13 March 2015 Fluorescence lifetime imaging ophthalmoscopy in type 2 diabetic patients who have no signs of diabetic retinopathy
Author Affiliations +
J. of Biomedical Optics, 20(6), 061106 (2015). doi:10.1117/1.JBO.20.6.061106
The time-resolved autofluorescence of the eye is used for the detection of metabolic alteration in diabetic patients who have no signs of diabetic retinopathy. One eye from 37 phakic and 11 pseudophakic patients with type 2 diabetes, and one eye from 25 phakic and 23 pseudophakic healthy subjects were included in the study. After a three-exponential fit of the decay of autofluorescence, histograms of lifetimes τi, amplitudes αi, and relative contributions Qi were statistically compared between corresponding groups in two spectral channels (490<ch1<560  nm, 560<ch2<700  nm). The change in single fluorophores was estimated by applying the Holm–Bonferroni method and by calculating differences in the sum histograms of lifetimes. Median and mean of the histograms of τ2, τ3, and α3 in ch1 show the greatest differences between phakic diabetic patients and age-matched controls (p<0.000004). The lack of pixels with a τ2 of ∼360  ps, the increased number of pixels with τ2<450  ps, and the shift of τ3 from ∼3000 to 3700 ps in ch1 of diabetic patients when compared with healthy subjects indicate an increased production of free flavin adenine dinucleotide, accumulation of advanced glycation end products (AGE), and, probably, a change from free to protein-bound reduced nicotinamide adenine dinucleotide at the fundus. AGE also accumulated in the crystalline lens.
Schweitzer, Deutsch, Klemm, Jentsch, Hammer, Peters, Haueisen, Müller, and Dawczynski: Fluorescence lifetime imaging ophthalmoscopy in type 2 diabetic patients who have no signs of diabetic retinopathy



Detection of the earliest signs of diabetic retinopathy (DR) increases the probability of reducing further pathological developments, provided that an appropriate therapy will be administered.1,2

Apoptosis of vascular and neural cells has been reported for the early onset of DR.3 Furthermore, in diabetes, neuronal degeneration due to apoptosis in the inner retina has been found to be 10-fold higher than that in vascular cells.4,5

Barber et al.3 found that apoptosis of neural cells is an earlier marker than is damage of vascular cells. No change in concentration of reduced nicotinamide adenine dinucleotide (NADH) was found between excited retina of diabetic rats and controls.6 Further studies were performed on excited retinas of rats under euglycemic and hyperglycemic conditions as well as in vivo on diabetic rats.7 In these studies, no elevated cytosolic NADH/NAD ratio was found under hypoglycemic conditions nor in diabetes, both in vitro and in vivo. But they found an increased polyol synthesis and metabolites upstream of glucose like the sorbitol pathway, which decreases NADPH.

Increased oxidation of sorbitol results in the generation of cytosolic NADH, which in early diabetes leads to metabolic imbalance as well as neural and vascular damage.8

In diabetic rats that were provided with a low dose of insulin, the thicknesses of the inner nuclear and plexiform layers were significantly reduced after 7.5 months. The density of ganglion cells was also reduced. These neurodegenerative changes were observed in the absence of morphologic vascular changes.9

Spectral domain optical coherence tomography showed that the total thickness of the retinal layer, particularly the retinal nerve fiber layer, was significantly thinner in diabetic Otsuka Long-Evans Tokushima fatty rats after 28 weeks than in nondiabetic Long-Evans Tokushima Otsuka rats. Additionally, a decreased number of ganglion cells and an increased frequency of apoptosis have been found by histological investigations.10

Isolated retinal capillary cells showed typical signs of mitochondrial dysfunction, including an increased release of cytochrome c into the cytosol and an accumulation of the proapoptotic protein Bax in mitochondria cells when incubated in 20 mM glucose. However, these changes were inhibited in the presence of superoxide dismutase.11

In diabetic patients, first metabolic alterations were found by Field et al.12 They excited the fundus using flashes of 467 nm light in a 3-deg field and detected the autofluorescence at 535 nm. Increased fluorescence intensity and a wider distribution of intensity of all pixels in the excited field were found in diabetic patients in relation to controls. These results were interpreted as the result of an impaired electron transport by energy-generating enzymes in the respiratory chain. Because of the specific excitation and emission wavelengths, an increased contribution of flavin adenine dinucleotide (FAD) was assumed.

The problem of in vivo fluorescence measurement is that the crystalline lens and the fundus layers are simultaneously excited. An exact interpretation of the spectrally resolved fundus fluorescence is possible only in eyes with artificial intraocular lenses (IOL) or when the fluorescence of the crystalline lens is separately detected and subtracted from the fluorescence of the whole eye.13,14

In clinical practice, the classification of DR is based on vascular changes according to the Early Treatment Diabetic Retinopathy Study Research Group.15

Using the highly accurate spectral domain optical coherence tomography, it was demonstrated that the thicknesses of the retinal nerve fiber layer, the ganglion cell layer, and the inner plexiform layer were reduced in patients with minimal signs of DR in the pericentral macular area compared with those in the controls. Moreover, in the peripheral macular area, the retinal nerve fiber layer and the inner plexiform layer were found to be thinner than in the controls. The DR status was the only significant explanatory variable for this reduced thickness.16

Measurements of the time-resolved fluorescence of endogenous fluorophores have the potential to detect metabolic alterations in the human fundus.17 The redox pairs of oxidized and reduced NAD-NADH and FAD-FADH2 act as electron carriers in the basic processes of energy metabolism (citrate acid cycle, respiratory chain). Additional fluorophores, such as lipofucin components, advanced glycation end products (AGE), collagens, elastin, products of the visual cycle (retinol), or components of heme synthesis (protoporphyrin IX), contribute to the autofluorescence of the fundus. In principle, these fluorophores can be discriminated by measuring specific excitation and emission spectra. Due to the transmission of the ocular media, however, excitation of endogenous fluorophores is only possible for wavelengths longer than 400 nm. Because the excitation maxima of most of these fluorophores are at wavelengths shorter than this value, any discrimination based on excitation spectra is difficult. The decay of fluorescence intensity (lifetime) after pulse excitation permits the differentiation of fluorophores with overlapping emission spectra. The free or protein-bound status of a fluorophore can also be distinguished if the emission spectrum is equal in both cases.

