2.1.

Fluorescence Excitation

The excitation of the autofluorescence is achieved by means of a filtered 300-W Xe lamp (5133 Combilight PDD: Wolf Endoskope, Knittlingen, Germany) in the violet-blue range (390 to 460 nm, custom-made filters). The lamp is also equipped with an IR blocking filter. The light is transmitted from the lamp to the endoscope via a liquid light guide (Type 4070.253: Wolf Endoskope, Knittlingen, Germany). The available excitation power at the distal end of a standard rigid bronchoscope (0 deg, Type 8465.30 endoscope: Wolf Endoskope, Knittlingen, Germany) is about 260 mW in the specified spectral range.

A dedicated flexible bronchoscope was also developed for this application (type 7265.006 bronchofiberscope equipped with UV illumination fibers: Wolf Endoskope, Knittlingen, Germany), enabling an excitation power of about 120 mW at its distal end in the same spectral domain.

2.2.

Fluorescence Detection

The photodetection system relies on one black and white (B/W) CCD camera (Type: CF 8/1 FMC; Kappa Messtechnik GmbH, Gleichen, Germany) and a dual-detection range: 500 to 590 nm (hereafter the green region) and 600 to 700 nm (hereafter the red region). In fact, at least two spectral domains are necessary for efficient contrast enhancement. However, detection over many spectral regions is shot-noise limited. A device, clipped at the eyepiece of a standard bronchoscope, framed the optical elements to separate the fluorescence light into its color components (dichroic mirror), to filter the unwanted light (long-pass filter, color filters), and to focus the two resulting light streams (one for the green region and one for the red region) on the same microhead B/W CCD camera. The signal was then processed by a delay line, which first duplicated the signal and then delays the two resulting signals (one for the green component and one for the red) from a given amount of time (13 μs for the green component; 51 μs for the red component), thus enabling the superimposition of the two components in false colors on a Sony trinitron color monitor (Type: PVM-2053 MD) (Figure 1). The gain and offset of the delayed signals can also be adjusted individually. The offset was adjusted to zero, and the gain levels of both the red and green images were adapted according to the following procedure. The endoscope was introduced in the trachea and maintained at a typical working distance to the carina. The gain of the red channel was then increased to generate a red image whose brightness corresponded to half the dynamic range of the system. The gain of the green channel was then enhanced to a level resulting in a green-yellowish appearance for the healthy tissues of the carina on the result image. The system was completed with standard VCRs, which enabled the recording of the raw B/W signal and the resulting false color signal. Note that a switchable mirror also enabled observation of the endoscopic site under standard white light illumination with a standard color camera.

Figure 1

Diagram of the photodetection apparatus.

Typical autofluorescence (AF) raw fluorescence images (B/W), the processed images and the corresponding white light (WL) image obtained with a flexible bronchoscope on a healthy tissue are presented in Figure 2.

Figure 2

Photodetection of a healthy tissue with a flexible bronchoscope.

An advantage of this system over existing ones is its ability to work at video frequency (no frame accumulation necessary) and to switch rapidly between the violet and WL illumination. In fact, the AF images are bright enough to enable the clinician to navigate in the tracheobronchial tree with the possibility to rapidly observe a suspicious site in a conventional way. Finally, this system is compact, reliable (no computer is used as long as the threshold calculation described later is not necessary), and relatively inexpensive as it avoids the use of costly devices such as image intensifiers and nonstandard components.

2.3.

Patients

Twenty patients with high lung cancer risk factors (positive sputum cytology, positive resection margin following lobectomy, patients within the regular follow-up of lung cancers) underwent WL and AF bronchoscopy. Thirty-three biopsies were taken at macroscopically suspicious (WL or AF positive) and control sites. These biopsies were analyzed using standard procedures by the Histopathology Unit of the Lausanne’s CHUV Hospital.

Photodetection took place in the following manner. A first survey of the bronchi was undertaken by the endoscopist under standard WL. This was intended to assess the state of the mucosa and the presence or absence of notable lesions under white light. The bronchi were then examined again under AF excitation light. Care was taken not to damage the mucosa during these examinations. A biopsy was taken on each suspicious WL or AF site and the results were compared to the histopathological analysis. Reddish brown sites under AF light excitation were considered positive and greenish sites negative.

