Endoscopic fluorescence lifetime imaging for in vivo intraoperative diagnosis of oral carcinoma.

FLIM Endoscope Probe and Instrumentation

FLIM techniques can be categorized into time-domain or frequency-domain, and wide-field imaging or scanning, according to the types of light sources and detectors used. Wide-field imaging time-domain FLIM is currently the most widely used modality for in vivo or clinical application because of its relative robustness and fast imaging speed. In this work, we built a portable time-domain wide-field FLIM apparatus that was coupled with a fiber image guide endoscopic probe. The apparatus schematic is shown in Figure 1 . The gated optically intensified CCD camera (ICCD, 4 Picos, Stanford Computer Optics, Berkeley, CA, USA) had a minimum gating time of 200 ps and a repetition rate up to 200 kHz. Tissue autofluorescence was induced by a fiber-delivered 337 nm pulsed nitrogen laser (MNL 205, LTB Lasertechnik, Berlin, Germany) with 700 ps pulse width and ≤50 Hz repetition rate. A fully integrated semiflexible fiber optic endoscope probe was built for fluorescence lifetime imaging application in vivo , as shown in Figure 1b . This included an optical delivery fiber (600 μ m, NA 0.48, Thorlabs, Newton, NJ, USA) and a 1.7-m-long, 0.6-mm-diameter 10,000 fiber image guide with a gradient index lens termination to image the fluorescence emission (4-mm working distance, 4-mm-diameter field of view). The total length was 3 m, and it consisted of three parts: a 2.4-mm-diameter, 16-mm-long hollow stainless steel rigid tip; a 1-m-long common part integrating the fiber image guide and optical delivery fiber; and a bifurcation into two legs. The light delivery fiber was optically aligned and cemented solidly to the fiber image guide within the stainless steel tube, which was sufficiently robust to allow operators to hold and point the endoscope to specific tissue regions of interest as indicated in Figure 1c . The common part of the probe had an outer diameter of 5 mm and a minimum bending radius of 10 cm, similar to that of the fiber image guide, and was contained within a coiled stainless steel furcation tube. After bifurcation, the optical fiber and fiber image guide were protected in a tube including a PVC jacket (3.8-mm outer diameter) and an inner tube (1-mm inner diameter), and were terminated with SMA connectors. The laser delivery fiber was longer than the fiber image guide to allow a higher freedom of movement during the in vivo work. A 20× microscope objective and 150-mm focal length doublet lens were used to magnify the proximal facet of the fiber image guide onto the ICCD chip. A filter wheel was inserted into the optical path to select up to six emission filters. Only one bandpass filter was used throughout this study: 460 ± 25 nm (central wavelength ± bandwidth). Signal synchronization was optimized for both the ICCD and the pulsed laser using the CCD camera as the master trigger at 30 Hz. This triggered the nitrogen laser, which in turn triggered the short intensifier gate. The precise delay between the gating time and the fluorescence signal was controlled by an internal delay generator that had a resolution of 10 ps. The electronic trigger of the laser had a jitter of less than 200 ps. The temporal gate width of the 4 Picos was varied from 0.2 to 1 ns depending on the signal level found for each experiment. The frame rate of the CCD camera was 30 Hz at the resolution of 480 × 736 pixels. For most tissue samples, one intensity gated image required the integration of 128 laser excitation pulses in order to obtain sufficient signal to noise. Data acquisition times for each measurement were ~2 min including one steady-state image and a series of up to 29 time-gated images (0.5 ns gate time, 0.5 ns relative delay time step). The camera “4 SPEC” software controlling the gated optical intensifier was customized for FLIM image acquisition. The overall system performance test followed procedures similar to those previously reported by our group ( Elson et al., 2007 ; Sun et al., 2009 ). To bring FLIM into the operating room for tumor demarcation in vivo on human subjects, the system was designed to meet specific clinical requirements, and the FLIM probe was integrated and mounted on a mobile cart. The process for acquiring measurements of in vivo tissue autofluorescence is demonstrated in Figure 1c , showing the endoscope held in the oral cavity by a surgeon.