The excitation and emission spectra and the lifetimes of the expected fundus fluorophores and of the anatomical structures of porcine eyes have been reported.18

The purpose of this study was to determine whether metabolic changes in the time-resolved autofluorescence of the fundus are detectable in diabetic patients who have no signs of DR.


Participants and Methods



The patients were recruited from the outpatient Clinic of Internal Medicine III of the Jena University Hospital. Inclusion criteria were a diagnosis of type 2 diabetes, no signs of DR, and no or minimal cataracts. No subject was suffering from posterior capsule opacification. The DR stage was blindly evaluated from color fundus images by a retina specialist. The subjects in the control group were volunteers or patients from the Eye Clinic of the Jena University Hospital and did not have diabetes mellitus. They had no ocular disease and had clear media or a mild cataract (nuclear or cortical stadium 1).19 Subjects suffering from systemic diseases that could influence ocular metabolism were not included. Because the fluorescence decay depends on age,20,21 the only subjects who were included in the age-matched comparison between diabetic patients who have no signs of DR and healthy subjects were older than 40 years. The investigation was performed predominantly on subjects with a crystalline lens. For comparison, small groups of pseudophakic (IOL) diabetic patients and healthy subjects were investigated.

All subjects provided written consent to participate in the study. All research procedures were performed according to the Declaration of Helsinki. Approval for the study was obtained from the ethics committee of the Jena University Hospital.

HbA1c was measured using high-performance liquid chromatography (official normal range 4.4% to 5.9%; mean 5.2; TOSOH-Glykohämoglobin-Analyzer-HLC-723-GHbV; Tosoh, Tokyo, Japan). The mean normal range measured in 1079 nondiabetic patients in 2009 was 5.65%±0.38%.22 Therefore, we adjusted the HbA1c according to the mean normal value of healthy people to the value of 5.05% (32mmol/mol) that was found in the Diabetes Control and Complication Trial Research Group (DCCT).23 Blood pressure was measured immediately before the autofluorescence measurement. Table 1 shows the data of the subjects who were included in the study. Values are given as the mean±standard deviation.

Table 1

Characterization of the subjects included in the study.

Diabetes type 2Healthy subjects
Crystalline lensIOLCrystalline lensIOL
Age (years)60.5±15.361.1±14.761.9±17.365.3±14.7
Gender22 M, 15 F8 M, 3 F14 M, 11 F10 M, 13 F
BMI (kg/m2)29.8±6.130±626.6±4.226.7±4
Systolic blood pressure (mmHg)138.9±15.4136.4±16.5134.8±12.2135.1±12.4
Diastolic blood pressure (mmHg)82.3±10.881.5±1078.9±8.878.1±9
Antihypertonic drugs811002839
Statins (%)4370813
Time since diagnosis of diabetes (years)14.7±9.816.4±10.7
Oral antihyperglycemic agents (%)3020
Treatment with insulin5760
Diet (%)1320
HbA1C DCCT adjusted (mmol/mol; %)49.9±13.9; 7.1±1.446.3±7.6; 6.8±0.7
Neuropathy (%)2420
Albumin/creatinine in urine (mg/g)76±233154±198
Albumin/creatinine in urine >20  mg/g (% patients)2730

Note: IOL, intraocular artificial lens.

The albumin/creatinine ratio in the urine indicates that some diabetic patients were suffering from microalbuminuria, and some were partly already suffering from macroalbuminuria. Microalbuminuria is assumed for albumin/creatinine ratios of 30 to 300mg/g, and macroalbuminuria is present for ratios >300mg/g.


Measuring System

The time-resolved autofluorescence of the fundus was measured using a modified confocal scanning laser ophthalmoscope (HRA II, Heidelberg Engineering GmbH, Heidelberg, Germany). The optomechanical unit of an HRA II was modified at the Eye Clinic of the University of Jena, Germany, for fundus excitation by a pulsed laser (BDL-440-SMC, Becker&Hickl GmbH, Berlin, Germany), which emitted 75-ps pulses at 448 nm at an 80 MHz repetition rate.18 The fluorescence was detected in two spectral channels (ch1: 490 to 560 nm, ch2: 560 to 700 nm) using multichannel plate photomultipliers (Ham-R 3809U-50, Hamamatsu, Herrsching, Germany). The fluorescence decay was detected using time-correlated single photon counting (TCSPC) by an SPC 150 board (Becker&Hickl GmbH). An HRT 41 router (Becker&Hickl GmbH) separated the photons from the two spectral channels. Fluorescence decay images were recorded from a 30-deg field with a resolution of 40×40μm2. The excitation power was 100μW in the cornea plane. Simultaneously with the excitation laser, the fundus was illuminated with an infrared laser (820 nm), which provided contrast-rich fundus images for automatic image registration in both spectral channels. This device was the forerunner model of the demonstrator developed by Heidelberg Engineering.21


Fitting of the Fluorescence Decay

The software SPCImage3.6 (Becker&Hickl GmbH) was used to evaluate the time-resolved autofluorescence images. To obtain a sufficient number of photons, a binning factor B=2 was used, which increased the number of photons at a single pixel by a factor of 25. The fluorescence decay was approximated by the three-exponential model function given in Eq. (1):


where I(t) is the number of photons at time t, IRF is the instrument response function, ai is the amplitude or pre-exponential factor, τi is the lifetime of exponent i, and b is the background.

To avoid the influence of the stepped slope of fluorescence intensity which is caused by the fluorescence of the crystalline lens, a tailfit was employed.

The criterion for an optimal approximation was the minimization of χr2 [Eq. (2)]:


where n represents the time channels (1024 applied here), q represents the free parameters (αi, τi, and b), N(tj) represents the detected photons in time channel j, and Nc(tj) represents the photons from the convolution of the model function with the IRF.

