2.1.

VLS Oximeter Design

Key clinical design goals included: 1. quantitative microvascular hemoglobin oxygen saturation measurement, 2. the ability to operate embedded into therapeutic catheters, 3. clinical ease of use, and 4. fast response times facilitating interventional procedures.

We began with a laptop-based VLS system that optically characterized or classified living tissues using partial least-squares (PLS) analysis and a nonscanning spectrophotometer,²⁶ and optimized this system over four generations of instruments.²⁷ While such a nonscanning device is becoming the obvious approach today for the rapid analysis of signals in vivo, when first constructed, this approach was novel. Briefly, we configured a charge-coupled device (CCD) spectrophotometer to be visible-light sensitive. The detector (ILX511 linear sensor, Sony, Japan) consisted of a 2048-element array with a $12.5\text{-}\mu \mathrm{m}$ width ( $14\mu \mathrm{m}$ center to center) by $200\text{-}\mu \mathrm{m}$ pixel area. Unused leading and trailing bins were masked to estimate dark count and offset voltage at each integration time (5 ms to 1 s). Sensitivity was 0.03 fJ per count at 600 nm (1 count/86 incident photons/pixel), but this fell to 1 count per 50,000 photons when accounting for fiber coupling (35%), grating efficiencies (70%), and photon capture (0.6% efficiency for multiply-scattered light into a 12-deg fiber-capture half angle). This yielded linear photon counts from 2 pW to 16 nW incident on the detector fiber.

The resulting clinical system was designed to meet United States [e.g., Food and Drug Administration (FDA)^{28, 29}] and European Union (e.g., CE mark^{29, 30}) medical device requirements, including software^{31, 32} and device³³ guidelines for mitigation of hazards, design control, verification, validation, biocompatibility, manufacturing, and safety. Manufacturing was transferred to an FDA-certified assembly facility. FDA approved VLS systems began shipping earlier this year.

2.2.

Analysis Software

Spectral standards were provided in software-readable Windows® INI files. For oximetry, hemoglobin standards were provided (oxy, deoxy, and met species). Integer wavelength alignment of collected spectra corrected for grating differences between systems. Standards were then used to solve for the concentration of each standard spectrum in the tissue using a scatter-corrected least-squares matrix fit on native or differential measured spectra. Iterative feature stripping allowed for complex analysis (e.g., solving for hemoglobin at one wavelength range, then solving chemotherapy drugs at another). Error checking allowed abnormalities to be detected (e.g., known abnormal hemoglobin in trial fits, or unknown spectral contaminants causing excessive fit residuals).

Data processing was governed by a readable INI file, written in a custom macro language script. In the oximeter, analysis is based on first differential spectroscopy, with a fit of scattering and absorbance from 476 to 586 nm. Calculation of absorbance, filtering, derivative calculation, and fitting were performed in real time, and required 30 ms per averaged spectrum set using an embedded single board computer.

Use of INI files for both standards and processing allows the VLS system to be user configured as required for tissue oximetry, tissue identification or characterization, or parametric probe guidance during interventional procedures, all using the same validated core software kernel. Examples of alternative analysis include the identification of chemotherapy agents such as Doxorubicin or Tirapazamine,³⁴ probe guidance for targeted injection of therapeutic substances, and monitoring of localized nonsurgical tissue ablation.^{35, 36}

For research use, a password-protected operating mode allowed standards and scripts to be modified, and data to be saved to an internal flash memory and later exported in Microsoft Excel-compatible form via the external USB port. Because an oximetry system with modified standards or scripts must never be confused with validated clinical oximeters, a self-test detects user-altered systems and prominently indicates at startup that the software has been modified, and that the device is for “research use only.”

2.3.

Light Source Development

There are significant drawbacks to the bulb sources typically used for broadband optical spectroscopy. Unlike narrow, coherent sources such as lasers, broadband sources are difficult to couple to a biological sample. They tend to emit light over a wide spherical angle from nonpoint sources, rendering inefficient attempts to direct their light either onto tissue samples or into fibers, and they tend to run hot, discouraging close approximation to living tissue.

