2.1.

Experimental Setup

We used a custom-built clinical spectrometer to collect white light reflectance and laser-induced fluorescence from skin sites on human volunteers. This system has been described in detail elsewhere.²¹ Briefly, the spectrometer system consisted of the following primary components [Fig. 1a]: 1. a tungsten halogen light source (Ocean Optics LS-1; Dunedin, Florida) for DRS, 2. a nitrogen gas laser (Newport VSL-337; Irvine, California) for LIFS, 3. fiber optic switch (Ocean Optics FOS-2×2-TTL; Dunedin, Florida), 4. a spectrograph (Princeton Instruments Acton SP2150i; Trenton, New Jersey), 5. a camera (Princeton Instruments Coolsnap HQ; Trenton, New Jersey), and 6. a 6-around-1 fiber optic probe (Fibertech Optica, NA = 0.22, 200-μm core diameter, 350-μm source-detector separation, 6.35-mm outer diameter; Kitchener Ontario, Canada) with the central fiber illuminating the skin and six surrounding fibers collecting the remitted light, shown in Fig. 1b. The system sequentially collected one reflectance and fluorescence spectrum every half second.

Fig. 1

(a) System schematic, (b) distal end of optical fiber probe, and (c) diagram of probe housed in a calibrated spring loaded case. The user held the probe by the spring-loaded case. The probe translated alongside the case, with the set crew moving accordingly. The spring was fully extended when the probe was completely extended and vice versa. The location of the set screw indicated the corresponding pressure applied on the skin.

To induce various pressures, we attached a calibrated spring-loaded case to the distal end of the fiber optic probe, shown in Fig. 1c. A pressure scale was marked on the case corresponding to the various pressures applied (P0 being only the weight of the probe, to P5 being the highest pressure applied). We characterized the accuracy of this pressure scale for each position using a separate scale (Adam AQT-600, Adam Equipment, Danbury, Connecticut).

Table 1 summarizes the results for pressures ranging from P0 to P5 (9, 22, 42, 77, 124, and 152 [mN/mm²], respectively. Error remained consistent for all pressures, ranging from 10 to 15 g. Pressure P5 of 152 [mN/mm²] is equal to approximately 22 [psi]. Typical pressures used in clinical measurements are mimicked by the P0 pressure. P0 is pressure resulting from only the weight of the probe supported by a flexible fixture, resulting in low pressure with high variance. For simplicity sake, we are reporting only the highest possible weight and pressure for P0.

Table 1

Pressure range.

2.2.

Acquisition Procedure and Volunteer Profile

Three replicates of one-minute measurements were taken on each volunteer's forehead, neck, and finger at each pressure separately. We took three replicate measurements to ensure that pressure effects are consistent. The replicate measurements’ locations did not overlap, but were within close proximity to each other. This is to prevent pressure effects from previous measurements to affect subsequent measurements. There is a half-minute break in-between measurements. We only measured P0 to P3 on the forehead as patients indicated discomfort for P4 and P5. Pressure (P1 to P5) was applied 5 s into the measurement, while P0 was acquired with the probe resting on the site supported by a flexible fixture. Measurements were acquired by the same user to eliminate interuser variability.

Five volunteers were involved in the study (University of Texas at Austin, IRB Protocol 2007-04-0099). Volunteers were within the age of 18 to 30, nonsmokers, and with no history of skin cancer. We restricted volunteers to Caucasians (Fitzpatrick skin type 1 to 2) to minimize variability due to skin optical properties across individuals.

2.3.

Data Analysis

A lookup table-based inverse model was fit to DRS data ^{21, 22} to obtain the following physiological parameters: [TeX:] ${\rm \mu^\prime }_{\kern-3pts}$ ${{\mu }^{\prime }}_s$ (630 nm), Mie theory slope, melanin, blood volume fraction (cHb), oxygen saturation (α), and hemoglobin packaging factor (HbF).