Different AF and WL images corresponding to different histopathological statuses are presented in Figures 3 4 5 6 7. Figure 7 presents a photodetection of an hyperplasia in a 61-year-old male patient. The patient underwent a bronchoscopy because of an extrinsic compression of the lower and middle right lobes. Hyperplasia is not a premalignant or early lesion, but the site appears positive in AF mode (and negative in WL mode). Nevertheless, note that as compared to Figures 3 4 5 6, the color aspect of the suspicious area is slightly different: it appears less reddish. This case is a false positive result, if the decisional algorithms presented later are not used.

Figure 3

Photodetection of a low-grade dysplasia on a 62-year-old male patient (flexible bronchoscope).

Figure 4

Photodetection and spectroscopy (spectra are normalized at 550 nm) of a low-grade dysplasia on a 62-year-old male patient (flexible bronchoscope).

Figure 5

Photodetection of an infiltrating epidermoid carcinoma on a 71-year-old male patient (flexible bronchoscope).

Figure 6

Photodetection of a moderate dysplasia on a 71-year-old male patient (flexible bronchoscope).

Figure 7

Photodetection of a hyperplasia on a 61-year-old male patient (rigid bronchoscope).

According to this result, the question of the assessment of the positivity of a site is of crucial interest. One approach is to find a way to quantify the positivity of the AF sites and to define a threshold of positivity. Indeed, some sites appear reddish and some other sites appear more orange-looking and, as we see later, this is correlated to the actual status of the tissue.

Several false positive results occurred during the early stages of our study and hinted at the necessity to quantitatively assess the “redness” of a spot and not just rely on the subjective visual assessment of the clinician. This is especially true since we initially lacked quantitative decision criteria and hence sometimes the reddest spot in the bronchi of one patient was assigned a positive result, even though it might have not been considered positive in another patient. This demonstrates that a quantitative approach using decision functions is not only useful, but also necessary for maximal efficiency of the method.

2.4.

Threshold Approach

We quantified the green and the red contribution in the images taken with our system and then calculated the resulting green/red ratio. These ratios were plotted against the histopathological status of the tissue and the result is given in Figure 8.

Figure 8

Green to red ratio of the healthy group (curve 1, group 1), the nonhealthy group (metaplastic/inflammatory: curve 2, group 2), and of the cancer group (malignant: curve 3, group 3) tissues as a function of the histopathological status (see Table 1). The errors bars represent the 67 confidence interval. The curves on the left axis are the projected normal distribution curves corresponding to the green/red ratios.

The points were calculated in the following way. For each site to be biopsied, an AF image was recorded with our system under broadband violet-blue light excitation. The resulting images were digitized using 8-bit quantification. The average intensity (and the corresponding standard deviation) of the green and red pixels was computed on the site of the biopsy and next to it on presumably healthy tissue. This is the reason why 33×2=66 points are presented in Figure 8. The corresponding green and red intensities were then divided, respectively, by the total green and the total red intensities recorded on one image of presumably healthy tissue to take into account the fact that the green and red gains are set independently for each patient (normalization). The green/red ratio of these normalized green and normalized red contributions was computed, together with its standard deviation and plotted versus the histopathological status of the site as given by the histopathologist. For the sake of the readability of Figure 8, the histopathological status has been coded with a number. These numbers are given in Table 1.

Table 1

The histopathological status codes have been separated in three groups. Healthy tissues are assigned to the first group (healthy group: green points). Chronic inflammation and metaplasias have been assigned to the second group (nonhealthy group: orange points). This intends to group together the tissue changes that are not malignant. The third group (cancer group: red points) includes the dysplasias and cancers. This grouping could be justified in a system of optical biopsy.

Figure 8 summarizes the quantification of the autofluorescence changes observed during our preliminary study. Each point is given with its standard deviation. This is due to the fact that the color contribution of the green and red components are not fully homogenous. They are fairly stable, however, as we can see in the small error bars. Assuming that these points represent a normal distribution, a curve can be drawn to express it. This has been done for each of the three groups (healthy group, curve 1, group 1; nonhealthy group, curve 2, group 2; cancer group, curve 3, group 3) and is represented in Figure 8 as three projected curves on the left side of the graph. We can see that the three curves are actually separated. Nevertheless, the overlap of the healthy group (μ=1.018, σ=0.143) and the nonhealthy group is nonnegligible and so is also the overlap of the nonhealthy group (μ=0.728, σ=0.103) and the cancer group (μ=0.575, σ=0.121) . The aim of such a quantification, however, is to fix an optimal threshold at an objective value to distinguish early cancers from nonmalignant changes. As the curves overlap, it is likely that there will always be some wrong results.

