2.1.

Samples

The study was conducted on tissue specimens from patients undergoing brain tumor surgery and histopathologically diagnosed as meningioma (total of 22 patients with various types of meningiomas: 8 transitional, 11 meningothelial, 1 psammomatous, 2 microcystic) and adjacent normal dura (4 patients). Samples of normal gray matter were obtained from 4 patients undergoing surgery for en-bloc resection of glioma brain tumor underlying cortex. The study was carried out with approval of the Cedars-Sinai Institutional Review Board. After excision, the specimens were kept in a sealed container and transported from the operating room to the spectroscopy laboratory. The spectroscopic investigations were performed within 30 to $120\min$ after excision. During the experimental procedure the samples were kept in a glass petri dish and soaked using 1 to 2 drops of saline to avoid drying of the tissue. Multiple distinct sites (total: 97) on the tissue were selected for spectroscopic investigation. After the completion of the spectroscopic procedure, the selected sites were marked using a dye and fixed in 10% buffered formalin, and then sent for histopathological studies. From each sample, transversely oriented sections $(4\mathrm{mm}$ thick) were then cut from the marked areas. The tissues were embedded in paraffin and stained with hematoxylin-eosin (H&E). The histological sections were evaluated by light microscopy.

2.2.

Instrumentation

The tissue samples were spectroscopically investigated with a prototype time-domain TR-LIFS apparatus (Fig. 1 ) similar to that used in previous studies.^{8, 9} A pulsed nitrogen laser (EG&G, model 2100; $337\mathrm{nm}$ , $1.2\mathrm{ns}$ , $10\mathrm{Hz}$ ) was used as excitation light source. The output of the light source was delivered to the tissue by a custom-made bifurcated probe, which consists of a central excitation fiber ( $600\mathrm{\mu }\mathrm{m}$ , numerical aperture 0.11) surrounded by a collection ring of 14 fibers ( $200\mathrm{\mu }\mathrm{m}$ , numerical aperture 0.22). The collection fibers are beveled at a $10°$ angle in order to improve light excitation-collection overlap for small tissue to probe distances.¹⁶ The pulse width measured at the tip of the fiber was $2.0\mathrm{ns}$ full width at half maximum. The collected fluorescence emission was focused into a scanning monochromator (ORIEL, model 77200) using an $F\#$ matcher (ORIEL, model 77529), and detected by a gated multichannel plate photomultiplier tube (MCP-PMT; Hamamatsu, model R2024U, rise time: $0.3\mathrm{ns}$ ) placed at the monochromator exit slit. The photomultiplier tube was gated using a high voltage gate (EG&G, ORTEC; $1\mathrm{GHz}$ ). The photomultiplier output and the entire fluorescent pulse from a single excitation laser pulse was directed to a digital oscilloscope (Tektronics, model TDS 680C, sampling frequency: 5 Gsamples/s, bandwidth: $1\mathrm{GHz}$ ) through a fast 1-GHz preamplifier (EG&G, ORTEC model 9306). The gating of the PMT was achieved using an optical switch triggered by laser pulse. To eliminate the scattering of the reflected excitation laser by the sample, a 360-nm long-pass filter was placed at the entrance slit of the monochromator and a 345-nm long-pass filter was placed in front of the PMT. A personal computer was used to control data acquisition, data transfer from oscilloscope, and monochromator wavelength scanning.

Fig. 1

Schematic of the instrumental setup for time-resolved fluorescence spectroscopy. (GPIB: General Purpose Interface Bus).

2.3.

Experimental Procedure

The fiber optic probe was placed $3\mathrm{mm}$ above the tissue specimen to optimize the probe light collection efficiency, as previously reported.¹⁶ Time-resolved emission of each sample was recorded in the 370 to 500 nm spectral range and scanned at 5 nm intervals. Five consecutive measurements of the fluorescence pulse emission at $390$ and $460\mathrm{nm}$ are performed to assess the reproducibility of the fluorescence lifetime measurement. The energy output of the laser (at the tip of the fiber) for sample excitation was adjusted to $3.0\mathrm{\mu }\mathrm{J}∕\mathrm{pulse}$ .

2.4.

Data Analysis

3.1.