Briefly, reduced scattering coefficient and Mie theory slope provide insight into the sampled site's density and scatterer size, described by the following equation:

We applied the model described by Zhang ²⁶ to the LIFS data to extract the intrinsic fluorescence (IF) from the fluorescence spectra distorted by absorption and scattering. NADH and collagen are the main endogenous fluorophores excited in this wavelength region. Using basis spectra of NADH and collagen, a least squares fitting model was fit to the IF data to obtain intrinsic fluorescence contribution by each fluorophore.

Figure 2 is a plot of a typical diffuse reflectance spectrum along with the lookup-table inverse model fit and the corresponding LIFS spectrum with IF processing and model fit. These spectra were acquired from the finger. Data acquisition for one minute provides 120 DRS and 120 corresponding LIFS spectra (two spectra per second for 60 s).

Fig. 2

(a) DRS and (b) corresponding LIFS spectra with model fits. Raw LIFS data (red line) was processed to eliminate distortions due to scattering and absorption, resulting in the IF data (blue line). Least squares models of NADH and collagen basis spectra were fit to the IF (black dashed line). (Color online only.)

DRS and LIFS data were processed, and the described models were fit to the acquired spectra to obtain physiological properties for the time course of one minute. We report the normalized physiological properties over time (Δ) for comparison between sites and pressures using the following formula:

3.1.

Probe Pressure Effects Observed on Diffuse Reflectance Spectroscopy

Several probe pressure effects were readily observed on the acquired DRS spectra. Figure 3 illustrates typical reflectance spectra acquired at t = 0, 7 (approximately 2 s after pressure was applied), 25, and 60 s collected on the neck [Fig. 3a] and forehead [Fig. 3b]. We noticed consistent and site-specific probe pressure effects.

Fig. 3

(a) and (b): typical reflectance at t = 0, 7, 2,5 and 60 s. Pressure (P4 on the neck, P2 on the forehead) was applied at approximately t = 5 s. Notice the progression of the DRS spectrum with time. Hemoglobin absorption decreased gradually after probe pressure application.

Absorption due to hemoglobin in the 550-nm region decreased with time, indicating decreased hemoglobin. Figure 3 demonstrates the decreasing progression of hemoglobin absorption in the Q-band (550-nm region) for one minute, with pressure applied at t = 5s. At t = 0 and 7 s (black and red lines, respectively), hemoglobin absorption in the Q-band is more prominent, resulting in a deeper “valley” in the reflectance spectrum compared to DRS spectra at t = 25 and 60 s (green and blue lines, respectively).

Oxygen saturation (α) decreased, indicating a change from oxyhemoglobin to deoxyhemoglobin. The dual absorption peaks characteristic of oxyhemoglobin (black and red lines of Fig. 3) transitioned to a single absorption peak characteristic of deoxyhemoglobin (green and blue lines of Fig. 3). This decrease in α was consistent on all sites, shown in Fig. 4.

Fig. 4

Plot of physiological properties versus time s. Pressure was applied at approximately t = 5 s (black dashed line). Plots between physiological properties (rows) are not on the same scale. Plots of the same physiological properties (columns) between different sites are on the same scale.

Overall reflectance on the neck decreased during pressure application, indicating a decrease in scattering. In addition, a slight increase in absorption is seen by stronger Q-band absorption features around 540 nm, as shown by the black and red plots (t = 0 and 7 s, respectively) in Fig. 3a. On the other hand, no initial decrease of reflectance was observed on the forehead, as blood is compressed out of the sampling site immediately during pressure application. We hypothesized this site-specific effect is due to differences in tissue morphology, discussed in the next section.

3.2.

Probe Pressure Effects on Physiological Properties over Time

Figure 4 summarizes probe pressure effects on each of the physiological parameters (columns) on three different anatomical sites (rows) over time. Each line tracked changes in physiological properties (y axis) over the time course of one minute (x axis) with the various pressures represented by different colors as shown in the legend at the bottom of the figure. Each row corresponds to data acquired at each site (neck, finger, and forehead, respectively from top to bottom). The following sections discuss Fig. 4 in greater detail.