Nevertheless, for our study, the thresholds have been set to the values where the healthy and nonhealthy curves cross for the healthy threshold (value of 0.85) and where the cancer and nonhealthy curves cross for the cancer threshold (value of 0.66). This involves that there is still an error of 25.4 (=0.66 σ) for a nonhealthy case to be in fact a cancer case, as well as an error of 4.9 (=1.96 σ) for a healthy case to be a cancer case, according to this preliminary study. Nevertheless, an analysis based on the positive predictive value (PPV) was performed to evaluate the efficiency of the method.

3.1.

Illustrations

Figure 3 presents a photodetection of a light dysplasia in a 62-year-old male patient. This patient underwent a bronchoscopy following a positive esophageal cytology (high-grade dysplasia). The presence of low-grade epithelial dysplasia in the bronchi was confirmed by the histopathological analysis. On this patient, we also made a spectroscopic measurement with our spectrofluorometer.¹ For this special measurement, the excitation light used was the broad violet-blue excitation light (390 to 460 nm) of the AF detection device. Two spectroscopic measurements were made: one on the lesion and one on the healthy tissue next to the lesion. The two spectra have been normalized at 550 nm to compare their shape. The resulting spectra are shown on Figure 4, associated with enlarged pictures of the measured area: the red spectrum is associated with the lesion (corresponding to red circles on photos), and the yellow one with the healthy tissue (yellow circles on photos). The decrease of the fluorescence signal intensity in the green region (on the lesion) is therefore seen as an increase in the red one. Moreover, in the raw image (B/W) of Figure 3, one can clearly observe that this intensity decrease in the green region is associated with an intensity decrease in the red region to a lesser extent. This is in full agreement with our spectroscopic observation.¹

Figures 5 and 6 present a photodetection of a 71-year-old male patient. This patient underwent a control bronchoscopy 4 months after a lobectomy (right upper lobe). He also had undergone a photodynamic therapy⁷ (PDT) in the left upper lobe 3 months earlier.

Two suspicious sites were found in the bronchi of this patient, one on the lobectomy scar (Figure 5) and one on the site of the previous PDT (Figure 6). Two biopsies were taken, one on each site. The histopathological analysis of the first biopsy (lobectomy scar, Figure 5) confirmed the presence of an infiltrating little-differenciated epidermoid carcinoma. Note that this site appeared erythroplasic under WL illumination. The second biopsy (on the site of the PDT, Figure 6) was analyzed as a moderate dysplasia. This site appeared also suspicious in WL mode, but the AF mode enabled us to clearly see the delineation of the suspicious area. This enabled us to retreat precisely this area by PDT, as the previous treatment had probably been underestimated.

3.2.

Overall Results

Thirty-three biopsies altogether were taken on sites that were suspicious (positives) during visual inspection of the bronchi in AF or WL modes on the 20 patients involved in this study. Nine of them were confirmed to be true AF positives by histopathological analysis. Thirteen of these 33 sites were positive in WL mode, whereas only 5 of them turned out to be true positives. The qualitative nature of the visual inspection in AF as well as in WL modes has the consequence that the number of false positives depends on the physician’s learning curve. Therefore, we added the aforementioned decision thresholds as a quantitative measure to reduce the number of false positives in AF mode. After the application of this criterion, only 12 of the 33 biopsies remained positive, thus 21 of the 24 false positives could be rejected. Together with the WL positives, a number of 20 biopsies taken on 15 patients remained for further analysis. The performance of AF and WL endoscopy on the remaining 20 positive sites are summarized in Table 2.

Table 2

Note that the hyperplasia presented in Figure 7 gives a normalized green/red ratio of 0.74±0.04 and is consequently rejected with this method.

The PPV gives the probability that a lesion is actually a lesion if visible with a given method (AF or WL), as opposed to the sensitivity that gives the capability of a method to find a lesion if there actually is a lesion. The PPV value is given as the ratio of the true positive (TP) results on the overall positive results [true and false positive (FP)].

The three false positive results remaining in AF mode after the threshold approach were not positive in the WL mode. This means that if we combine the AF thresholding mode with the WL mode, the resulting PPV is 100.

Note that four lesions were detected by AF only, whereas no lesions were detected by WL only. This means that the AF sensitivity (SENS _AF ) is higher than the WL sensitivity (SENS _WL ). Although it is not possible to calculate a value for SENS _AF (as the false-negative rate is not measurable in this organ),³ one can nevertheless estimate this parameter from SENS _WL by the expression SENS _AF =SENS _WL (TP _AF /TP _WL ), where TP _AF and TP _WL are the true-positive results under AF and WL, respectively. This leads to a SENS _AF 1.8 times higher than SENS _WL (Table 2).