Assuming that the detection of photons is a Poisson process, the limiting value in Eq. (2) is 1.

In addition to all of the individual amplitudes and lifetimes, the mean lifetime τm and the relative contribution Qi parameters are important for evaluating fluorescence lifetime imaging ophthalmoscopy (FLIO) measurements. Here, the mean lifetime is defined as



The relative contribution Qi of the component i corresponds to its respective area under the decay curve.



Pixel-by-pixel fitting of the measured fluorescence decay to a three-exponential model [Eq. (1)] results in images of the lifetimes τ1, τ2, and τ3, and the amplitudes α1, α2, and α3 for the fundus of each subject in both channels. Images of τm and of Qi can be calculated from the lifetimes and amplitudes.

Because the system did not contain an internal fixation target, the 30-deg field was not exactly at the same position at the fundus for all subjects. Thus, the comparison of the FLIO parameter of the complete images was not appropriate. Therefore, after approximation of the fluorescence decay of each pixel of the whole image, a region of interest (RoI) of 71×101pixels, which was located in the superior temporal quadrant that included the macula, was selected for further calculations in all measurements. This region was contained in all images.


Evaluation of Fitting Results

In our study, we calculated the fitting results for each pixel within the RoI (71×101pixels) and determined how frequently each value was determined for each lifetime τi, amplitudes αi, and Qi. In this way, one histogram of each τi, αi, and Qi was calculated in both spectral channels for each subject. The histograms of phakic diabetic patients and healthy subjects were compared in two steps.

First, the derived quantities’ mode, mean, and median from these histograms were compared for each individual between the two groups. The statistical comparison was performed using SPSS 21 (IBM Deutschland GmbH, Ehningen, Germany). The two-tailed t test was used if the error probability in the Kolmogorov-Smirnov test was >0.2 (normal distribution). Equality of variances was assumed if the error probability in the Levene-test was also >0.2. The Mann-Whitney U test was used if the parameters were not normally distributed. Significant differences in the fitting parameters were assumed for p<0.05. The area under curve of the receiver operating characteristics (ROC) was calculated. The ideal value is 1, while a value of 0.5 indicates that the separation between groups is only stochastic. Sensitivity and specificity were given for ROC>0.72. These results are global because each lifetime τi and amplitude αi results from the sum of the contributions from several fluorophores.

In a second step, changes in the single fluorophores were estimated. For this process, the Holm–Bonferroni method24 was applied using the software program FLIMX, which was developed at the University of Ilmenau.25,26

Here, the distribution of each fit parameter was divided into n intervals. The lifetimes in such intervals are, to a certain degree, determined by specific fluorophores. Differences between the fitting parameters in single intervals are considered significant if the error probability p is lower than the significance level α divided by the number n of intervals. The Wilcoxon test was performed for statistical comparison of the fit parameters of diabetic patients and healthy subjects in each interval.

The crystalline lens has a long fluorescence lifetime which influences the FLIO parameters, especially in channel 1. In contrast, an artificial IOL exhibits no fluorescence.

For that reason, the range of fit parameters was different for subjects with crystalline lens and for pseudophakic subjects. The size of the intervals was iteratively determined until the sum of the sensitivity and of specificity as well as ROC was maximal. The ranges and the size of the intervals for the fitting parameters are given in Table 2 for phakic subjects and in Table 3 for pseudophakic subjects.

Table 2

Range of fitting parameters for phakic subjects.

ParameterInterval sizeRange ch1Range ch2
α1 in %274 to 9260 to 88
α2 in %18 to 1914 to 34
α1 in %0.51 to 6.51.5 to 5
τ1 in ps535 to 10545 to 105
τ2 in ps10330 to 690300 to 590
τ3 in ps502500 to 46001800 to 3500
τm in ps10120 to 370150 to 360
Q1 in %212 to 369 to 35
Q2 in %113 to 3730 to 47
Q3 in %226 to 7420 to 50

Table 3

Range of fitting parameters for pseudophakic subjects.

ParameterInterval sizeRange ch1Range ch2
α1 in %272 to 9060 to 80
α2 in %110 to 2418 to 29
α1 in %0.50.5 to 52 to 6
τ1 in ps545 to 8545 to 85
τ2 in ps10330 to 460330 to 460
τ3 in ps502000 to 30001800 to 2300
τm in ps10100 to 250160 to 270
Q1 in %219 to 4112 to 34
Q2 in %128 to 4538 to 49
Q3 in %222 to 4622 to 46



The decay of the fluorescence intensity was fitted by the three-exponential model function of Eq. (1) for each pixel in both spectral channels. As demonstrated in Fig. 1, the model function sufficiently approximates the measured decay.

Fig. 1

Tailfit of a paramacular pixel by a three-exponential model function. The fit parameters are α1=82.3%, α2=11.3%, α3=6.4%, τ1=105ps, τ2=639.8ps, τ3=4294.4ps, and χr2=1.14.


After the fitting, images of the lifetimes, amplitudes, and relative contributions were calculated for both spectral channels. Figure 2 shows these images in the short-wavelength channel for a healthy subject and for a patient with diabetes but no signs of DR. The bluer color in the images of the patient with diabetes indicates a prolongation of the lifetimes.

Fig. 2

Paired comparison of the amplitudes αi (upper two rows of images), lifetimes τi (middle two rows of image), and relative contributions Qi (lower two rows) of a diabetic patient and a healthy subject in the short-wavelength channel. Images of the diabetic patients are in the upper part, and images of the healthy subject are in the lower part of each double row of images. The regions of interest are highlighted (white box).


The size and the position of the ROIs selected for quantitative comparison of the histograms of FLIO parameters are also shown in Fig. 2.

Generally, each histogram shows how often the same value of each parameter (lifetime, amplitude, and relative contribution) was determined in the analyzed field of a subject.

Figure 3, as an example, depicts the histogram of the lifetime τ2 of a diabetic patient in ch1 and ch2. The differences in the distribution of the lifetime τ2 in the two spectral channels indicate that the fluorescence results from different substances. Furthermore, this difference was detected equally in both the diabetic patients and the healthy subjects. From these distributions, statistical parameters were extracted for each subject.