Histology

Tumors were classified using the World Health Organization (WHO) criteria.¹⁹ Of the total of 97 sites that underwent spectroscopic analysis, 75 were histologically classified as a low-grade meningioma (WHO grade I), 17 as normal dura, and 5 as normal cortex. Most meningiomas, regardless of subtype, demonstrate some fibrous tissue component whether fibrosis, entrapped dura, or fibroblast-like meningioma tumor cells. The extent of extracellular fibrosis or sclerosis ranged from approximately 5% of visualized surface area to approximately 80%. Most meningiomas had a few psammoma bodies focally. A few cases had no psammoma bodies. Two cases demonstrated numerous psammoma bodies. Normal dura samples demonstrate dense fibrous tissue without evidence of tumor. The brain tissue demonstrates layered neurons indicative of cerebral cortex also without evidence of tumor.

3.2.

Time-Integrated and Time-Resolved Fluorescence

3.2.1.

Meningioma

The time-resolved fluorescence emission of meningioma samples showed a relatively narrow broadband emission characterized by a well-defined peak at a 385- to 390-nm wavelength [Figs. 2(a) and 3 ] lasting for more than $15\mathrm{ns}$ . The fluorescence intensity decay [Figs. 4(a) and 4(b)] was found wavelength-dependent with longer time-decay values in the region of peak fluorescence ( ${\tau }_{390}=2.04\pm 0.72\mathrm{ns}$ ; ${\tau }_{390:10\%}=8.0\pm 1.84\mathrm{ns}$ ) when compared with the red-shifted wavelengths ( ${\tau }_{460}=1.2\pm 0.46\mathrm{ns}$ ; ${\tau }_{460:10\%}=5.1\pm 1.8\mathrm{ns}$ ).

Fig. 4

Time-dependent parameters of meningioma, dura mater, and normal cerebral cortex: (a) average lifetime ${\tau }_{\lambda }$ and (b) time-decay to 10% ${\tau }_{\lambda 10\%}$ . Results are presented as $\mathrm{mean}\pm \mathrm{standard}$ error of the data from each independent measurement.

Fig. 2

Representative time-resolved fluorescence emission of (a) meningioma and (b) dura mater.

Fig. 3

Normalized integrated fluorescence emission spectra of meningioma, dura mater, and normal cerebral cortex. Results are presented as $\mathrm{mean}\pm \mathrm{standard}$ error of the data from each independent measurement.

3.2.5.

Discriminant analysis and classification

The classification accuracy (sensitivity and specificity values) for the discrimination of the meningioma from normal dura and cortex is summarized in Table 1 . When only spectral parameters were used as input, the stepwise linear discriminant algorithm selected five parameters ( $I_{480}∕I_{370}$ , $I_{440}∕I_{370}$ , $I_{460}∕I_{370}$ , $I_{380}∕I_{370}$ , $I_{390}∕I_{460}$ ). When the classification was performed using only the temporal parameters as input, the discriminant algorithm selected three parameters ( ${\tau }_{370:10\%}$ , ${\tau }_{500:10\%}$ , ${\tau }_{440}$ ). When both spectral and temporal parameters were used for the analysis, eight significant parameters were selected, $I_{390}∕I_{460}$ , $I_{380}∕I_{370}$ , $I_{390}∕I_{370}$ , and $I_{480}∕I_{370}$ from the spectral domain and ${\tau }_{370:10\%}$ , ${\tau }_{370}$ , ${\tau }_{440:10\%}$ , and ${\tau }_{500:10\%}$ values from the temporal domain. Figure 5 depicts a 2-D scatterplot of two discriminant functions constructed using variables derived from both spectral- and time-resolved domains. It can be observed that function 1 discriminates cortex from dura mater and meningioma, while function 2 enabled discrimination of meningioma from dura mater. In summary, parameters derived from either only spectral features or only temporal features provided means of discrimination between the three tissue groups. However, the overall accuracy of classification improved (91.8% cross-validated set; 92.8% training set) when both spectral and time-resolved features were used for discrimination. In contrast, the overall classification accuracy was lower when either only spectral features (63.9% cross-validated set; 67% training set) or only time-resolved features (83.5% cross-validated set; 85.6% training set) were used as input predictor variables.

Fig. 5

Results of the linear discriminant function analysis. Scatterplot depicts the discriminant score for the three groups of tissue (meningioma, dura mater, and cerebral cortex).

Table 1

Classification accuracy (sensitivity and specificity values).

