Instrumentation

The FLIM apparatus (Fig. 1 ) consisted of a gated intensified charge-coupled device (ICCD) camera, a pulsed laser, a flexible fiber-image guide-based endoscope, and a filter wheel. This system was adapted from our previously reported FLIM apparatus.⁹ Modifications were made for intraoperative use, including portability, remote fiber optic access to patients, fiber-probe sterilization, and medical safety. Figure 1 is a schematic of the instrument. The positioning of the fiber probe on the brain is also depicted. Briefly, tissue autofluorescence was induced by a pulsed nitrogen laser ( $337\mathrm{nm}$ , $700\mathrm{ps}$ , MNL 205 nitrogen, LTB Lasertechnik, Berlin). A customized semiflexible endoscope probe ( $3\mathrm{m}$ long) remotely delivered the excitation laser, and the fluorescence emission was imaged using a gradient index (GRIN) lens ( $\mathrm{NA}=0.5$ , $0.5\mathrm{mm}$ diam, and $4\text{-}\mathrm{mm}$ field of view) cemented to a fiber image guide ( $0.6\mathrm{mm}$ diameter, 10,000 fibers). The fluorescence emitted from the proximal end of the fiber bundle was projected onto the fast-gated ICCD (4 Picos, Stanford Computer Optics, Berkeley, California). A bandpass filter with a center wavelength of $460\mathrm{nm}$ and a bandwidth of $50\mathrm{nm}$ was used. Data acquisition time for each measurement was $\sim 2\min$ , including one steady-state image and a series of up to 29 time-gated images ( $0.5\text{-}\mathrm{ns}$ gating time and $0.5\text{-}\mathrm{ns}$ relative delay-time interval). During imaging, the probe was gently positioned perpendicular to the tissue surface and held with a Greenberg device to minimize the moving artifacts.

Fig. 1

(a) Schematic of the FLIM instrument setup, including a picture of the tip of the imaging bundle probe positioned on the interrogated area of the cortex. (b) Fluorescence intensity image of rhodamine B and (c) a cross section of the intensity distribution (profile). Note that The fluorescence intensity is higher in the center of the image when compared to the edge of the image. (d) Fluorescence average lifetime image and (e) fluorescence average lifetime histogram of the rhodamine B solution measured and analyzed with the FLIM system. The pixel coordinates are shown on the left and bottom side of the images in (b) and (d).

Sample illumination through endoscope probe

The energy density delivered at the tissue surface was $0.16\mathrm{mJ}∕{\mathrm{cm}}^2$ per pulse (20 times lower than the maximum permissible exposure value of $3.2\mathrm{mJ}∕{\mathrm{cm}}^2$ for UV lasers according to the American National Standard for Safe Use of Lasers). Figures 1 and 1 demonstrate the illumination intensity at a plane located $4\mathrm{mm}$ in front of the endoscope probe for a solution of rhodamine B ( $0.1\mathrm{mM}$ in methanol). The fluorescence intensity is strong in the center but drops significantly at the edge. This is due to the current endoscope that did not permit uniform illumination of the sample. Vignetting also occurs at the edge of the field. Despite this, the mean average fluorescence lifetime $(0.44\pm 0.03\mathrm{ns})$ is uniform across the entire surface [Figs. 1 and 1] after numerical deconvolution of lifetime values, as described next.

Image processing

Images were analyzed using the Laguerre polynomial deconvolution technique to calculate the fluorescence impulse response function, average fluorescence lifetime $\left({\tau }_f\right)$ , integrated intensity, and Laguerre coefficients (LECs).¹⁰ The Laguerre functions contain a built-in exponential term that results in a convenient expansion of exponential decays while also forming a complete orthogonal set that allows fast and complete expansion. The first four Laguerre functions were sufficient to recreate the fluorescence decay. The resulting function can then be used to calculate ${\tau }_f$ (by computing the interpolated time at which the intensity falls to 1/e of the initial intensity) and integrated intensity of the data. This technique¹⁰ enables fast lifetime processing. Tissue FLIM images $(480\times 736\text{pixels})$ presented took less than $60\mathrm{s}$ to process on a PC with an Intel Core 2 CPU 6600 at $2.40\mathrm{GHz}$ and $1\text{-}\mathrm{GB}$ RAM running Matlab.

Statistical analysis

To distinguish between tissue types (tumor versus normal), FLIM data represented by multiple parameters (fluorescence intensity, ${\tau }_f$ , and LECs values) were evaluated using one-way analysis of variances (ANOVA). A $p$ -value $<0.05$ was used as criteria for achieving statistical significance.

Validation on human subjects

FLIM experiments were conducted on three patients undergoing craniotomy and resection of glioblastoma. 13 sites were examined, four from normal cortex (NC), seven from GBM-infiltrated cortex (GBM), and two from brain-tumor interface (BTI). Areas were identified by the neurosurgeon based on an intraoperative MRI-guided neuronavigation system and by visual inspection of the operative site. The study was approved by the University of California Davis Institutional Review Board. The FLIM instrument (on a mobile cart) was brought to the operating room and the distal end of the endoscopic probe was placed in a sterilized protective tube that extended $4\mathrm{mm}$ beyond the end of the probe. For intraoperative measurement, the protective sterile tube was positioned perpendicular to the tissue surface, gently resting on the brain. Tissue biopsies were taken from tumor areas for histopathologic correlation.