3.3.

Illustration of Skin under Probe Pressure

Figure 5 provides a simplified cross sectional view of skin under probe pressure with the optical path of sampling light. Normal epidermal thickness ranges from 70 to 90 μm, ^{29, 30} and normal dermal thickness ranges from 1 mm (eyelids) to 5 mm (back), varying according to the region's functional requirement.³¹

Fig. 5

Cartoon illustration of skin under probe pressure. (a) Cross section of the human skin without probe pressure exposure. (b) and (c): skin cross sections on the neck and forehead, respectively, under probe pressure. Notice light samples of the collagen-rich reticular dermis layer on the forehead.

On the neck, probe pressure compressed blood out of the sampling site, displaced water, and reduced the thickness of epidermis. As water is displaced, group refractive index increased, reducing refractive index mismatch. As a result, hemoglobin absorption and scattering decreased.

Similar events took place on the forehead, with one exception. Unlike the neck, the skull on the forehead prevented the dermis from collapsing along with probe pressure into the hypodermis. This caused light to sample the dense connective tissue layer of dermis that is rich in collagen, as shown in Fig. 5c. Blood and water evacuated the pressure site due to compression, and light sampled more collagen from dense connective tissue layers of the dermis. Scattering due to sampling of collagen dominates the decrease of refractive index mismatch, resulting in overall increase of [TeX:] ${\rm \mu^\prime }_{\kern-3pts}$ ${{\mu }^{\prime }}_s$ . This resulted in a decrease of hemoglobin absorption and increase of scattering and fluorescence due to collagen.

We are currently conducting experiments and simulations to determine the sampling depth of our probe. Preliminary results conducted on phantoms with optical properties similar to human skin have shown mean sampling depths of approximately 300 μm. It is unlikely for muscle to be sampled on the finger and neck, which predominantly consist of ligaments and fat. Whether sampling depths reach muscle on the forehead will depend on how human skin collapses under probe pressure, which might be different for each individual. While it is possible that muscle on the forehead is sampled, it is very unlikely for the skull to be sampled due to the relatively shallow sampling depth.

3.4.

Pressure Effects on Clinical Measurements and Implications on Skin Cancer Diagnostics

The described combined DRS and LIFS system is currently being tested as a noninvasive diagnostic of skin cancer.³² In this section we explore how probe pressure effects can implicate our clinical measurements and influence skin cancer diagnostics. In other words, we would like to know if probe pressure effects can negatively influence the outcome of skin cancer diagnostics.

As discussed in the previous section, our results demonstrate that long-term probe pressure effects can have significant distortions on physiological properties, even at the lowest pressure of P0, as illustrated by the black lines in Fig. 4 (cHb and α increased by 20 and 10% on the neck, and both decreased by 15 and 5% on the forehead). However, short-term light probe pressure (P0) effects are minimal. We focus on the initial portion of light probe pressure (P0) measurements, as it coincides with our clinical measurement acquisition procedure.

All of our clinical measurements consisted of an average of three replicates acquired within two seconds after the probe is in contact with the suspected lesion. The probe (contact area diameter = 6.35 mm) is handheld by an operator to be in full contact but not fully resting on the lesion. The resulting clinical acquisition pressure is effectively less than the weight of the probe resting on the lesion (P_clinic ≤ P0), verified using the same weigh balance (Adam AQT-600, Adam Equipment, Danbury, Connecticut). Thus, pressure effects on our clinical measurements rest comfortably within the first two seconds of the pressure P0 measurements. Figure 6 summarizes the percentage mean and standard deviation of physiological properties during the first two seconds under probe pressure P0. Measurements taken during the first two seconds after probe contact pressure (P0) ranged within ±10% for all extracted physiological properties. Pressure effects for diagnostically significant physiological properties are smaller (0 ± 2% for [TeX:] ${\rm \mu^\prime }_{\kern-3pts}$ ${{\mu }^{\prime }}_s$ , 0 ± 5% for HbF).

