2.1.

Time-Resolved Fluorescence Measurements

The instrumentation for the time-resolved spectroscopic fluorescence measurements has been described in detail elsewhere.³⁰ Briefly, the excitation light pulses (typical pulse duration about $500\mathrm{ps}$ , repetition ratio $10\mathrm{Hz}$ ) of a nitrogen laser-pumped dye laser emitting at $405\mathrm{nm}$ (MSG 803TD, LTB GmbH, Berlin, Germany) were injected into a $550\text{-}\mu \mathrm{m}$ -core-diam optical fiber using a UV-coated lens and a dichroic long-pass filter (cutoff wavelength $450\mathrm{nm}$ , $45\text{-}\mathrm{deg}$ arrangement). The fiber could be easily inserted into the working channels of conventional flexible bronchoscopes, and its flat distal cut-end was applied in direct gentle contact with the lesions or normal respiratory mucous membrane. This geometry, combined with the limited penetration depth of violet $\left(405\mathrm{nm}\right)$ light, defines a probed volume of typically $0.1{\mathrm{mm}}^3$ . The laser light delivered by the fiber excited the mucosa AF that was collected by the same fiber. The fluorescence signal was transmitted through the dichroic filter and injected into a spectrograph (250IS, Chromex, Albuquerque, New Mexico). An additional $430\text{-}\mathrm{nm}$ long-pass filter positioned in front of the spectrograph removed the remaining backscattered excitation light, and an entire time-resolved spectrum could thus be acquired by means of a streak camera (C4334, Hamamatsu Photonics K.K., Hamamatsu City, Japan) that was operated in the analog mode in the $20\text{-}\mathrm{ns}$ time window. The spectral resolution was approximately $15\mathrm{nm}$ , and the temporal resolution was $280\mathrm{ps}$ . Each individual measurement consisted of the integration of 150 acquisitions (see the following below). The entire setup was installed on a trolley for transport to the procedure room.

Measurements performed on a reference sample (quartz cuvette for optical spectroscopy containing $3\mathrm{ml}$ of rhodamine 6G, $10\mu \mathrm{M}$ in water) were carried out prior to the endoscopic analysis in order to compensate for the variations of the excitation light intensities, optical alignment, and detector performance. The AF intensities were subsequently normalized by the signal intensity of this reference and expressed in relative units (r.u.). This procedure allowed a comparison of the fluorescence intensities from different patients. This reference sample was also used to validate our fluorescence lifetime measurement system and data analysis procedure. As reported in detail by Glanzmann,³⁰ the lifetime measured on this reference sample was compatible with the literature. In addition, a mono-exponential decay was obtained with an excellent reproducibility (variations on the order of 2%) on this sample. Our group also performed measurements on reference samples containing a mixture of rhodamines. The individual fluorescence lifetimes could be identified with a higher than 20% precision using our fluorescence lifetime measurement system and data analysis procedure. Last, it should be noted that similar measurements were performed on various luminescence standards exhibiting luminescence lifetimes ranging between 0.175 and $3700\mathrm{ns}$ . The excellent agreement between these measurements and the literature indicate that the response of our instrument is reliable within this time range.³⁰ However, it is worth noting here that the shortest measured lifetimes, the component ${\tau }_1$ defined below, were inferior to the width of the excitation laser pulse and were thus likely to be biased. For this reason, these shortest lifetimes were not taken into account in the comparison, presented here, of the results from the spectrometric and imaging studies.

2.2.

The AF Imaging Device

The AF imaging device has been described in detail elsewhere.¹² Briefly, it consisted of a filtered endoscopic light source and a filtered 3 CCD endoscopic camera (both from Richard Wolf GmbH, Knittlingen, Germany). The light source contained a $300\text{-}\mathrm{W}$ xenon lamp and was equipped with a flip-flop filter holder containing (1) a custom-made blue-violet bandpass filter $\left(390\mathrm{nm}\text{to}460\mathrm{nm}\right)$ for AF excitation, and (2) a grid for conventional white-light (WL) endoscopy. The light was led to the endoscope optics via a liquid light guide, and the endoscopic image (WL and AF) was detected by the 3 CCD camera. A $490\text{-}\mathrm{nm}$ long-pass filter in the camera objective rejected all blue-violet excitation light. When the light source was employed in the AF mode, i.e., emitting blue-violet light, the blue color channel of the camera was suppressed. Thus, the AF images obtained with this system contained only the green $\left(500\mathrm{nm}\text{to}590\mathrm{nm}\right)$ and red $(>600\mathrm{nm})$ color information. The camera head could be clipped to conventional rigid bronchoscopes or to a dedicated bronchofiberscope. The bronchoscopic images were displayed on a video monitor (Sony, Japan) and recorded with a video recorder to enable their subsequent analysis by a computer after digitization. This procedure provided quantitative information regarding the green and red AF intensities. Due to the spectral and intensity changes mentioned earlier, healthy bronchial mucosa appeared greenish (AF negative), whereas (pre-)neoplastic lesions presented a brown-reddish color (AF positive).