Fig. 3

Histograms of lifetime τ2 of 7171 pixels in channels 1 and 2. The statistical parameters in picoseconds are as follows: for ch1: mode=470, median=549, mean=528, standard deviation=134; for ch 2: mode=420, median=442, mean=435, standard deviation =36.4.


The results of the statistical comparison are given for subjects with a crystalline lens in Table 4 for channel 1 (490 to 560 nm) and in Table 5 for channel 2 (560 to 700 nm).

Table 4

Statistical comparison based on the fit parameters of the fundus fluorescence in channel 1 (490 to 560 nm) for the phakic eyes of healthy subjects and diabetic patients who have no signs of diabetic retinopathy (DR).

Fit parameterTestModalMedianMeanSDSkewnessKurtosis
Channel 450 to 560 nmα1t test0.0080.011nsns
α3t test1E-061E-06
τ1t test0.0040.008
τ3t test4E-064E-06nsns
τmt test3E-063E-06ns
Q1t test1E-063E-06nsns
Q2t test1E-071E-07nsnsns
Q3t testns1E-071E-07ns0.029ns

Note: ns, nonsignificant; ROC, receiver operating characteristic.

Table 5

Statistical comparison based on the fit parameters of the fundus fluorescence in channel 2 (560 to 700 nm) for the phakic eyes of healthy subjects and diabetic patients who have no signs of DR.

Fit parameterTestModalMedianMeanSDSkewnessKurtosis
Channel 560 to 700 nmα3t test0.0040.004ns
τ1t test0.0220.0380.018nsns
τ2t test8E-059E-050.001
τ3t test1E-062E-052E-052E-05nsns
τmt test0.0010.0010.001ns
Q1t test0.0050.0053E-05ns0.001
Q2t test1E-077E-050.001nsns
Q3t test4E-056E-056E-055E-06nsns

In the short-wavelength channel, the best discrimination between healthy subjects and diabetic patients who have no signs of DR was achieved for the median and mean parameters in the distributions of α3, τ2, τ3, τm, Q1, Q2, and Q3. The statistical values of skewness and kurtosis mostly did not reach significance. The highest ROC value (0.875) with 83.8% sensitivity and 76% specificity was found for the median of τ2 in ch1 with an error probability p=0.000001. Good discrimination was also achieved for the medians of α3 and τ3 in the short-wavelength channel with error probabilities of p=0.000005.

No significant difference between the two groups was obtained for the amplitude α2. In contrast, the difference in the lifetime τ2 between the two groups is highly significant, supporting the assumption that the composition of the fluorescent substances changes.

The best discrimination was obtained in the long-wavelength channel (560 to 700 nm) using the median and mean of the distributions of τ2, τ3, Q1, Q2, and Q3 in the selected RoIs at the fundus (Table 5).

The statistical comparison was also performed for pseudophakic patients and controls (Table 6). Only 10 subjects were included in the diabetic group; thus, the value of the statistical comparison is limited. The median and mean were only significantly different for τ2 in the short-wavelength channel, but on a much lower level than in phakic eyes. Surprisingly, the best discrimination was possible for the kurtosis of the amplitude α3. In the long-wavelength channel, the differences for the median and mean reached significance for τ1 and τ3.

Table 6

Statistical comparison based on the fit parameters of the fundus fluorescence of the pseudophakic eyes of healthy subjects and diabetic patients who have no signs of DR.

Fit parameterTestModalMedianMeanSDSkewnessKurtosis
Channel 490 to 560 nmα3Mann-Whitneynsnsnsnsns0.007
τ2t testns0.0420.031nsnsns
τ3t test0.031nsnsns0.047ns
τmt test0.035nsnsnsns
Q3t testnsnsns0.036ns
Channel 560 to 700 nmα3Mann-Whitneynsnsnsnsns0.042
τ1t testns0.0170.014nsnsns

Figure 4 shows median boxplots for all FLIO parameters in ch1 and ch2 for phakic subjects. For pseudophakic subjects, only significantly different parameters are given.

Fig. 4

Comparison of the median boxplots of different parameters between diabetic patients and healthy subjects: (a) crystalline lens, (b) intraocular artificial lens. H, healthy subjects (light gray); D, diabetic patients (dark gray); ns, not significant; ****p<0.0001, **p<0.05.


To obtain more information about the substance-specific changes in the eyes of diabetic patients as an early sign of metabolic alterations before the signs of DR are visible, the Holm–Bonferroni test was applied. Many fluorophores are excited in the eye. Thus, each lifetime τi in a three-exponential approximation includes lifetimes from different fluorophores. Thus, the histograms that are created from the fitting parameters for all of the RoI pixels reflect the contributions of several fluorophores.

Therefore, similar to each approximated τi, the contribution of individual fluorophores with slightly different lifetimes can be estimated if the range of τi is divided into intervals. In each interval, the difference in each fitting parameter was tested between the distributions corresponding to healthy subjects and diabetic patients. The discriminations between the groups are independent of the intervals; thus, the results of intervals with high sensitivity and high specificity can be combined to obtain the best separation. Table 7 summarizes the results of the Holm–Bonferroni test for phakic diabetic patients and healthy subjects.

Table 7

Results of the Holm–Bonferroni test for diabetic patients who have no signs of DR and for healthy subjects.