For illustration, consider a 1-cm filament bulb. The bulb’s surface area is $105{\mathrm{mm}}^2$ , while the area contacting a 1-mm tissue sample placed against the glass is only $0.8{\mathrm{mm}}^2$ . Thus, the target tissue intercepts less than 1% of the bulb’s visible light output. In practice, because this bulb is hot, a fiber is required to transmit the light to the tissue. When accounting for both the low conversion efficiency of input energy to visible light (about 4% for conventional bulbs) and the poor transfer of light to tissue via a $100\mu \mathrm{m}$ fiber, only 0.0003% of the energy flowing into this bulb ends up transmitted to the tissue—equivalent to 333 W of input energy required for each mW of visible light delivered. To address this issue, we developed two miniaturized light sources.³⁷

First, we developed a high-intensity fiber-coupled halogen source [Fig. 1(a) ]. In this system, an integrated lens in the light bulb housing, placed only a few millimeters away from the filament, couples light into a collimated beam, while a reversed beam expander (NA 0.55, FL mm) separated by an air gap from the bulb for thermal isolation, focuses the light in a collimated beam directly onto the fiber face. Reducing the distance from the filament to the lens allowed use of a shorter focal length and higher NA lens, enabling the fiber to intercept a larger fraction of the emitted light.

Fig. 1

Light sources developed. (a) A small but intense light source developed under SBIR support produces $3.5\mathrm{W}∕{\mathrm{cm}}^2$ . The device measures $25\times 25\times 38\mathrm{mm}$ and requires no cooling fan. (b) A broadband, low-thermal-transfer LED light source deployed into the tip of the probe itself, delivering a similar intensity of illumination for much less power and lowered local heating of the sample.

Second, we developed a white LED source for integration directly into the probe itself for high light coupling with low transmission losses [Fig. 1(b)]. Because a broadband LED emits primarily in the desired band, it runs more efficiently, allowing for cool operation and close juxtaposition of the tissue and source. This proximity, in turn, further raises the efficiency of light transfer, as a narrow-aperture coupling fiber is no longer required. An LED also allows for an inexpensive electrical connection of the bulb to the monitor, rather than a fiber coupling, or even for no electrical connection at all if a disposable power source is embedded in the probe.

2.4.

Probe Design

A goal of probe design was interchangeability, allowing different therapeutic catheters and probes to be attached to the oximeter. Early probe designs²⁷ were refined and reduced to a reproducible set of application-specific clinical probes for human use. These probes include a $6\times 200\text{-}\mathrm{mm}$ hand-held wand constructed for neurosurgical, plastic, cardiac, and vascular surgery applications, used here for skin studies [Fig. 2(a) ]; a 2-mm-diameter endoscopic catheter designed for invasive studies, used here for human gastrointestinal mucosal studies [Fig. 2(b)]; a clip-on probe designed for buccal placement [Fig. 2(c)]; a 27-Ga needle probe developed for tumor oximetry and chemotherapy level estimation [Fig. 2(d)]; a 5-mm-diameter esophageal monitoring catheter, used here in human gastrointestinal studies [Figure 2(e)], and a 12-mm-diameter flexible colonic probe, used here for human gastrointestinal studies [Fig. 2(f)]

Fig. 2

Clinical probes used in human trials. Application-specific VLS oximetry probes, including (a) a $6\times 200\text{-}\mathrm{mm}$ hand-held wand, (b) a 2-mm-diam endoscopic catheter, (c) a clip-on buccal probe, (d) a 27-Ga needle probe, (e) a 5-mm-diam esophageal monitoring catheter, and (f) a 12-mm-diam flexible colonic probe used here for human gastrointestinal studies.

2.5.

Calibration Standards

2.6.

Ex Vivo Methods

2.7.

In Vivo Methods

VLS oximetry was recorded using the clinical probes at one or more sites in each subject from the skin, and from mucosal surfaces in the cheek, esophagus, stomach, intestine, and/or colon.