2.3.

In Vivo Measurements

Eleven high-risk lung cancer patients undergoing conventional white-light bronchoscopy (WLB) and diagnostic AFB were included in the present study. A total of 14 bronchoscopies, referred to in the following as a case, were performed on these patients. The inclusion criteria were positive sputum cytology, positive resection margins following previous lung cancer surgery, and routine follow-up bronchoscopy. Both WLB and AFB were performed with the AF bronchofiberscope described earlier, and the locations of AF-positive as well as AF-negative, i.e., healthy, sites were identified. For the time-resolved measurements, the optical fiber of the spectrofluorometer was inserted into the working channel of the bronchoscope and placed in gentle contact with the mucous membrane under visual control. The acquisition time was approximately $15\mathrm{s}$ . For each measurement, two acquisitions were performed on a single spot without moving the fiber, with the aim of verifying whether the contact with the mucosa had been established through the entire duration of the measurement. The two acquisitions were considered void if they were not identical. Whenever possible, measurements were repeated on the same site after slightly displacing the fiber in order to account for intrapatient variations due to local heterogeneities of the bronchial mucosa. Measurements were performed on all lesions identified with the AFB imaging system. Additional measurements on the healthy mucosa were carried out at various locations in the bronchial tree so as to assess the anatomy-dependent intra- and interpatient variations. More precisely, measurements were performed on the carina, the walls of the main bronchi, as well as on the spurs of the upper-and lower-lobe bronchi. Biopsies were taken with WLB or AFB on suspect sites, and the presence or absence of (pre-)neoplasia was evaluated by subsequent histopathological analysis. The whole procedure was recorded on video. In histopathologic analysis, the specimens were classified into the following three categories: (1) reactive change including basal cell hyperplasia and squamous metaplasia; (2) preneoplasia without invasion of the basement membrane, including mild, moderate, and severe dysplasia, as well as CIS; and (3) neoplasia, including microinvasive carcinoma and invasive squamous cell carcinoma.

2.4.

Lifetime Analysis

The fluorescence decays were analyzed using the Marquardt-Levenberg iterative nonlinear least-squares reconvolution fitting algorithm, assuming an a priori multiexponential decay (Software HPD-TA-Fit 2.1.3, Hamamatsu Photonics Germany GmbH, Herrsching, Germany). The signal-to-noise ratio of the obtained decays enabled a fit with up to three exponentials. The resulting curves of the weighted residuals as well as the autocorrelation of the residuals versus channel number are shown for graphical assessment of the goodness of fit (Fig. 1 ). Good fits should yield randomly distributed residuals about zero. The presence of any pattern in the distribution of either the residuals or the autocorrelation function is an indication that the model used is inadequate to describe the data. Bad fits provide low-frequency periodicity in an autocorrelation plot that can be easily detected.³⁵ The AF decays were analyzed for two wavelength ranges $510\mathrm{nm}\text{to}550\mathrm{nm}$ and $600\mathrm{nm}\text{to}640\mathrm{nm}$ , corresponding to the green and red spectral detection ranges, respectively, of the AFB imaging device. Figure 1 shows the AF decay between $510\mathrm{nm}$ and $550\mathrm{nm}$ on healthy bronchial mucosa together with the excitation pulse and the curves of both the residuals and the autocorrelation function. As demonstrated in the figure, the fits with three decay parameters gave rise to curves of both the residuals and the autocorrelation function. The slight modulation of both curves as well as a weak odd-even effect was due to the response characteristics of the streak camera. As can be seen on the left part of the autocorrelation curve, the range of the fit could not be extended all the way to the beginning of the decay profile. This was probably the result of a slight difference in pulse and decay profile due to the nonlinearity of the so-called cross talk of the streak camera. Nevertheless, the fitted results were very stable, and the 3-exponential fits gave rise to considerably enhanced residual curves as opposed to their 2-exponential counterparts. The 4-exponential fits, on the other hand, led to unstable results, and no improvement of the residuals was obtained.

