin color and age can be adjusted for in measurements. Time-resolved fluorescence may hold promise in diabetes detection by distinguishing different fluorophores emitting at the same wavelength and potentially being able to detect AGEs directly.

3.1.

Experimental Apparatus

The study reported here made use of the TCSPC technique to measure fluorescence lifetimes of skin. A schematic of the experimental setup is shown in Fig. 1 . Excitation and emission light were channeled by means of liquid light guides attached to a sensing head. Liquid light guides were used instead of commonly used fiber optics as their transmission and coupling efficiency are higher compared with fiber optics. Both light guides had a diameter of $8\mathrm{mm}$ , and the excitation light guide was held $1.59\mathrm{cm}$ above the skin. The receiving end of the collection liquid light guide was held abut against the excitation liquid light guide and angled at $45\mathrm{deg}$ to the normal. The sensing head also served to shield the sample surface from ambient light. A total surface area of $6.41{\mathrm{cm}}^2$ was exposed to excitation light.

Fig. 1

Schematic of the experimental setup, including (1) diode driver, (2) LED, (3) UV liquid light guide, (4) sample probe, (5) visible liquid light guide, (6) monochromator, (7) photomultiplier tube, and (8) personal computer.

The excitation source used was a light emitting diode or LED (PLS 370 diode, PicoQuant GmbH) with its center wavelength at $375\mathrm{nm}$ with a spectral width of around $10\mathrm{nm}$ . The full width half maximum of the diode pulses at a repetition rate of $2.5\mathrm{MHz}$ and average power of $4.8\mu \mathrm{W}$ per pulse was $664\mathrm{ps}$ . The LED was driven by a diode driver (PDL-800 B diode driver, PicoQuant GmbH, Berlin, Germany) at a repetition rate of $5\mathrm{MHz}$ for all measurements, producing an estimated energy fluence rate of $2.07\times {10}^{-6}\mathrm{W}∕{\mathrm{cm}}^2$ at the skin surface for a pulse energy of $2.65\mathrm{pJ}$ and pulse width of $664\mathrm{ps}$ . A liquid light guide designed for the UV to visible part of the spectrum (77628, from Spectra-Physics, U.S.A) carried light from the LED to the sensing head. Another liquid light guide designed for the visible part of the spectrum (77631, from Spectra-Physics, U.S.A), was used to collect the fluorescence emission and carry it to a monochromator (SpectraPro-150, Acton Research Corporation, U.S.A.). A photomultiplier tube (PMT) assembly (PMA-M 165 from PicoQuant GmbH) was coupled to the monochromator on the other end. The monochromator slits on the entry and exit of the light path were adjusted manually to maintain a bandwidth of $3\mathrm{nm}$ .

Finally, for data acquisition, the “sync” output signal from the diode driver and the “start” signal from the PMT assembly were fed to the respective channels on the data acquisition board (TimeHarp 200 from PicoQuant GmbH), via standard $50\text{-}\Omega$ coaxial cables. The “sync” output from the diode driver provided the signal to synchronize the timing electronics on the data acquisition board mounted on a personal computer (PC). The time resolution of our device was estimated to be $200\mathrm{ps}$ .

3.2.

Experimental Procedure and Parameters

The measurement procedure and parameters were arrived at by conducting preliminary studies on the palms, arms, cheeks, and feet of a 31-yr-old male Caucasian subject. The fluorescence lifetimes were measured at four different emission wavelengths: 442, 460, 478, and $496\mathrm{nm}$ . The first wavelength was chosen based on studies reporting that collagen fluorescence exhibited a maximum at around $440\mathrm{nm}$ when excited by light at $370\mathrm{nm}$ , corresponding to collagenase-digestible collagen cross-links, including AGEs on collagen.⁴⁰ The remaining wavelengths were arbitrarily chosen to cover the spectral range up to $500\mathrm{nm}$ .

To measure the fluorescence lifetimes, the monochromator was set to the emission wavelength of interest and the monochromator slits were adjusted to match that particular wavelength. The probe was then placed on the skin surface and the LED was turned on.