First, surface oximetry was performed on skin in a series of noninvasive measures on the skin. Hand measurements from a Caucasian, the dorsal surface of an African American, the ear lobe of a Caucasian, and the lip of a Caucasian were obtained using the wand probe [Fig. 2(a)].

Second, mucosal oximetry was measured in a series of nonpenetrating, reflection measurements of the inner surface of the gastrointestinal tract during a routine endoscopic exam. Endoscopic measures were collected after Institutional Review Board (IRB) review, Human Studies approval, and with written informed consent. Measurements were made using the clinical endoscopic VLS catheter [Fig. 2(b)] or rectal catheters [Figs. 2(e), 2(f)] held a few millimeters away from the mucosal surface.

Third, buccal oximetry was measured from the inner surface of the cheek during planned cardiac arrest during cardiac repair. Buccal measures were collected after IRB review, Human Studies approval, and with written informed consent. Measurements were made using the buccal VLS clip [Fig. 2(c)], as shown on a human subject (Fig. 3 ). The percentage of time during which the tissue oximeter and the pulse oximeter collected data during cardiac surgery was monitored. Because there are periods during which there is no pulse, up time was determined as the percentage of 5-s windows during which at least one valid oximetry signal was recorded.

Fig. 3

Buccal probe on adult cheek. The white LED light illuminates the inner mucosal side of the cheek.

The goal of these in vivo tests was to demonstrate that the expected VLS spectral features were seen in living subjects, thus confirming that VLS accurately records spectroscopic information from human subjects. Validation of VLS oximetry measurements in a large population of patients is reported in a companion article (see discussion).

3.1.

Hardware Performance Results

3.1.1.

Light source performance

Halogen lamp source visible light intensity was captured and measured using a $100\text{-}\mu \mathrm{m}$ fiber either directly coupled to the bare bulb (3.4 to $7.5\mathrm{mW}∕{\mathrm{mm}}^2$ ) or coupled through an integral lens in the bulb (67 to $152\mathrm{mW}∕{\mathrm{mm}}^2$ ; 1.20 mW peak total). Incorporation of an integral lens, inside the bulb and close to the filament, allowed for a 10 to 45 fold improvement in light collection.

The cool LED light illumination of tissue was measured using illumination either captured and delivered through a $100\text{-}\mu \mathrm{m}$ fiber (0.37 mW delivered to the tissue) or by direct LED illumination of the tissue after placement of the LED into the tip of a medical instrument (3.2 mW delivered to tissue). The spectrum of light measured when embedded into the buccal probe is shown in Fig 4 . Direct coupling of the cool source increased light delivery to the tissue by 10 fold. This exceeded the visible light delivery from the lens-coupled halogen bulb by 3 fold, despite a much lower light density of the LED source, due to the larger effective aperture achieved through direct illumination.

Fig. 4

Output of the phosphor-coated blue LED. Power is measured as per the ANSI standard test for eye safety for light emitting devices, in which the light is passed through a 3.5-mm slit and measured at 100-mm distance. The peak emission from the blue LED is visible at 463 nm, while the phosphor produces an overlapping broadband peak extending the range of illumination to more than 700 nm.

3.1.2.

Probe performance

Returning light levels for the end-emitting needles [Fig. 2(d)] was 56 nW in 2% IL, or about 0.01% of the transmitted input fiber light of $520\mu \mathrm{W}$ ( $67\mathrm{mW}∕{\mathrm{mm}}^2$ into a $100\text{-}\mu \mathrm{m}$ core fiber). Throughput for the optical forceps varied with separation thickness, but was not detectable for tissue thicknesses of 1 cm or greater. Cross talk varied significantly by probe configuration. End-emitting needle probes were susceptible to significant cross talk in air, from 0.01% to more than 1% of total input signal, typically from scattering through the cladding at the needle end. Side-emitting, fiber array, and forceps probes had little or no detectable cross talk. Cross talk was best reduced by the use of dense black epoxy potting material as well as by the presence of a metallic barrier between the fibers.