Fig. 1

A 3-exponential least-square deconvolution fit of an AF decay in the range of $510\text{to}550\mathrm{nm}$ . Upper panel: The dashed line depicts the excitation pulse at $405\mathrm{nm}$ , whereas the black solid line and the gray patterned line represent the decay data and fit function, respectively. The lower figures display the autocorrelation function as well as the residuals.

The relative photonic weight of each lifetime was calculated using the expression ${\alpha }_i\ast {\tau }_i∕({\sum }_i{\alpha }_i\ast {\tau }_i)$ , where ${\tau }_i$ is the AF lifetime, and ${\alpha }_i$ is a preexponential factor. In order to characterize the speed of fluorescence decay with a single parameter, the 10%-decay-time, designating the time period during which the fluorescence decreased from its maximum value to 10% of its intensity, was introduced. This approach enabled, in typical conditions, a quantification of the differences in decay time at the center of the decay curve, since the time-window of the streak camera was $20\mathrm{ns}$ , as can be seen in Fig. 1.

2.5.

AF Image Analysis

For each case, one WL and one AF image per spectroscopically measured and biopsied site were digitized in the $24\text{-}\text{bit}$ RGB color scheme. The average intensity of the red (R) and green (G) pixels of the biopsied lesion as well as of the presumably healthy surrounding mucosa (reference) were computed together with their standard deviations. These values were used to compute (1) the intensity ratios in the green and the red color channels {ratio $R_{\text{green}\left(\text{red}\right)}^{\prime }$ between the green (red) intensities on the lesion and the corresponding intensities on the reference area [Eq. 1]} and (2) the normalized red-to-green color ratio $R_{\mathrm{norm}}^{\prime }$ between the lesion and the reference area [Eq. 2]:

Equations 1, 2 show the computation of the red and green intensity ratios $R^{\prime }$ [Eq. 1] and of the normalized color ratio $R_{\mathrm{norm}}^{\prime }$ , where $R_{\text{lesion}}$ $\left(G_{\text{lesion}}\right)$ and $R_{\text{reference}}$ $\left(G_{\text{reference}}\right)$ represent the red (green) pixel intensities calculated from the lesion and the reference area, respectively.

3.1.

Intra- and Interpatient Variations of the Lifetime Parameters

The intrapatient variations of the AF lifetime due to mucosa heterogeneities and changes in the anatomical structure were assessed from 14 cases, including 11 patients as described in Sec. 2. Figure 2 compares the calculated AF lifetimes measured from healthy mucosa at various locations of the tracheo-bronchial tree during the 14 exams against their normalized relative photonic weight. All lifetimes extracted from decays measured during a single exam are indicated with the same sign. This renders it possible to compare the intra- and the interpatient variation.

Fig. 2

Lifetimes and normalized photonic weights of the AF decays on the healthy bronchial mucosa of 11 patients between 510 and $550\mathrm{nm}$ at $405\text{-}\mathrm{nm}$ excitation. (Each patient corresponds to one symbol.)

The mean values of three AF lifetimes, ${\tau }_1$ , ${\tau }_2$ , and ${\tau }_3$ , were $0.17\pm 0.07\mathrm{ns}$ , $2.02\pm 0.13\mathrm{ns}$ , and $6.84\pm 0.39\mathrm{ns}$ , respectively, with variations on the order of 39%, 6.4%, and 5.7%, respectively. The corresponding variations for the photonic weights were 17.9%, 4.0%, and 5.9%, respectively. Approximately 50% of the fluorescence stems from the long lifetime ${\tau }_3$ , about 40% from the intermediate lifetime ${\tau }_2$ , and close to 10% from the short lifetime ${\tau }_1$ .

The mean intrapatient variations in the fluorescence lifetimes were on the order of 17.3% (range 6.5% to 35.4%) for the shortest lifetime ${\tau }_1$ and approximately 4% (range 1.5% to 7%) for the longer lifetimes ${\tau }_2$ and ${\tau }_3$ . The corresponding intrapatient variations of the photonic weights ranged from 3.5% to 7.7%. Although the measurements of the individual patients were somewhat grouped, the interpatient variations were no more pronounced than their intrapatient counterparts. As already stated, the intrapatient variations result partially from mucosa heterogeneities, but also from changes in the AF of the tissues related to the anatomical structure of the bronchi.