Fit parameterSignificanceROCSensitivitySpecificityBest separationFurther separation intervals
Channel 490 to 560 nmα1ns0.70.6220.882
α30.0010.730.9460.521.51.5, 5.5
τ20.0010.860.8380.84400400, 410, 480 to 580
τ30.0010.840.7570.8430502800 to 3100, 4400
τm0.010.770.8380.64150140 to 160, 270, 330 to 360
Q10.0010.770.8910.643434 to 36
Q20.0010.850.8920.722921, 28 to 31
Q30.0010.780.8380.724038 to 40
Channel 560 to 700 nmα1ns0.650.6490.6874
α30.050.710.5680.844.54 to 5
τ10.010.770.6760.766565, 80
τ20.010.790.6760.845037 to 380, 430 to 460
τ30.0010.850.8920.821002050 to 2150, 2650 to 2900
Q10.050.70.5410.881818, 20
Q20.0010.870.7570.964335 to 39, 43 to 45
Q30.010.760.7570.764040 to 44

The best discrimination between healthy subjects and diabetic patients who have no signs of DR was obtained for the lifetime τ2 in ch1 (490 to 560 nm) in the interval of 400 to 410 ps (ROC=86%, sensitivity=83.8%, specificity=84%), τ3 for 3050 ps in the interval 2800–3100 ps (ROC=84%, sensitivity=75.7%, specificity=86%), and Q2 for 29% (ROC=85%, sensitivity=89.2%, specificity=72%).

In the long-wavelength channel, the best separations were found for τ3 at 2100 ps (ROC=85%, sensitivity=89.2%, specificity=80%) and for Q2 at 43% (ROC=87%, sensitivity=75.7%, specificity=96%).

In the comparison between the controls and the diabetic patients with an artificial IOL, significance in the Holm–Bonferroni test was only reached for τ1 at 65 ps (ROC=81%, sensitivity=80%, specificity=69.6%) in ch1 (Table 8). The intervals of the fitting parameters are given for the parameters whose ROC values are at least 0.7. We speculate that the low number of diabetic patients with an artificial IOL led to the lower number of significant parameters compared with the results in Table 7. On the other hand, these results point at the considerable influence of changes in the fluorescence of diabetic lenses.

Table 8

Results of the Holm-Bonferroni test for pseudophakic diabetic patients who have no signs of DR and for healthy subjects.

Fit parameterSignificanceROCSensitivitySpecificitySeparation interval
Channel 490 to 560 nmα2ns0.70.70.69615
Channel 560 to 700 nmτ1ns0.710.90.56585

Graphical analysis of the histograms is a further step in the interpretation of metabolic alterations at the fundus of diabetic patients. The sum and difference histograms of τ2 in ch1 for diabetic patients and for healthy subjects with a crystalline lens are given in Fig. 5.

Fig. 5

Sum and difference histograms of lifetime τ2 in the short-wavelength channel (490 to 560 nm).


We observe a shift in the sum histograms for τ2 to longer fluorescence decay times in the fundus of diabetic patients who show no signs of DR compared with those for the controls. The curve in the difference histogram of τ2 in Fig. 5 clearly shows that in the RoI of diabetic patients, the number of pixels with a decay time of 380ps is reduced, whereas more pixels with fluorescence decay times longer than 450ps are present compared with those in the controls.

As demonstrated in Fig. 6, there is also a shift in the sum histograms of τ3 to longer fluorescence decay times in the fundus of diabetic patients who show no signs of DR when compared with the controls.

Fig. 6

Sum and difference histograms of lifetime τ3 in the short-wavelength channel (490 to 560 nm).


Likewise, considering the difference histogram for the decay time τ3 in Fig. 6, fewer pixels with a lifetime of 3000ps are present in diabetes, and more pixels with longer decay times of 3900ps are present in the selected RoIs.

Longer decay times for τ2 and τ3 in the short-wavelength channel as well as higher values of α3 were observed in the diabetic patients versus controls.



The detection of early metabolic changes in the fundus is important for the development of a specific treatment for DR. First in vivo spectral measurements of metabolic alterations at the fundus of diabetic patients were published by Field et al.12 It is assumed that alterations in the retinal vessel system3 or thinning of fundus layers16 are most likely consequences of metabolic malfunction.

The new technique of FLIO permits the discrimination of fluorophores according to the decay of fluorescence after pulse excitation and, to a certain degree, according to the emission spectra. Based on spectral and time-resolved in vitro measurements of autofluorescence on isolated structures27of porcine eyes and of fundi in toto28 as well as of human fundus structures,29 we assume a relation between FLIO parameters and fundus layers. Additionally, cross-sections in FLIO images also show a relation between anatomical structures and lifetimes as well as amplitudes.30

Thus, the amplitude α1 and the decay time τ1 might correspond to some extent with the retinal pigment epithelium, α2 and τ2 likely correspond to the neuronal retina, and α3 and τ3 likely demonstrate the influence of connective tissue and of the lens.

The decay times and the amplitudes in the short-wavelength channel (490 to 560 nm) differ from those in the long-wavelength channel (560 to 700 nm); thus, different substances contribute to the total fluorescence in the two channels.

In particular, the long fluorescence decay of the crystalline lens is present in the short-wavelength channel, where the τ3 is much longer than the τ3 in the long-wavelength channel. This contribution is confirmed because the τ3 in the short-wavelength channel in pseudophakic eyes is much shorter than that in eyes with a crystalline lens.

To investigate such metabolic alterations, diabetic patients who have no signs of DR and age-matched healthy subjects were investigated by FLIO. Although the mean duration of diabetes mellitus was 15 years, the mean HbA1c value was 7.1±1.4%, which is in the 6.5 to 7.5% range recommended by the EASD/ADA31 and by the National German Guideline.32

In contrast to optical coherence tomography measurements of the fundus layers and visible changes in the retinal vessel system, FLIO measurements detect functional metabolic changes.

Considering amplitudes and lifetimes after a three-exponential fit of fluorescence decay in fields of 71×101pixels, the best discriminating parameters were the median and mean in histograms of τ2. Because τ2 presumably corresponds to the neuronal retina, metabolic alterations in this anatomical structure might be assumed to occur in the fundus of diabetic patients before DR is detectable in color fundus photographs. Early alterations in the vessel system are detectable by fluorescein angiography. As this is an invasive test, it was not applied for ethical reasons in people who have had no visual complaints.

In addition to the statistical evaluation of the distribution of individual FLIO parameters, the sum of the individual histograms of the amplitudes, decay times, and relative contributions was considered for both groups. Similar to each individual histogram, the sum histograms contain the fluorescence of several fluorophores. Thus, it is difficult to attribute any alterations to changes of only one fluorophore.