		Meningioma N=75	Normal Dura N=17	Normal Cortex N=5
Spectral values only	Sensitivity	61%	65%	100%
	Specificity	73%	66%	98%
Time-resolved values only	Sensitivity	84%	77%	100%
	Specificity	100%	91%	90%
Spectral $+$ time-resolved values	Sensitivity	89%	100%	100%
	Specificity	100%	91%	99%

Total number of sites: 97; Meningioma

4.1.

Time-Resolved Fluorescence Spectroscopy in Tissue Characterization

In this study, we employed a time-resolved fluorescence spectroscopy apparatus, enabling the acquisition of fluorescence temporal response across the emission spectrum of tissue specimens, thus both steady state spectra and time-resolved intensity decay can be recovered from a single measurement sequence. We found that combining spectroscopic parameters derived from both spectral and time domains, not only facilitates a better understanding of the kinetics of the biochemical and physiological processes occurring in meningiomas, dura mater, and cortex but also contributed to the improved accuracy of tissue recognition. Although numerous research groups^{4, 7, 12, 14} have shown that steady state fluorescence spectroscopy techniques alone are potential tools for diagnosis of tumors, including brain tumors, detection of cancer using only spectral information is subject to several limitations. For example, the relatively broad emission brands of tissue fluorescent components may reduce the capability of this technique to resolve the spectrally overlapping components. Also, the presence of endogenous chromophores (hemoglobin) or changes of excitation/collection geometry can strongly influence the acquired spectral profile. The time-resolved measurement resolves fluorescence intensity decay in terms of lifetimes and thus provides additional information about the underlying fluorescence dynamics. This property has intrinsic advantages, including (1) biomolecules with overlapping fluorescence emission spectra but with different fluorescence decay times can be discriminated; (2) the measurements are sensitive to various parameters of biological microenvironments (including pH and enzymatic activity); and (3) the measurement is independent of fluorescence emission intensity as long as the SNR is commensurable, and consequently independent of the presence of the endogenous chromophores in tissue (hemoglobin) and excitation-collection geometry or optical assembly.⁸ These are important features for further in vivo studies and intraoperative diagnosis of meningiomas as well as other types of neoplasm.

4.2.

Fluorescence Characteristics and Tissue Recognition

Two wavelength ranges appeared important for developing diagnostic algorithm for distinguishing meningioma from normal dura and cortex: $370$ to $400\mathrm{nm}$ (the region of main peak emission for meningioma and dura) and $440$ to $480\mathrm{nm}$ (the region of main peak emission for cortex). Note that the dominant narrowband blue-shifted peak emission is a hallmark for meningioma, but it is not always observed in healthy or diseased neuronal tissue including gliomas.^{6, 7, 13, 14} The application of multivariate statistical analysis has shown that meningioma, normal dura, and cortex tissues can be differentiated using a relatively limited number of predictor variables (16 in total) from these two spectral ranges. Parameters obtained from both spectral (intensity values) and time-resolved emission contributed to the accuracy of tissue classification. The use of both sets of spectroscopic parameters, spectral and time-resolved, have improved the overall accuracy of meningioma detection (91.8%), compared to the case when only spectral variables were employed (63.9%). Note that the temporal features alone used in discriminant function analysis demonstrate superior ability to classify the tissue than spectral feature alone. Although in this study we used only a small dataset, these results indicate that the time-resolved fluorescence spectroscopy technique is robust enough to enable good discrimination between the three types of tissue investigated in this study. Moreover, the use of time-resolved information enhances the ability to discriminate tumor tissue from the surrounding normal tissue. The classification accuracy may be further improved once the number of samples for dura and normal cortex in the training set increases.

4.3.

Fluorescence of Meningioma, Dura, and Cortex: Interpretation in Terms of Intrinsic Fluorescent Constituents

The fluorescence intensity time-decay characteristics were found both tissue- and wavelength-dependent. The fluorescence lifetimes ${\tau }_{\lambda }$ of meningioma and dura were longer when compared to cortex and decreased significantly with the emission wavelength. These trends suggest that at least two distinct fluorophores are likely to contribute to the fluorescence emission of meningioma and dura. In contrast, the fluorescence emission of normal cortex, characterized by a short-lived emission with nearly constant lifetime values across the emission spectrum, most likely originated from a single fluorophore. Potential sources of fluorescence are outlined in the following.