The variations in AF lifetime due to mucosa heterogeneities are illustrated by the open triangles and open diamonds in Fig. 2, representing seven and five measurements, respectively, conducted at neighboring spots on the healthy carina of two patients. These intrapatient variations were on the order of 35%, 3.8%, and 3.9% for the lifetimes ${\tau }_1$ , ${\tau }_2$ , and ${\tau }_3$ , respectively.

Figures 3a and 3b show the distribution of the AF lifetimes from healthy bronchial mucosa for the green [ $510\text{to}550{\mathrm{nm}}^3$ ; Fig. 3a] and red [ $600\text{to}640\mathrm{nm}$ ; Fig. 3b] spectral regions, respectively. The figures represent data from a subgroup of eight patients for which data in both spectral regions were available. The mean AF lifetimes computed for the fluorescence decays in the red spectral region ( ${\tau }_1$ : $0.25\pm 0.08\mathrm{ns}$ , ${\tau }_2$ : $2.06\pm 0.20\mathrm{ns}$ , ${\tau }_3$ : $6.98\pm 1.08\mathrm{ns}$ ) were relatively similar to those in the green spectral region ( ${\tau }_1$ : $0.15\pm 0.06\mathrm{ns}$ , ${\tau }_2$ : $2.04\pm 0.19\mathrm{ns}$ , ${\tau }_3$ : $6.97\pm 0.41\mathrm{ns}$ ). [95% confidence interval (CI) and p-value of the "red-green" lifetimes populations differences: ${\tau }_1$ : (0.065 to 0.135; $p<0.05$ ), ${\tau }_2$ : ( $-0.076$ to 0.116, $p>0.05$ ), ${\tau }_3$ : ( $-0.392$ to 0.411, $p>0.05$ ).] In addition, the figures show the correlation between the AF lifetime and the anatomic localization of the measured site. No dependence on the location in the bronchi of either the AF lifetimes or the relative photonic weight was observed in either of the spectral regions. Moreover, the changes in the AF lifetimes originating from measurements performed at various locations in the bronchi were comparable to the AF lifetime variations due to the heterogeneity of the mucosa response, as shown in Fig. 2.

Fig. 3

Photonic weights versus AF lifetimes measured on healthy mucosa on (1) the main carina (squares); (2) superior spurs (open circles); (3) inferior spurs (circles); and (4) the walls of the stem bronchi (triangles). Figure 3a shows the results for the green spectral region $\left(510\mathrm{nm}\text{to}550\mathrm{nm}\right)$ , whereas Fig. 3b displays those for the red spectral region $\left(600\mathrm{nm}\text{to}640\mathrm{nm}\right)$ .

Due to the similarity of the principal AF lifetimes in the green and the red spectral regions, all data presented in the following paragraphs refer exclusively to the green spectral region. This approach could also be justified by the fact that the decrease in intensity observed in spatially resolved spectrofluorometry^{8, 23, 32} and imaging^{11, 12, 16} was most important in the green spectral region and was thus the principal cause for the AF contrast.

It is worth noting that although the lifetimes did not depend on the localization of the measurement spot within the bronchial tree, the fluorescence intensities did. At wavelengths ranging from $510\mathrm{nm}\text{to}550\mathrm{nm}$ , the fluorescence intensities measured on the spurs of the upper and lower bronchi were twofold lower (relative intensities $I_{\mathrm{rel}}=42.6\pm 15.8$ and $33.6\pm 19.6$ , respectively) than those measured on the carina $(I_{\mathrm{rel}}=81\pm 37.2)$ and on the wall of the main bronchi $(I_{\mathrm{rel}}=79.5\pm 38.8)$ ; data not shown.

3.2.

AF Lifetimes of (Pre-)Neoplastic Lesions

A total number of eight lesions were detected by AF bronchoscopy in 5 of the 11 patients. All lesions were invisible with conventional white-light bronchoscopy. Histopathologic analysis of the biopsies taken from the lesions confirmed five (pre-)neoplasia (three severe dysplasia and two microinvasive carcinoma) and three reactive changes (squamous metaplasia). The former are here referred to as true positives (TP) and the latter as false positives (FP). The photonic weights of the various AF lifetimes detected in the green spectral region for the TP lesions (solid symbols) and the surrounding healthy mucosa (open symbols) are presented in Fig. 4 .