Calculations of the difference between the sum histogram of diabetic patients and that of healthy subjects revealed decreased abundance at 380ps for τ2 and at 3000ps for τ3; however, more pixels had longer τ2 decay times of 480ps and a longer τ3 of >3800ps.

There are several possibilities for interpretation of this change. A change in the composition of elastin (lifetime τ1=380ps/amplitude α1=72% and τ2=3.59ns/amplitude α2=28%) would be thinkable, in principle. But we found no hints for such changes of elastin in diabetes in the literature.

More relevant is the interpretation in connection with the underlying chemical species. The decay time of 380 ps corresponds to free NADH; thus, a lack of free NADH versus protein-bound NADH, which has decay times of several nanoseconds, can be assumed.33 The increased contribution of protein-bound NADH is indicated by the longer τ3 decay time and by the increased amplitude α3 in the short-wavelength channel. This reason for the changed lifetime is unlikely because the maximal excitation of NADH is at 340 nm and the emission maximum is at 460 nm. In earlier studies,17,20 we excited pure NADH at 446 nm and detected a weak fluorescence with a maximum at 530 nm. Because it was unlikely to excite NADH at such long wavelength, we interpreted this fluorescence as result of pollution.17,34 Apparently, the minimal contribution of the NADH fluorescence can be detected by the highly sensitive TCSPC technique.

A shift of NADH from the free to the protein-bound state results from an increased contribution of glycolysis to energy production.35

The elongated fluorescence lifetime τ2 in diabetic subjects, especially in the short-wavelengths channel, is most likely interpretable as an increased contribution of free FAD. Field et al.12 published an increased fluorescence intensity of FAD in diabetic eyes. The applied excitation wavelength and emission range are optimally suited for detection of FAD. In our own measurements, we found the lifetimes of free FAD using a biexponential fit to be τ1=330ps with α1=18% and τ2=2810ps with α2=82%.17 The lifetime of protein-bound FAD is 100 ps.36

Though the number of subjects with an artificial lens is low, the results of these groups demonstrate the influence of the fluorescence of the crystalline lens. This fluorescence contributes to the fluorescence that is detected from the eye. The confocal detection principle only reduces the contribution of the fluorescence of the lens, but does not avoid it completely.37 As the excited volume of the lens is much higher in fundus cameras, using the principle of aperture separation, the contribution of fluorescence of the lens cannot be neglected in fluorescence measurements of the eye. Thus, diabetic metabolic changes can be assumed, which accumulate in the crystalline lens, too. As mitochondria are missing in the lens, the accumulation of AGEs can be assumed in the lens. In in-house measurements, AGEs were excited at 446 nm and emit with a maximum at 520 nm. The corresponding lifetimes using a biexponential fit were τ1=865ps with α1=62% and τ2=4170ps with α2=28%.17

Considering all these aspects, the detected changes in the time-resolved autofluorescence in type 2 diabetes are the result of several metabolic alterations, most probably the accumulation of AGEs in the lens and the production of free FAD and accumulation of AGEs in the neuronal retina at the fundus. A change from free to protein-bound NADH cannot completely be excluded.

In conclusion, FLIO is a promising new method for investigation of the metabolism of the eye based on fluorescence lifetime of endogenous fluorophores. To get information about the fundus fluorescence alone, the fluorescence decay of the lens should be measured separately and be used in the fit of the fluorescence decay, detected from the whole eye. To specify the interpretation of metabolic alterations, techniques should be developed, permitting FLIO measurements in single fundus layers of the living human eye.


This research was supported by the German Federal Ministry of Education and Research (Grant No. 03IPT605A) and the German Research Council (DFG HA 2899/19-1).



Y. Ohkubo et al., “Intensive insulin therapy prevents the progression of diabetic microvascular complications in Japanese patients with non-insulin dependent diabetes mellitus: a randomized prospective 5-year study,” Diab. Res. Clin. Pract. 28, 103–117 (1995).DRCPE90168-8227http://dx.doi.org/10.1016/0168-8227(95)01064-KGoogle Scholar


UK Prospektive Diabetes Study (UKPDS) Group, “Intensive blood-glucose control with sulfonyl ureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33),” Lancet 352, 837–853 (1998).LANCAO0140-6736http://dx.doi.org/10.1016/S0140-6736(98)07019-6Google Scholar


A. J. Barber, T. W. Gardner and S. F. Abcouwer, “The significance of vascular and neural apoptosis to the pathology of diabetic retinopathy,” Invest. Ophthalmol. Vis. Sci. 52, 1156–1162 (2011).IOVSDA0146-0404http://dx.doi.org/10.1167/iovs.10-6293Google Scholar


A. J. Barber et al., “Neural apoptosis in the retina during experimental and human diabetes: early onset and effect of insulin,” J. Clin. Invest. 102, 783–791 (1998).JCINAO0021-9738http://dx.doi.org/10.1172/JCI2425Google Scholar


A. B. El-Remessy et al., “Neuroprotective and blood-retinal barrier-preserving effect of cannabitol in experimental diabetes,” Am. J. Pathol. 168, 235–244 (2006).AJPAA40002-9440http://dx.doi.org/10.2353/ajpath.2006.050500Google Scholar


R. M. H. Diederen et al., “Reexamining the hyperglycemic pseudohypoxia hypothesis of diabetic oculopathy,” Invest. Ophthalmol. Vis. Sci. 47, 2726–2731 (2006).IOVSDA0146-0404http://dx.doi.org/10.1167/iovs.06-0076Google Scholar


M. S. Ola et al., “Analysis of glucose metabolism in diabetic rat retinas,” Am. J. Physiol. Endocrinol. Metab. 290, E1057–E1067 (2006).0193-1849http://dx.doi.org/10.1152/ajpendo.00323.2005Google Scholar


Y. Ido et al., “Early neural and vascular dysfunction in diabetic rats are largely sequelae of increased sorbitol oxidation,” Antioxid. Redox Signal. 12, 39–51 (2010).1523-0864http://dx.doi.org/10.1089/ars.2009.2502Google Scholar