The PMT shutter was opened once the sensing head was firmly placed on the skin surface and then the histograms were collected using the TimeHarp software. The probe was held in contact with the palm until 1500 to 5000 counts were acquired in the peak channel. This number was arrived at by considering a trade-off between the duration of the experiments and the signal-to-noise ratio. The duration of a single measurement ranged from 5 to $20\min$ , based on the sensing location and emission wavelength. It was observed that the time taken for each measurement was shortest on the palm, possibly due to the low concentration of melanin. Moreover, higher emission wavelengths required longer durations to achieve these counts due to a lower intensity of the fluorescence. Figure 2 shows a typical sample of in vivo time-resolved fluorescence measurement on human skin along with the ${\chi }^2$ and associated residuals indicating the goodness of the fit. For an acceptable fit, the ${\chi }^2$ must be around 1.0 and the residuals should be randomly distributed around 0.

Fig. 2

Typical measurement of fluorescence decay from human skin (dark gray), IRF (light gray), and exponential decay retrieved from reconvolution (black line) with the TimeHarp Fluofit 4.0 software package.

The instrument response function (IRF) was measured by placing the probe on a Teflon block for the same number of maximum photon counts of at least 1500 as in the in vivo measurements, with the PMT collecting at $403\mathrm{nm}$ . The reconvolution software FluoFit version 4.0 was used to recover the lifetimes from the measured fluorescence decays. This software utilized the Marquardt-Levenberg algorithm with a weighting factor $W\left(i\right)$ defined as

Table 2

Population characteristics of healthy and diabetic subjects recruited for this study.

In addition to time-resolved fluorescence measurements, mean HbA1c values from the previous $4\mathrm{yr}$ were recorded for diabetic subjects by obtaining medical records with Institutional Review Board (IRB) approved consent (UCLA IRB #04-12-012-02). Subjects were also recruited from the local VA hospital with VA hospital IRB approved consent (VA IRB #PCC 2005-030449). All consent forms were signed prior to testing and collection of medical record data. All available data for each subject from the previous year were averaged and two subsets of these data were compared, one with “high” average HbA1c values and the other with “low” average HbA1c values. Data on duration of diabetes were obtained directly from the subject being tested.

4.1.

Effect of Skin Complexion and Gender

Figure 5 shows the fluorescence lifetimes ${\tau }_1$ , ${\tau }_2$ , and ${\tau }_3$ and their corresponding amplitudes $A_1$ , $A_2$ , and $A_3$ as a function skin complexion for healthy subjects at an emission wavelength of $442\mathrm{nm}$ . The figure indicates that the fluorescence lifetimes do not significantly vary for different skin complexions. Using the Fisher’s exact test,⁴¹ the fluorescence lifetimes and fractional amplitudes of healthy subjects with dark complexion and light complexion were compared. The light-complexioned group consisted of 10 Caucasian subjects, age $42\pm 12\mathrm{yr}$ , while the dark-complexioned group was composed of 10 Hispanic, Persian, South Asian, and African-American subjects, age $39\pm 20\mathrm{yr}$ . Based on these test results, no statistically significant difference in average fluorescence lifetimes and fractional amplitudes exists between these two groups ( $p>0.05$ in all cases). Note that measurements took longer for dark skin. However, skin complexion had no effect on the lifetimes and fractional amplitudes.

Fig. 5

Effect of skin complexion on the fluorescence lifetimes ${\tau }_1$ , ${\tau }_2$ , and ${\tau }_3$ and fractional amplitudes $A_1$ , $A_2$ , and $A_3$ as a function of age for healthy subjects at excitation and emission wavelengths of 375 nm and $442\mathrm{nm}$ , respectively.

Moreover, Fig. 6 shows the fluorescence lifetimes and amplitudes as a function of gender for healthy subjects in the same age group at the same emission wavelength of $442\mathrm{nm}$ . Again, there are no noticeable trends in the fluorescence lifetimes with respect to gender. The comparison of the fluorescence lifetimes and fractional amplitudes of 15 healthy female subjects, average age $42\pm 12\mathrm{yr}$ , and 22 healthy male subjects, average age $48\pm 20\mathrm{yr}$ , did not produce statistically significant differences ( $p>0.05$ for all cases). Thus, Figs. 5 and 6 as well as the statistical analysis suggest that fluorescence lifetimes and fractional amplitudes do not vary with skin complexion and gender. The same conclusions were made for other emission wavelengths (data not shown).

Fig. 6

Effect of gender on the fluorescence lifetimes ${\tau }_1$ , ${\tau }_2$ , and ${\tau }_3$ and fractional amplitudes $A_1$ , $A_2$ , and $A_3$ as a function of age for healthy subjects at a excitation and emission wavelengths of 375 nm and $442\mathrm{nm}$ , respectively.

4.2.