Fig. 4

Lifetimes and normalized photonic weights (wavelength range $500\mathrm{nm}\text{to}590\mathrm{nm}$ ) of the autofluorescence decays on three severe dysplasia, two microinvasive carcinoma, and healthy mucosa measured in five cases. Each symbol corresponds to one measurement. Solid symbols depict measurements on TP lesions, whereas open symbols refer to measurements on surrounding healthy mucosa.

Figure 4 does not demonstrate a systematic difference in the photonic weight or AF lifetime computed for healthy mucosa and (pre-)neoplasia, respectively. This was confirmed by statistical analysis giving the 95% CI and the $p$ -values for the three lifetimes as follows: ${\tau }_1$ (95% CI: $-0.0285$ to 0.085, $p>0.05$ ), ${\tau }_2$ (95% CI: $-0.0282$ to 0.1482, $p>0.05$ ), and ${\tau }_3$ (95% CI: $-0.0209$ to 0.5609, $p>0.034$ ). The results from the AF lifetime measurements were compared to the color aspect and intensity ratios computed from the AFB images. For four of the five (pre-)neoplasias (two severe dysplasia and two microinvasive carcinoma), a complete set of data, including AFB and WLB images, as well as AF lifetime measurements for both the lesion and its surrounding healthy mucosa, was available for analysis. Figures 5a and 5b display the AF lifetimes in the green spectral region versus the intensity ratios $R^{\prime }$ and $R_{\mathrm{norm}}^{\prime }$ . Although the ratios computed from the AFB images clearly show evidence of a spectral and intensity contrast between the lesions and their surrounding healthy mucosa, no difference in the AF lifetimes measured on the respective sites could be observed.

Fig. 5

AF lifetimes (wavelength range $500\mathrm{nm}\text{to}590\mathrm{nm}$ ) ${\tau }_2$ and ${\tau }_3$ of four true positive (TP) lesions and three false positive (FP) lesions found in two patients versus the intensity ratios as computed from the AFB images. (a) AF lifetimes measured on the lesion versus the green ( $R^{\prime }$ , diamonds) intensity ratios between TP (solid symbols) and FP (open symbols) lesions and healthy mucosa (gray circles). (b) AF lifetimes measured on the lesion versus the normalized $R_{\mathrm{norm}}^{\prime }$ intensity ratio for TP (solid squares) and FP (open squares) lesions. The gray circles depict the AF lifetime measured on healthy mucosa versus the $R_{\mathrm{norm}}^{\prime }$ for the same sites.

As mentioned earlier, it is worth noting here that the shortest measured lifetimes ${\tau }_1$ were inferior to the width of the excitation laser pulse and were thus likely to be biased. For this reason, the shortest lifetimes were not taken into account in the comparison of the results from the spectrometric and imaging study.

Figure 6 displays the relative peak AF intensities measured between $510\mathrm{nm}$ and $550\mathrm{nm}$ as functions of the 10% decay time for 47 measurements performed on the healthy mucosa of 11 patients. Each symbol depicts values measured at various locations in the same patient. As described earlier, the 10% decay time permitted a characterization of the speed of the AF decay by a single parameter.

Fig. 6

The relative peak fluorescence intensity between $510\mathrm{nm}$ and $560\mathrm{nm}$ versus the 10% decay time measured on healthy bronchial mucosa in 11 patients. (Each symbol corresponds to one patient.)

The relative inter- and intrapatient variations in the AF intensities were 56% and 40%, respectively. Furthermore, the inter- and intrapatient variations of the 10% decay time were on the order of 10% and 5.4% (range $7\mathrm{ns}\text{to}11.5\mathrm{ns}$ ), respectively, and thus markedly smaller than the intensity variations. The measurements of the relative fluorescence intensity with an optical fiber were extremely difficult to perform on small spurs due to the patient’s heartbeats and the often complex geometries in the bronchi. It was thus first believed that the analysis conditions were at the origin of the significant AF intensity variations. However, the measurements conducted at the easily accessible carina of the same patient—shown in Fig. 2 (open triangles)—demonstrated that the local heterogeneities in the bronchial wall in terms of intensities and decay parameters were as high as the intrapatient variation resulting from measurements on other, less accessible locations in the bronchi. This contradicted the idea that poor measurement conditions were responsible for the fluctuations observed. The 10% decay time and AF intensities from the measurements on the carina are also displayed as open triangles in Fig. 6.

Since the variations in the 10 % decay time were small and very poorly correlated with the intensity, the intensity variations could not be attributed to fluorescence quenching, but were rather believed to be due to changes in the concentrations of fluorochromes and/or absorber or to variations in the structure of the bronchial wall.