E. Lieth et al., “Retinal neurodegeneration: early pathology in diabetes,” Clin. Exp. Ophthalmol. 28, 3–8 (2000).1442-6404http://dx.doi.org/10.1046/j.1442-9071.2000.00222.xGoogle Scholar


J. H. Yang et al., “Retinal neurodegeneration in type II diabetic Otsuka Long-Evans Tokusshima fatty rats,” Invest. Ophthalmol. Vis. Sci. 54, 3844–3851 (2013).IOVSDA0146-0404http://dx.doi.org/10.1167/iovs.12-11309Google Scholar


R. U. Kowluru and S. N. Abbas, “Diabetes-induced mitochondrial dysfunction in the retina,” Invest. Ophthalmol. Vis. Sci. 44, 5327–5334 (2003).IOVSDA0146-0404http://dx.doi.org/10.1167/iovs.03-0353Google Scholar


M. G. Field et al., “Rapid, noninvasive detection of diabetes-induced retinal metabolic stress,” Arch. Ophthalmol. 126(7), 934–938 (2008).AROPAW0003-9950http://dx.doi.org/10.1001/archopht.126.7.934Google Scholar


F. C. Delori et al., “In vivo fluorescence of the ocular fundus exhibits retinal pigment epithelium lipofuscin characteristics,” Invest. Ophthalmol. Vis. Sci. 36(3), 718–729 (1995).IOVSDA0146-0404Google Scholar


D. Schweitzer, “Ophthalmic applications of FLIM,” Chapter 20 in Fluorescence Lifetime Spectroscopy and Imaging, L. Marcu, P. M. W. French and D. S. Elson, Eds., pp. 423–447, CRC Press, Boca Raton, FL (2014).Google Scholar


Early Treatment Diabetic Retinopathy Study Research Group., “Grading diabetic retinopathy from stereoscopic color fundus photographs—an extension of the modified Airlie House classification. ETDRS report number 10,” Ophthalmology 98, 786–806 (1991).OPANEW0743-751Xhttp://dx.doi.org/10.1016/S0161-6420(13)38012-9Google Scholar


H. W. van Dijk et al., “Early neurodegeneration in the retina of type 2 diabetic patients,” Invest. Ophthalmol. Vis. Sci. 53, 2715–2719 (2012).IOVSDA0146-0404http://dx.doi.org/10.1167/iovs.11-8997Google Scholar


D. Schweitzer et al., “Towards metabolic mapping of the human retina,” Microsc. Res. Tech. 70, 410–419 (2007).MRTEEO1059-910Xhttp://dx.doi.org/10.1002/(ISSN)1097-0029Google Scholar


D. Schweitzer, “Quantifying fundus autofluorescence,” Chapter 8 in Fundus Autofluorescence, N. Lois and J. V. Forrester, Eds., pp. 78–95, Wolters Kluwer Lippincott Williams & Wilkins, Philadelphia, PA (2009).Google Scholar


E. Y. Chew et al., “Evaluation of the age-related eye disease study clinical lens grading system AREDS report no. 31,” Ophthalmology 117, 2112–2119 (2010).OPANEW0743-751Xhttp://dx.doi.org/10.1016/j.ophtha.2010.02.033Google Scholar


L. Deutsch, “Evaluierung des Fluorescence Lifetime Imaging vom Augenhintergrund bei Patienten mit Diabetes mellitus,” MD Dissertation, FSU Jena, Germany (2012).Google Scholar


C. Dysli et al., “Quantitative analysis of fluorescence lifetime measurements of the macula using the fluorescence lifetime imaging ophthalmoscope in healthy subjects,” Invest. Ophthalmol. Vis. Sci. 55, 2106–2113 (2014).IOVSDA0146-0404http://dx.doi.org/10.1167/iovs.13-13627Google Scholar


C. Kloos et al., “Unnoticed shift in the normal range of glycated HbA1c with TOSOH high performance liquid chromatography,” Diabetologia 54(Suppl. 1), 401 (2011).DBTGAJ0012-186XGoogle Scholar


Diabetes Control and Complication Trial Research Group, “The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus,” N. Engl. J. Med. 329, 977–986 (1993).NEJMAG0028-4793http://dx.doi.org/10.1056/NEJM199309303291401Google Scholar


S. Holm, “A simple sequentially rejective multiple test procedure,” Scand. J. Stat. 6, 65–70 (1979).SJSADG0303-6898Google Scholar


M. Klemm, private communication, University of Technology, Institute of Biomedical Engineering and Informatics, Ilmenau, Germany (2010).Google Scholar


M. Klemm et al., “Repeatability of autofluorescence lifetime imaging at the human fundus in healthy volunteers,” Curr. Eye Res. 38, 793–801 (2013).CEYRDM0271-3683http://dx.doi.org/10.3109/02713683.2013.779723Google Scholar


D. Schweitzer et al., “Spectral and time-resolved studies on ocular structures,” Proc. SPIE 6628, 662807 (2007).PSISDG0277-786Xhttp://dx.doi.org/10.1117/12.726701Google Scholar


S. Peters, M. Hammer and D. Schweitzer, “Two-photon excited fluorescence microscopy application for ex vivo investigation of ocular fundus samples,” Proc. SPIE 8086, 808605 (2011).PSISDG0277-786Xhttp://dx.doi.org/10.1117/12.889807Google Scholar


D. Schweitzer et al., “Time-resolved autofluorescence imaging of human donor retina tissue from donors with significant extramacular drusen,” Invest. Ophthalmol. Vis. Sci. 53, 3376–3386 (2012).IOVSDA0146-0404http://dx.doi.org/10.1167/iovs.11-8970Google Scholar


D. Schweitzer et al., “Detection of early metabolic alterations in the ocular fundus of diabetic patients by time-resolved autofluorescence of endogenous fluorophores,” Proc. SPIE 8087, 80871G (2011).PSISDG0277-786Xhttp://dx.doi.org/10.1117/12.889109Google Scholar