Clinical Validation in Human Subjects

Ten patients with suspected HNSCC were included for FLIM measurement and a total of 26 sites were examined. The study was approved by the Institutional Review Board at the University of California at Davis and all patients involved in this research were consented for the study. The FLIM instrument was placed on a mobile cart so that it could be easily transferred between operating rooms before the experiment. Before data collection from a patient, the rigid distal end of the endoscopic probe was placed in a PMMA protective tube that could be sterilized and sealed with a sapphire window. This tube extended beyond the end of the probe and acted as a spacer between the probe and the tissue in order to maintain the 4-mm working distance of the probe. For intraoperative measurement, the protective sterile tube was gently positioned perpendicular to the interrogated tissue surface. After the measurement was completed, tissue biopsies were taken from the measured region and histopathological analysis was independently conducted by a clinical pathologist. The energy density delivered at the tissue surface was 0.16 mJ/cm 2 per pulse. This was 20 times lower than the skin maximum permissible exposure value of 3.2 mJ/cm 2 for UV lasers according to the American National Standard for Safe Use of Lasers.

Image and Data Processing

Image processing and lifetime deconvolution were conducted using a custom-built graphical user interface written using MATLAB. A rapid deconvolution was achieved through the polynomial Laguerre expansion, which allowed the fluorescence impulse response function, the fluorescence lifetime, the integrated intensity, and the Laguerre coefficients (LECs) to be calculated ( Jo et al., 2006 ). The Laguerre technique is based on a nonparametric model that allows the evaluation of fluorescence decays from complex fluorescent systems such as biological tissues without a priori assumptions about the decay function or number of fluorescent molecular species within the fluorescent system ( Liu et al., 2012 ). In addition, the Laguerre functions contain a built-in exponential term and are orthogonal, which results in a fast, convenient, unique, and complete expansion of the exponential decays, and in addition produces an additional set of decay parameters (i.e., LECs) for enhanced analysis of the fluorescence decay features. In this work, up to four Laguerre polynomials were used in the expansion to the fluorescence impulse response profile for each measurement based on linear least-square error, resulting in a set of four LECs. The parameter α found within the Laguerre polynomials was fixed at 0.8 based on the Kernel memory length and the number of Laguerre functions, allowing for the functions to decay sufficiently close to zero by the end of the fluorescence decay ( Jo et al., 2006 ). The resulting function could then be used to calculate the integrated intensity and the average fluorescence lifetime by computing the interpolated time at which the intensity falls to 1/ e of the initial intensity. The tissue FLIM images (480 × 736 pixels) presented in this paper took less than 60 s to process using an algorithm that was implemented on a PC with an Intel Core 2 CPU 6600 at 2.40 GHz and 1 GB RAM.

Statistical Analysis

Statistical analysis was conducted for all 26 sites to investigate the differentiation of HNSCC and the surrounding normal tissue for different parameters including the average fluorescence intensity, lifetime, and LECs values found from each image field. The main results are summarized in Table 1 , including the mean values and standard errors for each parameter for normal and tumor samples (13 for normal and 13 for tumor). At the bottom of the table, the p -value between the two groups is listed to evaluate the significance. The overall average lifetime in tumor tissue decreased by 18.9% from the normal tissue value τ normal = 1.49 ±0.06 ns to τ tumor = 1.21 ± 0.04 ns. The statistical difference was significant based on the p -value of 0.0009. In addition, the average intensity decreased clearly in the tumor tissue, showing a 47.8% drop compared with normal tissue, but the p -value was slightly higher at 0.012. One-way analysis of variance (ANOVA) was applied to the statistical analysis of fluorescence lifetimes and intensities. Figures 4a and 4b are ANOVA box plots showing the difference of means, standard deviations, and distribution of lifetimes and intensities between normal and tumor, respectively. The fluorescence lifetimes of tumor and normal groups demonstrated an obvious separation with a low p -value