2.2.1.

Histologic measurements

Each sample was reviewed by the pool of clinical cytologists or histopathologists at the two cancer centers: the MDA and the BCCA. This review was followed with a blinded review by the study pathologists at each site; each site had one expert in cytology and one expert in cervical histopathology. A third review was conducted if there was disagreement between reviews 1 and 2. Approximately 20% of biopsies went on to third review, while for 80% there was agreement between the first and second reviews. The final consensus diagnosis was used as the gold standard for each spectroscopy measurement. We conducted multiple analyses of the interpathologist and intrapathologist readings as well as the interinstitutional and intrainstitutional readings. Clinical comparisons are often made using the kappa statistic for which there is a clinical scale. The kappa statistic demonstrates agreement above the level of chance. The kappa scale shows values: 0 agreement due to chance alone, 0 to 0.2 slight agreement, 0.2 to 0.4 fair agreement, 0.4 to 0.6 moderate agreement, 0.6 to 0.8 substantial agreement, and 0.8 to 1.0 almost perfect agreement. Kappas can be calculated using exact agreement (unweighted kappa) and agreement within a well-defined diagnosis above and below the exact diagnosis (weighted kappa). We also conducted studies of randomly selected slides, and had them reviewed at the other cancer center and by an outside expert pathologist at the Brigham and Women’s Hospital in Boston. The summary of these data shows substantial agreement between institutions (0.658). Each study pathologist, when compared to themselves, also had moderate to substantial agreement (0.652, 0.571, and 0.464 unweighted kappas; 0.718. 0.731, and 0.653 weighted kappas). Further detailed results of the histopathologic review and detailed assessment of agreement among the study pathologist readings can be found in Malpica ²⁴

2.3.1.

Illumination components

In both FastEEM systems, an arc lamp provides quasimonochromatic light for fluorescence measurements and broadband excitation light for reflectance measurements. Xenon arc lamps were used as the light source because a xenon light can produce wavelengths that range from 250 to $700\mathrm{nm}$ . Shutters control the exposure time of light that is directed toward the tissue. Reflectance spectra were measured after fluorescence spectroscopic measurements were acquired.

Fluorescence excitation light for the FastEEM2 originated from a 150-W xenon arc lamp where light output was focused into a step driven monochromator. A monochromator relies on a grating to disperse light spectrally, and a controlled stepper motor selects the wavelength in 10-nm increments from 300 to $480\mathrm{nm}$ for fluorescence measurements. In the FastEEM2, the lamp source housing has a separate port to direct white into the illumination fiber and bypasses the monochromator to take reflectance measurements. A shutter switches the port on or off according to which spectroscopic measurement is taken. The diffuse reflectance light passes through a rotating filter wheel containing three long-pass filters that are coated to transmit 50% of the broadband light intensity.

In the FastEEM3, a 75-W xenon arc lamp served as the light source where the output was directed through bandpass filters. Bandpass filters in the FastEEM3 are placed on a triple filter wheel assembly in which each wheel is individually rotated according to the excitation wavelength selected. Less power was needed from the FastEEM3 light source because bandpass filters were used to produce monochromatic light, which increases light intensity approximately 6 times relative to a monochromator. A cold mirror reflected $90\mathrm{deg}$ from the lamp output ensures that unwanted IR light does not pass through the filter wheel assembly. Excitation wavelength ranging from 290 to $530\mathrm{nm}$ in increments of $10\mathrm{nm}$ are delivered through excitation fibers in the fiber optic probe. The FastEEM3 light used for reflectance originates from the same lamp as the fluorescence excitation light. In the reflectance measurements, though, neutral density filters are also contained in the filter wheel assembly that only allow a certain percentage of light intensity to pass (10, 20, or 50%). The fluorescence and reflectance illumination fibers are located on a multiport translational stage, which moves with respect to the fixed light source depending on the type of illumination.

2.3.2.

Fiber optic probes

The fiber optic probe is used as a conduit for light to excite and collect an area of tissue for both fluorescence and reflectance spectroscopy.²⁶ The probes used in these studies were about the size of a pencil and illuminated an area of the cervix that was approximately $2\mathrm{mm}$ in diameter. The fluorescent illumination light travels through a bundle of excitation fibers, which is coupled to a mixing element. A mixing element enables uniform illumination and collection of light from the same area, specifically, $1.8\mathrm{mm}$ in diameter, on the tissue surface. The remitted light travels through the mixing element to the collection fibers. The excitation and collection fibers are randomly distributed across the mixing element.

Reflectance measurements were made at different source detector separations where the illumination fiber is located at various distances from the collection fiber. Each fiber is $200\mathrm{\mu }\mathrm{m}$ in diameter with a numerical aperture of 0.2, while the entire probe is approximately $4\mathrm{mm}$ in diameter. The probe is surrounded by stainless steel tubing that is approximately $20\mathrm{cm}$ long. A protective coating surrounds the fiber bundle to guard against excessive bending and mechanical damage. Figure 3 provides an illustration of the general characteristics for each system probe.