Effect of Age

Finally, an attempt was made to model the fluorescence lifetimes of healthy subjects shown in Fig. 4 as a function of age using linear regression. The square of the correlation coefficient $R^2$ was less than 0.03 for all three lifetimes, thus indicating that variations in fluorescence lifetimes with age cannot be described by a linear model (data not shown). Moreover, the small regression coefficients in all the cases also indicate that the three lifetimes do not depend on the age of the subject. Student’s $t$ -tests were performed on all six parameters (fluorescence lifetimes and fractional amplitudes) comparing young and old healthy subjects. In addition, 15 subjects in both the young and old groups were compared with average ages of $31\pm 7$ and $63\pm 11\mathrm{yr}$ , respectively. No statistically significant difference was detected for any of the parameters $(p>0.05)$ . Thus, the lifetimes and fractional amplitudes obtained for healthy subjects are statistically independent of age, gender, and skin complexion. Table 4 presents the mean values and the standard deviations of the fluorescence lifetimes and their corresponding fractional amplitudes for healthy subjects, providing reference values for healthy human skin. This clearly demonstrates the advantage of time-resolved fluorescence over steady-state fluorescence in that it is insensitive to age and skin complexion.

Table 4

Statistics of the decay parameters obtained on the left palm of 18 healthy and 27 diabetic subjects at an emission wavelength of 442nm ; the reported p values were obtained using the unpaired Student’s t -test to compare healthy and diabetic populations.

4.3.

Effect of Diabetes

Table 4 summarizes the experimental data collected for 18 healthy subjects and 27 age-matched diabetic subjects. Their fluorescence lifetimes and fractional amplitudes were compared by means of an unpaired Student’s $t$ -test. Only the fractional amplitudes $A_1$ and $A_2$ produced statistically significant differences in the means of the diabetic and the healthy patients. This was based on the criteria that $p<0.05$ for the difference between average values to be considered significant. Additionally, the relative contributions of each lifetime to the overall intensity of diabetic and healthy subjects were compared using the Student’s $t$ -test. The fractional contribution of a single lifetime to the steady-state intensity can be defined as⁴

Further testing was performed on only the diabetic subjects, comparing duration of diabetes as well as average HbA1c values over the past $4\mathrm{yr}$ . To test the effect of diabetes duration, the diabetic subjects were divided into two groups with an average of $11\pm 5\mathrm{yr}$ since diagnosis in the shorter duration group and $24\pm 7\mathrm{yr}$ in the longer duration group. However, no statistically significant difference was found between these two groups when performing the unpaired Student’s $t$ -test. Similarly, any correlation with average HbA1c from the previous $4\mathrm{yr}$ was tested by dividing the diabetic subjects into two groups: one with a low average HbA1c of $6.9\pm 0.7\%$ and higher average HbA1C of $9.8\pm 0.9\%$ . No significant difference was detected between these two groups ( $p>0.05$ for all cases).

Additionally, 18 diabetic subjects with foot ulcers, aged $62\pm 8\mathrm{yr}$ , were compared with 18 age-matched healthy subjects, average age $57\pm 15\mathrm{yr}$ , using the Student’s $t$ -test. Diabetic subjects with foot ulcers were specifically chosen because many diabetic subjects who develop ulcers also have other comorbidities associated with diabetes.^{42, 43} Thus, this group represents subjects whose AGE accumulation is expected to be much higher than other diabetic subjects. Not only were statistically significant differences noted between the two groups for fractional amplitudes $A_1$ and $A_2$ , but also differences in lifetime ${\tau }_1$ between the two groups was also statistically significant, albeit with a $p$ value of 0.038. All the test results are presented in Table 5 . Given the long half-life of collagen in skin⁴⁴ $(\sim 15\mathrm{yr})$ , one can expect changes in the time-resolved fluorescence properties of skin to correlate more with the occurrence of diabetic complications than with HbA1c values. This measurement would indicate the long-term accumulation of AGEs and other oxidative damage. The rate at which tissue accumulates AGEs is also dependent on various genetic or metabolic factors.⁴⁵ Therefore, not all diabetic subjects are as susceptible to the development of complications, in spite of poor blood glucose control and high HbA1c levels. The development of complications cannot be predicted by conventional treatment monitoring such as blood glucose measurements and HbA1c tests. Finally, nine diabetic subjects without foot ulcers were compared with 18 age-matched healthy subjects using the Student’s $t$ -test. No statistically significant differences between the two groups were detected in any of the parameters except in amplitude $A_2$ ( $p$ value of 0.045). This indicates that the differences in the amplitude $A_1$ and ${\tau }_1$ between 18 healthy and 27 diabetic subjects reported earlier might be mainly due to the diabetic subjects with foot ulcers.