2.1.3.

End angle polish

Medical needles tend to be sharp, with narrow end angles to facilitate tissue penetration with minimum local blunt trauma. However, light exiting the fiber in tissue was substantially diminished due to internal reflection when the angle of the tip was less than 30 deg, even when in contact with moist tissue (refractive index about 1.36) rather than air $(\text{index}=1.00)$ . This is likely due to the high refractive index of the optical fibers ( $n=1.46$ to 1.50). Because of this, sharp polishing to a medical angle of 15-deg results may not be practical.

3.2.

Standards

3.2.1.

Reflectance standards and probe calibration

The spectral flatness and total reflectance of the tested standards varied widely (Fig. 5 ). Some standards exhibited spectral features (such as water and fat features in the intralipid standard above 900 nm, or the 912-nm feature in the silicone standard). The noncontact and surface-contact probes were best calibrated using dully surfaced diffuse reflectance materials, which reduced specular reflections, allowed the angle of measurement to be of low importance in the standardization, and exhibited good reproducibility of the reflected intensity.

Fig. 5

The spectral flatness of the reference materials tested. Absolute reference is an optically flat NIST-traceable diffuse spectral standard. All other standards demonstrate a degree of spectral response. Small-fiber paper standards, such as thick, quantitative filter paper or commercial dust-free wipes, were the most consistent, easily made standards.

The invasive needle probes, unlike surface contact probes, demonstrated large variations in coupling to solid standards, with returning light intensity for a given probe showing a repeated measure variation (1 SD) of 12.8% of mean intensity. Liquid standards produced less variation (0.2% of mean intensity). However, lipid-based aqueous standards also produced unpredictable wavelike variations in the reference curves (Fig. 6 ), especially after the lipid samples were left standing. These oscillations, attributable to Mie-like scattering effects, introduced error into the reference spectrum, which then carried into clinical measurement. This error was unacceptably high when testing for low-concentration compounds such as chemotherapy agents, or when using higher-order differential spectroscopy. In contrast, the soft solid polysiloxane-based standards did not exhibit such similar oscillation, but exhibited a return intensity that varied with needle pressure. Polysiloxane coupling variability was reduced using a consistent compressive force between the standard and the probe (0.3 to 0.6 N), and by lowering the scattering coefficient of the standard. Such intensity-related measures would be of less importance in approaches that are not intensity based (e.g., frequency and time-domain measures). Last, the addition of a small amount of dye to the standard allowed for estimation of volume measured.

Fig. 6

Oscillations from liquid Intralipid-based standards. Use of Intralipid solutions for reference scattering for the needle probes at times produced wavelike variations in the reference curves, consistent with Mie-theory variations in some studies. These oscillations were most notable after differential spectroscopy, and introduced error into the quantitation of concentration. As a result, we discontinued use of liquid standards for the VLS calibration.

3.2.2.

Dye

The spectral effects introduced by scattering were reduced by use of differential spectroscopy.⁴⁶ In the absence of a scattering correction, the match between the unscattered and noncontact scattering waveform spectral errors improved using differential spectroscopy, which emphasizes nonlinear effects over the baseline offset and exponential effects induced by scattering and variations in the coupling. When further including a scattering correction in the fitting method, calculated dye concentration was independent of scattering over a wide range of scattering levels (scattering 0.5/mm to 1.5/mm $(r^2=0.97)$ , absorbance 0.005 to 0.025/mm $[r^2=0.99]$ , Fig. 7 ). A ratio of the measured concentration of two dyes, a model for the determination of saturation used in tissue oximetry, was independent of scattering over the range of scattering coefficients seen in most tissues ( $\text{ratio}=0.550\pm 0.020$ , or a variance of 3.6%). This suggests that ratiometric measures, such as hemoglobin saturation, should be stable and independent of the tissue monitored in vivo.

Fig. 7

Predicted versus actual absorbance and scattering plots. Scatter- and differential-corrected plots for measured (a) absorbance and (b) scattering are shown for a range of lipid and dye concentrations intended to cover the physiologic range of expected values in tissue in vivo. The line of identity is shown as a dashed line in each plot.