Fig. 6

Plots (mean and standard deviation) of physiological property changes for the first 2 s of acquisition.

We developed a logistic regression-based classification algorithm using two physiological properties ( [TeX:] ${\rm \mu^\prime }_{\kern-3pts}$ ${{\mu }^{\prime }}_s$ and HbF) to classify nonmelanoma cancers such as basal cell carcinoma (BCC), actinic keratosis (AK), and squamous cell carcinoma (SCC).³² The physiological properties were chosen by one-way analysis of variance (ANOVA), and ranked so that they showed the highest statistically significant difference between cancerous groups. The reduced scattering coefficient ranked the highest, followed by hemoglobin correction factor. The difference between physiological properties of normal and cancerous groups is large (>25%), such that probe pressure effects on diagnostic features (0 ± 2% for [TeX:] ${\rm \mu^\prime }_{\kern-3pts}$ ${{\mu }^{\prime }}_s$ , 0 ± 5% for HbF) should not negatively influence diagnostic outcome. We think short-term (t < 2 s) light probe pressure (P0) effects should be tolerable for this particular application. However, light probe pressure (P0) will eventually have significant site-specific effects (black lines of Fig. 4). Our results demonstrate that pressures greater than 9 [mN/mm²] (approximately 1.3 psi) for longer than 5 s induced significant distortions in collected spectra, and resulted in significant changes on physiological properties as shown in Fig. 4. This emphasized the importance to acquire measurements immediately on probe contact.

We noticed that measurements acquired on the finger and forehead are more prone to motion artifacts. Activities such as blinking, talking, and pulsation can contribute to significant spectral changes. On the other hand, areas with thicker muscle and tissue mass have a larger “buffer” zone, damping noise by motion artifacts. This can be seen in Fig. 6, where standard deviations for measurements on forehead and finger are generally larger compared to the neck. Care must be taken to mitigate motion artifacts while collecting spectra at low pressures. We can request the patient to refrain from speaking and keep their eyes closed during acquisition on the forehead, and avoid sites sensitive to pulsation as best as possible.

3.5.

Alternatives to Alleviate Probe Pressure Effects

Given that the effects of probe pressure can be significant under certain conditions, several alternatives exist to alleviate the effects of pressure. One technique to alleviate probe pressure effects is to reduce pressure by increasing the probe's contact area. Cerussi reported less than 5 and 2% changes in absorption and scattering, respectively, for probe pressure on in-vivo human breast tissue for forces up to 2 N.¹⁹ Our results showed much bigger probe pressure effects on absorption for similar amounts of force (P3 = 2.45 N). However, there are significant differences in our probe geometry and sampling depths. Their probe contact area is approximately 3000 mm² with a source detector separation of 28.5 mm, both substantially larger than our probe's geometry. The much deeper sampling depth contributed to the small probe pressure effects on optical properties. Increasing our sampling depth to a similar distance is not an option, as we would like to sample the human skin that is relatively superficial. However, a larger contact area will distribute the force over a larger area and reduce probe pressure, and as a result might “dampen” probe pressure effects. Another technique that will alleviate probe pressure effects are noncontact probes or imaging systems. ^{11, 33, 34}

Zhu proposed a statistical technique called error removal by orthogonal subtraction (EROS) that reduced variability in spectra from replicates measurements of the same sample.³⁵ Some of the sources of spectra variability are small changes in angle and probe pressure. EROS was reported to reduce complexity and increase accuracy of qualitative classification on colon lesions³⁵ and Barrett's esophagus.³⁶ Our results showed that short-term light probe pressure has negligible effects under clinical measurements conditions, thus we do not think it necessary to add an additional step in our postprocessing procedure. Furthermore, our current diagnostic model only used two parameters, thus reducing model complexity is not necessary. Improvement in diagnostic accuracy makes EROS worth exploring in the future.