The fluorescence of connective tissue proteins, in particular collagens, is most likely to dominate the fluorescence emission of dura and meningioma. Several types of collagen and their cross-links are known to absorb light above $300\mathrm{nm}$ and to exhibit a strong fluorescence emission in 370-nm to 500-nm spectral range.^{5, 8, 20} This assumption is supported by our histopathological analysis, which identified fibrous tissue in both meningioma and dura samples. It is also in agreement with the biochemical and chromatographical analysis of the composition of meningioma and dura reported in other studies that demonstrated the presence of high concentration of various types of collagen in these tissues.²¹ Our results also showed that meningioma and dura were characterized by distinct fluorescence decay dynamic [e.g., dura ${\tau }_{\lambda :10\%}<\mathrm{meningioma}{\tau }_{\lambda :10\%}$ , Fig. 4(b)], albeit having common spectroscopic features such as an intense fluorescence emission in the 370- to 400-nm wavelengths range (peak at $\sim 390\mathrm{nm}$ ) with relatively long-lived fluorescence $({\tau }_{\lambda }>1.5\mathrm{ns})$ . These differences suggest that fluorescence of meningioma and dura have originated from distinct types of collagens and their cross-links.

For example, studies of fluorophores from collagen-rich fractions of human dura mater using chromatography have demonstrated that two fluorophores are associated with dura fluorescence emission.²² These are a “P” fluorophore (a pyridinium compound similar to pyridinolines collagen cross-link, peaks excitation/emission: $335∕385\mathrm{nm}$ ) and an “M” fluorophore (an age-related fluorophore due to acceleration of collagen browning; peaks excitation/emission: $370∕440\mathrm{nm}$ ). Our TR-LIFS measurement on dura mater tissue corroborates with this early study by revealing two peaks emission, one at about 390 and the other at $440\mathrm{nm}$ . In addition, the change of the fluorescence decay dynamics as a function of wavelength (decrease of ${\tau }_{\lambda }$ with wavelength, and a near constant ${\tau }_{\lambda :10\%}$ across the spectrum), as observed in our study, not only confirm the presence of at least two fluorophores in dura matter, but also indicates that these fluorophores are characterized by distinct lifetimes. Our results also suggest that the valley that modulates the spectral emission may not be entirely due to blood presence in tissue or fluorescence reabsorption, as initially assumed.

On the other hand, several immunocytochemical studies have demonstrated the presence in meningioma of collagen I, collagen III, collagen IV, procollagens, laminins, and vimentine.²¹ Evidence has also shown that meningioma cells are derived closely from the leptomeningeal arachnoid cells and are “fibrous response” of the leptomeninges to trauma, infection or other pathologies such as tumor infiltration.^{21, 23, 24} The time-resolved spectra of meningiomas in our study, characterized by a relatively narrow-band blue-shifted emission [Fig. 2(a)], closely resemble collagen type I (peak emission $380\mathrm{nm}$ ) and collagen type III (peak $390\mathrm{nm}$ ) fluorescence.^{5, 8, 20} Such trends indicate that these two types of collagen most likely dominated the fluorescence emission of meningioma at the blue-shifted wavelengths.

The short-lived $(<1.5\mathrm{ns})$ fluorescence emission (peak at about $460\mathrm{nm}$ ) found for cortex corresponds to the NAD(P)H fluorescence emission fluorescence. These features are in agreement with previous studies that have reported the spectral^{6, 7, 13, 14} as well the lifetime⁹ fluorescence emission from cortex. Both forms of NAD(P)H, free in the cytoplasm [low quantum yield of about 2% and lifetimes $<1\mathrm{ns}$ (Refs. 25, 26, 27)] as well as bound in mitochondria [high quantum yield about fourfold higher than the free form and radiative $\mathrm{lifetimes}>1\mathrm{ns}$ (Refs. 25, 26, 27)] are widely recognized as endogenous fluorophore involved in cellular metabolism,^{25, 27, 28} and probably have originated the fluorescence emission measured from cortex samples. Also, in the red-shifted spectral range, NAD(P)H fluorescence likely contributed to the fluorescence emission of dura mater and meningioma, as depicted by the short lifetimes values determined for these tissues at emission wavelengths above $440\mathrm{nm}$ .