3.2.3.

Hemoglobin

The measured spectra of oxygenated and deoxygenated hemoglobin compare well to published values. Spectral peaks agree with a mean difference of $1.0\pm 0.8\mathrm{nm}$ ( ${\mathrm{HbO}}_2$ peaks measured at 543 and 578 nm versus published values of 542 and 578 nm, Hb peaks measured at 557 nm versus published values of 556 nm). These differences became insignificant using differential spectroscopy. A plot of calculated hemoglobin saturation versus measured electrode ${\mathrm{pO}}_2$ demonstrates an oxygen dissociation curve with the expected sigmoid shape (Fig. 8 ) and good p50 agreement with values published in the literature³⁸ (p50 measured $=35\text{-}\mathrm{mm\; Hg}$ , p50 $\text{expected}=33\pm 2\text{-}\mathrm{mm\; Hg}$ under the conditions of pH 6.6, no 2,3-DPG, and 20°C).

Fig. 8

Deoxygenation binding curve as measured using VLS. The measured p50 (the point at which 50% of the hemoglobin is saturated with oxygen) is 35-mm Hg. The expected p50 for human hemoglobin is $33\pm 2\text{-}\mathrm{mm}\mathrm{Hg}$ under the conditions of pH 6.6, no 2,3-DPG, ${\mathrm{pCO}}_2=0.6\text{-}\mathrm{mm}\mathrm{Hg}$ , and 20° C.

3.3.

In Vivo Spectrophotometry

The spectral signals recorded from noninvasive measures on the hand of a Caucasian subject, the dorsal hand of an African American subject, on the ear lobe of a Caucasian subject, the lip of a Caucasian subject, and the colonic mucosa of a human subject undergoing endoscopy are shown in Fig. 9(a) . Each spectrum required between 25 to 45 ms to collect. The signals show substantial variation in absolute reflectance intensity, background scattering, and hemoglobin signal strength. A fit assuming the presence of adult hemoglobin plus scattering accounted for the most of the observed spectral shape (mean rms fit $\text{error}=2.7\%$ ). Background variation and coupling errors are reduced using differential spectroscopy [Fig. 9(b)], with the variance in calculated saturation using repeated measures on the same skin sample falling from 2.7% for absorbance to 1.5% for first differential and 1.3% for second differential spectroscopy.

Fig. 9

Differential spectroscopy applied to in vivo spectra. (a) Absorbance plots show differing baseline and slope in human tissues measured in vivo. (b) Second differential plots remove much of this baseline variation, and enhance the nonlinear spectral features, such that the hemoglobin peaks from different tissues substantially overlap. Note the 480- to 520-nm feature in the colonic mucosal spectrum seen more clearly in the differential plot.

For the human gastrointestinal studies, 804 measurements were performed in various segments of the gastrointestinal tract in 45 normal patients. Normal mucosal hemoglobin saturation $(\text{mean}\pm \mathrm{SD})$ was $69\pm 4\%$ , while in tissue later determined to be tumors, this value was $45\pm 23\%$ ( $n=14$ , $p<0.0001$ ). The normal values were demonstrated to be normally distributed (Kolmogorov-Smirnov test of normal distribution for continuous variables, $p<0.005$ ). The tight variation in the normal values suggests that VLS oximetry is measuring reliably and reproducibly in tissue; the low oxygenation measured in human spontaneous gastrointestinal tumors suggests that VLS spectroscopy is sensitive to baseline tumor ischemia.

For the human cardiac surgery studies, the up time of the system during various stages of cardiac surgery is shown in Table 1 . The VLS oximeter had little down time during these studies. Unlike pulse oximetry, hand or body motion did not interfere with the ability of the VLS oximeter to collect spectral data, and to report oximetry results.

Table 1

Percent up time of T-Stat® oximeter versus pulse oximeter during cardiac surgery. Up time was calculated as percent of 5-s intervals during which at least one valid data point was recorded.