Table 5

Statistics of the decay parameters on the left palm of 18 diabetic subjects with foot ulcers and 18 age-matched healthy subjects at an emission wavelength of 442nm ; the reported p values were obtained using the unpaired Student’s t -test to compare healthy and diabetic populations.

While a previous study by Meerwaldt ³⁴ showed that steady-state autofluorescence measurements correlated with age, diabetes duration, and mean HbA1c of the previous year, this study did not find any significant relationships between the fluorescence lifetimes and amplitudes and these variables. Meerwaldt ³⁴ used a polychromatic excitation source emitting at wavelengths ranging from 300 to $420\mathrm{nm}$ and fluorescence was measured over the entire window of 420 to $600\mathrm{nm}$ . This is in contrast with this study, where the excitation source was a monochromatic diode emitting at $375\mathrm{nm}$ (spectral width of around $10\mathrm{nm}$ ) and fluorescence was measured at 442, 460, 478, and $496\mathrm{nm}$ with spectral width of 3 nm. It is speculated that these differences in the excitation wavelength might be the source of the different conclusions obtained in the two studies. This study should be repeated for a different excitation wavelength between 300 and $420\mathrm{nm}$ and emission wavelength between 420 and $600\mathrm{nm}$ to corroborate results by Meerwaldt ³⁴

4.4.

Identification of Fluorophores

In an earlier study on fluorescence lifetime measurements on human skin,³³ two fluorescence lifetimes of 0.4 and $2.4\mathrm{ns}$ at the stratum spinosum were reported using multiphoton excitation at $740\mathrm{nm}$ , equivalent to single-photon excitation at $370\mathrm{nm}$ . In the presented study, similar values were found when the decays were fit to a triexponential model. This suggests that the fluorophores with the high fractional amplitudes $A_1$ and $A_2$ in our measurements originate, at least in part, from the stratum spinosum. Our measurements are detecting another fluorophore as well, perhaps deeper in the tissue, which would explain the lower contribution of this fluorophore to the overall fluorescence signal. The similarities between our results and the previously reported data help to validate our results in spite of using different multiexponential models. Moreover, the shorter two lifetimes ${\tau }_1$ around $0.4\mathrm{ns}$ and ${\tau }_2$ around $2.7\mathrm{ns}$ observed in our measurements were similar to the lifetimes of free and protein-bound NADH reported in literature.^{31, 46} Indeed, Schneckenburger ⁴⁶ studied the autofluorescence from cultivated endothelial cells and determined the fluorescence lifetime of free and protein-bound NADH to be between 0.4 and 0.5 and between 2.0 to $2.5\mathrm{ns}$ , respectively. In addition, Konig and Riemann³¹ also reported a fluorescence lifetime of $2.7\mathrm{ns}$ at a depth of $50\mu \mathrm{m}$ within the skin, which could possibly be attributed to bound NADH. Based on these studies, it is speculated that the two shorter fluorescence lifetimes ${\tau }_1$ and ${\tau }_2$ correspond to free and protein-bound NADH, respectively. This suggests that the differences in fractional amplitudes $A_1$ and $A_2$ observed between diabetic and healthy subjects are related to the metabolic states of their tissues rather than directly detecting the presence of glycation products. However, as the presence of glycation products is known to increase tissue oxidative stress,^{47, 48} they may have an indirect effect on our measurements.

Furthermore, it is speculated that the third observed average lifetime ${\tau }_3$ equal to $9.2\mathrm{ns}$ for healthy subjects corresponds to collagen cross-links formed^{34, 38} by AGEs. However, our data on diabetic subjects seem to contradict this speculation. The longest lifetime ${\tau }_3$ and the corresponding fractional amplitude $A_3$ are not significantly different between diabetic and healthy subjects. Given the known correlation between AGEs and diabetes and skin steady-state autofluorescence,³⁴ one would expect the fractional amplitude of the third lifetime to increase with increasing AGE accumulation, if the lifetime could be attributed to AGEs. With elevated concentration of AGEs in the skin, their fluorescence signal would be stronger and, thus, contribute a higher fractional intensity. Further in vitro experimentation is required to precisely determine the origin of this lifetime.