Fluorescence intensity image

In NC, a clear network of blood vessels was observed in the intensity image [Fig. 2] most likely caused by the strong absorption of hemoglobin.¹¹ The center region of the intensity image was bright due to the uneven illumination of the endoscopic probe. Similar features were observed for intensity images of GBM [Fig. 2] and BTI [Fig. 2] sites. Overall, the fluorescence intensity was higher for areas identified as NC compared to GBM (Table 1). These findings are in agreement with earlier LIFS studies,^{4, 5, 11, 12} which demonstrated that fluorescence intensity in GBM was lower compared with NC between 450 and $480\mathrm{nm}$ . It is well documented^{4, 5} that upon UV excitation, the main brain tissue fluorophores emitting at these wavelengths are the reduced nicotinamide adenine dinucleotides NADH and reduced nicotinamide adenine phosphate dinucleotides NADPH in both free and protein-bound form. The concentration of the NADH is approximately five times greater than that of the NADPH.¹³ Changes in autofluorescence in this spectral band were attributed to an alteration of the contribution of these dinucleotides to the overall emission, and to changes of the total amount of redox equilibrium and the free/bound condition of these coenzymes.^{4, 13}

Fluorescence lifetime image

NC showed relatively uniform ${\tau }_f$ values across the entire area [Fig. 2], as demonstrated by a narrow ${\tau }_f$ histogram [Fig. 2] with a value of ${\tau }_{\mathrm{NC}\text{-}\mathrm{Imean}}=1.23\pm 0.09\mathrm{ns}$ . Neither the nonuniform illumination at the tissue surface nor the presence of blood affected the ${\tau }_f$ values. This emphasizes the more robust nature of time-resolved measurement versus intensity measurements when implemented in vivo, where uniform illumination can be difficult to achieve due to tissue irregularities as well as the presence of blood in the surgical field. For a few measurements of NC, we noticed a slight increase in ${\tau }_f$ to $\sim 1.5\mathrm{ns}$ in areas containing blood vessels. While the exact origin of this trend is not known, this trend is clearly related to the presence of the blood vessel, as observed in the fluorescence intensity image and by the eye. Possible explanations include the contribution to the fluorescence from structural protein in the blood vessel, or the decrease of oxygen concentration in these regions, which in turn increases the relative concentration of bound NADH, or the selective absorption of the fluorescence by the blood.

GBM sites exhibited nonuniform distribution of ${\tau }_f$ values across the interrogated area. The lifetime histogram [Fig. 2] depicts a higher ${\tau }_f$ with a broad distribution $({\tau }_{\mathrm{GBM}\text{-}\mathrm{Imean}}=1.64\pm 0.33\mathrm{ns})$ compared with NC. A few subareas exhibited shorter ${\tau }_f$ values comparable with that of normal cortex. We ascribe this to the variable depth at which the tumor is infiltrated within the cortex with respect to the surface of the cortex. While the entire area is identified as tumor, in some subareas the tumor can be below the penetration depth $(\sim 200\text{to}300\mu \mathrm{m})$ of the excitation wavelength, thus fluorescence was mainly collected from normal tissue on top of the tumor. The BTI sites also exhibited a nonuniform distribution of the fluorescence lifetime values. However, the ${\tau }_f$ values were segregated, with longer values $({\tau }_{\mathrm{Imean}}=1.66\pm 0.20\mathrm{ns})$ in tumor versus normal cortex $({\tau }_{\mathrm{Imean}}=1.40\pm 0.09\mathrm{ns})$ [Fig. 2]. Parameters derived from FLIM data (both ${\tau }_f$ and LECs) demonstrated significant differences between GBM and NC. Overall, we determine faster decay dynamics for NC compared to GBM. We note also that LEC-0, which is associated with the fast decay component of the fluorescence decay curve,¹⁰ presents the largest relative change $(>80\mathrm{\%})$ with a lower value in NC versus GBM.

The ${\tau }_f$ of NADH and NADPH depends on their bound-free forms, with a range from subnanosecond values in the free form to a few nanoseconds in the protein-bound form.^{14, 15} The ${\tau }_f$ of brain tissues in this study ranged between $1.2\text{to}2.6\mathrm{ns}$ , most likely reflecting changes in the bound-free NADH/NADPH equilibrium between normal and tumor conditions. These values suggest that fluorescence emission of brain tissue is dominated by the protein-bound form of these coenzymes, with a relatively higher contribution of the bound form to the GBM fluorescence when compared to that of NC. The fluorescence lifetime of NADH and NADPH in brain tissues in vivo has been underinvestigated. Vishwasrao ¹³ reported NADH kinetics in brain hippocampus slices in normoxic and hypoxic conditions, where the latter showed a shorter ${\tau }_f$ than the former. In our previous TR-LIFS studies,⁶ we also observed similar or slightly longer ${\tau }_f$ at $460\mathrm{nm}$ for GBM compared with normal cortex, a trend that is in agreement with the current study. Due to different experimental conditions, no extensive comparison with other studies can be made, however our current findings demonstrate that imaging lifetimes allow normal versus tumor to be distinguished

The main challenge in the validation of FLIM results is the ability to conduct conventional histopathological analysis of all interrogated areas, so that optical results can be correlated to local pathologies. This is a significant challenge for brain tissue diagnosis, where biopsy is limited to areas of tumor involvement. Neurosurgical techniques for GBM resection do not typically involve obtaining pathologically tumor-free tissue margins, as is the case for cancers operated on in other regions of the body. Instead, GBM resection is limited to the areas of gross tumor involvement, or in the case of malignant GBMs, to the enhancing and necrotic portions. Thus, in the current study the validation of the FLIM results was mainly based on gross pathology, MRI data, and neurosurgeon experience. Another limitation is the sampling error that results from the heterogeneous tissue composition of the tumor, which can contain heterotopic areas of hypervascularity, necrosis, ischemia, and degrees of malignant degeneration.