Further investigations are required for a more complete understanding of the factors of intrinsic fluorophores causing differences in the fluorescence emission properties between meningioma dura, and cortex. Moreover, the metabolic properties of tissue may be different ex vivo versus in vivo, and these differences can be reflected in tissue fluorescence response.¹³ In vivo studies will account for such differences and enable direct comparison with the current ex vivo studies and better our understanding of how metabolic changes affect the fluorescence data. This study, however, indicated that time-resolved fluorescence spectroscopy represents a potential pathway for nonintrusively gaining insights into the biochemical composition of these tissues.

4.4.

Time-Resolved Fluorescence Spectroscopy Technology: Clinical Perspective

Neurosurgeons often request frozen section intraoperative tissue specimen analysis from one to many sites during a single surgery for tumor resection. This is generally the case with surgical resection of meningiomas where one to several dural margins are often sent to the neuropathologist for intraoperative diagnosis beyond the initial tumor mass specimen. Rapid identification of neoplastic versus normal tissue using fluorescence-based diagnostics would be a highly beneficial tool for both the surgical pathologist as well as the neurosurgeon.

Moreover, meningiomas are believed to arise from the “arachnoid cap” cells of the arachnoid, although they are intimately associated with the dura mater. Because of their histogenesis, meningiomas develop along any of the external surfaces of the brain as well as within the ventricular system. Common sites of tumor include the convexity, the falx, and the skull base of the anterior, middle, or posterior fossae. Meningiomas occurring “en plaque”, or located in anatomically challenging areas such as the petroclival region, present difficulties in achieving complete surgical resections. Thus, a fluorescence spectroscopy may also prove to be valuable intraoperative adjunct for improving completeness of resection in a setting where margins of the neoplastic tissue are indistinct.

In this study, we used the autofluorescent properties of tissue as a diagnostic tool. For a clinical or intraoperative diagnostic of diseased tissue, such an approach has inherent advantages over the use of exogenous fluorophores (molecular probes). The measurement of endogenous fluorescence does not require any systemic or local administration of an external imaging agent, thus problems associated to toxicity and pharmacodynamics/kinetics of the external agent are not encountered. Because diagnosis based on autofluorescent properties of the tissue enables a direct and safe evaluation of the TR-LIFS technique in patients, we have recently developed a clinically compatible TR-LIFS apparatus.²⁹ Studies are currently being conducted at Cedars-Sinai Medical Center using this apparatus to intraoperatively record the fluorescence of brain tumors including meningioma. These studies will also aid in developing efficient clinical diagnostic procedures that include the detection of tumor margins. Studies have shown that the $1∕e$ optical penetration depth for 337-nm excitation wavelength ranges between 200 and $250\mathrm{\mu }\mathrm{m}$ depth in a variety of tissues including in brain tissue.¹³ Thus, using fiber optic probes with optimal excitation-collection geometry, the presence of tumor at the margins can be detected with relatively high spatial accuracy.

In addition, the analytical methods employed in this study (Laguerre expansion of kernels and linear discriminant analysis) facilitate near real-time data analysis. In a previous study, we demonstrated the ability of the Laguerre expansion of kernels technique for fast (less than $35\mathrm{ms}$ per wavelength) deconvolution of the fluorescence impulse function and analysis of biological systems.¹⁷ This method enables a direct recovery of the intrinsic properties of a dynamic system from the experimental input-output data.³⁰ The discriminant function analysis, also, has been explored for automated classification of ${}^1\mathrm{H}$ magnetic rotation spectroscopy (MRS) spectra from brain tumors.¹⁸ A fast real-time classification algorithm can be developed by assigning a discriminant score for different types of tissues. This discriminant score can then be compared with the acquired fluorescence from the unknown tissue so as to classify the tissue in real time. The accuracy of classification by linear discriminant analysis can be increased by continuously acquiring sample for training. Such a combination of analytical methods can aid the neurosurgeon with a real-time optical biopsy of tissue.

Furthermore, our current time-resolved fluorescence spectroscopic system has a potential to be integrated with the current stereotactic-image-guided surgery apparatus; thus, it can provide the surgeon with the diagnostic map of the tissue in view. This particular feature would not only enhance the surgeon's ability to ensure complete resection of various tumors, but may also help in future advances such as robotic surgery where the surgeon can access additional information about the tumor beyond just the visual image.