S. E. Inzucchi et al., “Management of hyperglycemia in type 2 diabetes: a patient-centered approach: position statement of the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD),” Diabetes Care 35, 1364–1379 (2012).DICAD20149-5992http://dx.doi.org/10.2337/dc12-0413Google Scholar


H. H. Abholz et al., “Nationale Versorgungs Leitlinie Therapie des Typ-2 diabetes,” Dtsch. Arztebl. 110(40), A-1875/B-1655/C-1623 (2013).Google Scholar


H. D. Vishwasrao et al., “Conformational dependence of intracellular NADH on metabolic state revealed by associated fluorescence anisotrophy,” J. Biol. Chem. 280, 25119–25126 (2005).JBCHA30021-9258http://dx.doi.org/10.1074/jbc.M502475200Google Scholar


D. Schweitzer, “Metabolic mapping,” Chapter 10 in Medical Retina, F. G. Holz and R. F. Spaide, Eds., pp. 107–123, Springer, Berlin Heidelberg (2010).Google Scholar


V. Jyothikumar, Y. Sun and A. Periasamy, “Investigation of tryptophan–NADH interactions in live human cells using three-photon fluorescence lifetime imaging and Förster resonance energy transfer microscopy,” J. Biomed. Opt. 18, 060501 (2013).JBOPFO1083-3668http://dx.doi.org/10.1117/1.JBO.18.6.060501Google Scholar


M. C. Skala et al., “In vivo multiphoton microscopy of NADH and FAD redox states, fluorescence lifetimes, and cellular morphology in precancerous epithelia,” PNAS 104(49) 19494–19499 (2007).PNASA60027-8424http://dx.doi.org/10.1073/pnas.0708425104Google Scholar


D. Schweitzer, M. Hammer and F. Schweitzer, “Limitations of the confocal laser scanning technique in measurements of time-resolved autofluorescence of the ocular fundus,” Biomed. Tech. (Berl.) 50, 263–267 (2005).Google Scholar


Dietrich Schweitzer received his Dipl.-Ing. in 1967, Dr.-Ing. in 1976, and the habilitation degree in 1990 from the Technical University of Ilmenau, Germany. From 1979, he was with the University of Jena, Germany, and was the head of experimental ophthalmology (EO) until his retirement in 2008. In 1995, he was appointed to a docent of EO. He developed a method for measuring the oxygen saturation, the optical density of xanthophyll, and was the founder of FLIO.

Lydia Deutsch: Biography is not available.

Matthias Klemm received his master’s degree in biomedical engineering from the Technische Universität Ilmenau. He is currently involved in the development of new methods for the analysis of time-resolved fluorescence data from the human eye. Present interests also include electrical current stimulation of the human eye and dynamic vessel analysis.

Susanne Jentsch has worked as a scientist in the Jena University Hospital, Department of Ophthalmology, since 2007. After receiving a graduate degree of optometry and vision science in 2007 at the ERNST-ABBE-University of Applied Science in Jena, she worked on clinical-experimental studies with focus on macular pigment density in age-dependent macular degeneration (AMD) as well as fluorescence lifetime imaging in AMD, glaucoma and Alzheimer ’s disease.

Martin Hammer studied physics at the University of Jena and received his diploma degree in 1983 and his PhD in 1998. From 1983 to 1990, he was with Carl Zeiss Jena. Since 1990, he has been with the University of Jena, Department of Ophthalmology. Currently, he is head of the experimental ophthalmology group. His recent research interests are functional imaging of the ocular fundus, retinal blood flow and oximetry, as well as autofluorescence lifetime imaging.

Sven Peters studied biochemistry at the University of Jena and received his diploma in 2005. Since he started his PhD he has been fascinated working with fluorescence lifetime imaging. Using this technique he investigated protein-protein interactions on microarrays at the Institute of Photonic Technology Jena. In 2010 he joined the experimental ophthalmology group at the University Hospital Jena and focuses now on early diagnostics of retinal disorders, using autofluorescence lifetime and anisotropy imaging.

Jens Haueisen received his MS in 1992 and his PhD in 1996 in electrical engineering from the Technical University Ilmenau. From 1998 to 2005, he was the head of the Biomagnetic Center, Friedrich-Schiller-University, Jena, Germany. Since 2005, he has been a professor of biomedical engineering and directs the Institute of Biomedical Engineering and Informatics at the Technical University Ilmenau. His research interests include the investigation of active and passive bioelectric and biomagnetic phenomena and medical technology for ophthalmology.

Ulrich A. Müller studied medicine from 1974 to 1980 at the University of Jena. He graduated in 1982 with a degree in pathological anatomy, in 1995 as Doctor habilitatus in diabetology, and in 1993 with a Master of Science in clinical biochemistry from the University of Aberdeen. In 2002, he was professor of internal medicine, and in 2003 head of the Department of Endocrinology and Metabolic Diseases at University Hospital Jena, Germany. He has done studies on diabetes health care delivery, diabetic late complication, insulin therapy, patient education.

Jens Dawczynski studied medicine from 1990 to 2001 at the University of Jena. He received his MD in 2001 and habilitation degree and postdoctoral lecture qualification in ophthalmology in 2008. In 2011, he was appointed as professor of fundus imaging at University of Leipzig, Department of Ophthalmology. He has been the associate director and senior physician in this department. He leads studies on diabetic retinopathy, retinal vessel occlusion, and age-related macular degeneration.

Dietrich Schweitzer, Lydia Deutsch, Matthias Klemm, Susanne Jentsch, Martin Hammer, Sven Peters, Jens Haueisen, Ulrich A. Müller, Jens Dawczynski, "Fluorescence lifetime imaging ophthalmoscopy in type 2 diabetic patients who have no signs of diabetic retinopathy," Journal of Biomedical Optics 20(6), 061106 (13 March 2015). http://dx.doi.org/10.1117/1.JBO.20.6.061106
Submission: Received ; Accepted


Picosecond phenomena


Fluorescence lifetime imaging


Control systems


Back to Top