Fig. 3

Fiber optic probe schematic for (a) FastEEM2 and (b) FastEEM3 and (c) the illuminated FastEEM3 probe.

In the FastEEM2 probes, a fluorescence bundle is surrounded by a circular arrangement of reflectance fibers. A total of eight reflectance collection fibers at four different SDSs are located away from one illumination fiber. The separations are notated as 0, 1, 2, 3 and are located 0.25, 1.16, 2.13, and $3.02\mathrm{mm}$ , respectively, away from the illumination fiber. Epoxy glue stabilizes the fibers at the four SDSs. The fluorescence bundle fibers are located inside of the reflectance assembly and are collectively glued with optical grade epoxy to a quartz mixing element.

In the FastEEM3 probe design, a fiber disk composed of stainless steel precisely aligned the reflectance fibers and the fluorescence bundle, thereby diminishing the necessity for epoxy glue at the distal end of the probe. The metal disk also provides protection for the mixing element and reflectance fibers in case of mechanical impact. For diffuse reflectance spectroscopy in the FastEEM3, many illumination fibers at six source detector separations surround one collection fiber. In between the 0 and 1 positions, two additional source detector separations $(a,b)$ were added in the FastEEM3 probe and are located 0.5 and $0.74\mathrm{mm}$ away from the detector fiber, respectively.

2.3.3.

Collection components

The remitted light from fluorescence measurements travels through the collection fibers of the fiber optic probe and passes through long-pass filters. Long-pass filters enable wavelengths above a chosen cut-on wavelength to be transmitted, depending on the excitation wavelength selected. They remove the excitation light and enable the fluorescent light to continue to the camera. These filters are located on a wheel that is placed in front of the imaging spectrograph. Long-pass-filtered light is then sent into an imaging spectrograph, where it is dispersed by a grating. The long-pass filters we used had cut-on wavelengths of approximately $30\mathrm{nm}$ longer than the excitation light; wavelengths are selected between 300 and $800\mathrm{nm}$ . The imaging spectrograph is coupled to a CCD camera, which records light intensity. In general, the camera was thermoelectrically cooled to approximately $-30°\mathrm{C}$ to reduce dark current noise from thermally generated electrons. Reflectance light remitted from the tissue is collected in the equivalent manner as fluorescence emission, except that light travels through reflectance collection fibers of the fiber optic probe and did not pass through long pass filters.

The main difference in the collection subsystem between the two FastEEM devices was the imaging spectrograph. In FastEEM2, the imaging spectrograph contains a movable diffraction grating to maximize the area of illumination on the CCD. The FastEEM3 utilizes a fixed grating in the spectrograph because the moving grating in the FastEEM2 was subjected to wear from heat and age, potentially reducing the accuracy of the wavelength calibration.

2.5.1.

Fluorescence quality assurance and data processing

Fluorescence data from each measurement is reported as an EEM that represents fluorophores such as NADH, FAD, collagen, elastin, and hemoglobin. These fluorophores change in concentration as lesions advance from normal to low-grade to high-grade to cancer. Biologic plausibility studies confirm that not only is neoplasia important, but also biographical variables (such as menopausal status) and tissue type (the cervix has both columnar and squamous tissue). Each EEM underwent 10 reviews by the two senior engineers, the senior statisticians, and the principal investigator over a 2-y period. A forthcoming paper will describe agreement among these reviews. Approximately 5000 spectra were reviewed from both studies, and 20% were eliminated due to autofluorescence of one probe. We are working to restore those measurements, thus we show these data as a preliminary analysis in which we hope to demonstrate that each device/probe combination is able to separate those categories that influence the EEM such as neoplasia, tissue type, and menopausal status. These multidimensional data are reduced through a principal component analysis followed by binary classification using Mahalanobis distance. Further analyses using other statistical methods are ongoing at the time of this paper.

2.5.2.

Reflectance quality assurance and data processing

Each reflectance matrix was similarly quality assured in the aforementioned process. The data from each source detector separation can be represented by a matrix containing calibrated reflectance intensity as a function of the emission wavelength and the SDS distance. While less biologic plausibility studies have been performed on the reflectance data, mathematical modeling and analyses of patient data demonstrate differences among hemoglobin absorption, oxygen saturation, strength of stromal scattering, spectral shape of stromal scattering, collagen absorption, collagen concentration, keratin concentration, NADH concentration, FAD concentration, and epithelial scattering coefficients. In this analysis, we examine only SDS 1 or $1\mathrm{mm}$ . Chang ⁹ discusses the data processing and analysis of both fluorescence and reflectance data in more detail.