2.1.

Camera System

The camera system is schematically depicted in Figure 1. An incoherent white light source (xenon lamp source, Oriel, Stratford, CT) was passed through a long-pass transmission filter (>500 nm wavelength) and a linear polarizer (Ealing Electrooptics, plc., Watford, UK) to yield linearly polarized light. The wavelength of illumination does not strongly influence the depth of imaging which is governed by the rates of depolarization by birefringent tissue structures such as collagen or actin-myosin fibers and depolarization by scattering. Laser light was avoided because laser speckle interfered with the polarization images. The light was collimated by a 20 cm focal length lens and delivered to the skin at an angle of 15° to the normal to the skin surface. The choice of angle is not critical and oblique angles of illumination beside 15° also work. A glass plate contacted the skin and was optically coupled to the skin by a drop of water. Glare from the air/glass/skin interfaces was directed away from the viewing camera. Because polarized light imaging is based on scattering rather than on absorption, there are no problems with the contact pressure of the skin against the glass plate blanching the cutaneous blood vessels. The camera (Princeton Instruments, Trenton, NJ; 16-bit CCD camera) with macro lens (Nikon, Melville, NY; f 105 mm, 1:2.8 D) only collected light that had entered the skin and been backscattered toward the camera. For some of the experiments reported here, a razor blade was placed in the illumination beam to cast a shadow on the skin surface. A second analyzing linear polarizer in front of the camera was manually aligned either parallel or perpendicular to the orientation of the polarization of the illumination light which was parallel to the plane defined by the source-skin-camera triangle. Two images were acquired, one called I _par when the analyzer was parallel to the illumination and the second called I _per when the analyzer was perpendicular to the illumination. The camera integration time for each image was 1 s, although shorter times are possible with stronger illumination. Images were acquired as 600×600 images with 34×34 μm ² pixels using a library of C programing code and a custom program written with MATLAB™ software to control the camera.

Figure 1

Schematic of camera system. Collimated white light source is filtered by a bandpass filter and passed through a linear polarizer oriented parallel to source/sample/camera plane. Sample is coupled to glass plate by drop of water. Specular reflectance from air/glass/skin surface is reflected away from the camera. Light that enters the tissue and backscatters toward the camera passes through an analyzing linear polarizer oriented either parallel or perpendicular to the orientation of the source polarization, yielding two images called I _par and I _per , respectively.

2.2.

Signal Processing

The analyzing linear polarizer in front of the camera was oriented parallel to the illumination light (I₀) to acquire an image called I _par . The I _par image consisted of the superficially reflected light (R_s) plus one half of the deeply penetrating light (R_d). The term “deeply penetrating” refers to light penetration into the reticular dermis perhaps to a depth of about 300 μm or more. Any epidermal melanin acted as an absorption filter on the skin surface with a roundtrip in/out transmission of T _mel . The I _par image is described:

2.3.

Illustrating Optical Coupling

Two subjects (Caucasian) were imaged at the Oregon Medical Laser Center, Providence St. Vincent Medical Center. One was a normal adult skin site. The second was a burn site on an adult who was burned as a youth. These images are shown in Figure 2.

Figure 2

Images of (left) normal skin site, and (right) skin burn scar. (a) I _par image. (b) I _per image. (c) I _par +I _per image. (d) Pol image. Notice I _par image is slightly brighter than I _per image due to reflectance of incident light by superficial layers. Pol image isolates this superficial scattering and reveals the structural details of the superficial layer. Note the boundary between areas coupled and not coupled by a drop of water, illustrating the importance of water coupling of the glass plate to the skin in order to reject the glare specularly reflected from the air/tissue surface. In the normal skin site (left), notice the dermal structure revealed in the Pol image. In the burn skin site (right), notice that the burn appears whiter than normal skin in the I _per and I _par images but appears darker than normal skin in the Pol images.

2.4.

Illustrating the Point Spread Function

A third subject (Caucasian) was imaged at the Oregon Medical Laser Center (same system as earlier) while casting a shadow from the edge of a razor blade onto the skin (Figure 3). The razor blade cast a sharply defined shadow on the skin. The light that entered the skin in the illuminated skin area was multiply scattered and migrated through the skin into the shaded area. Light collected as I _par and I _per from the shaded area had experienced multiple scattering events within the tissue. The Pol image was calculated and the distance to which the I _par , I _per , and Pol signals extended into the shadow provided an estimate of the point spread function for crosstalk between pixels in Pol images of skin. A white paper card was imaged to specify the position of the shadow edge as the midpoint between the maximum values and minimum values. For each image, the I _par and I _per values along a single x-axis line perpendicular to the edge of the shadow was acquired using MATLAB™. The average of 300 such line traces was determined. This average was plotted versus x to indicate the I _par and I _per behavior relative to the edge of the shadow. The Pol values versus x were calculated using these average I _par and I _per traces.

2.5.

Pilot Clinical Study

In a pilot clinical study of human patients [Figures 4(a)–4(o)], the color filter was omitted and the images used a full-spectrum of white light source (Dolan-Jenner Industries, Woburn, MA). Images were acquired as 435×548 images with 39×39 μm ² pixels using IPLab™ to control the camera. Analysis used IPLab™ software.

The CCD camera was mounted on a universal joint on a balanced photographic boom arm (Red Wing) so that the camera could be positioned normal to any skin surface of a patient. Ten patients in the dermatology clinic of the Veterans’ Administration Hospital, Oregon Health and Science University, who presented various skin pathologies were imaged. The camera was coupled to the skin by a drop of water. Two images were acquired in a few seconds while the skin site was held stationary against the glass plate in the image plane of the camera. Image processing was accomplished at a later time.

3.1.

Normal Skin and Skin Burn Scar

Figure 2 shows four images each for a normal skin site and a skin burn site that was acquired by the adult subject as a youth and had undergone scarring. The four images are I _par , I _per , I _par +I _per , and Pol. There was a slight increase of reflected light in I _par relative to I _per due to the superficial scattering of incident polarized light. The I _par +I _per was equivalent to the total reflectance that one would observe by eye. The I _par , I _per , and (I _par +I _per ) images were all very similar. The Pol was very different and emphasized the scattering of incident polarized light by superficial tissues.

The tissue was coupled to the glass plate by a drop of water and the water-line margin between coupled and uncoupled areas is visible in the figures. The Pol image shows how the specular reflectance from the air/tissue surface dominated the image when there was no optical coupling between the glass plate and the skin. The skin area not coupled by water to the glass plate presented a large amount of specularly reflected light. Since the skin surface was rough, there was surface glare despite the illumination light being delivered from an angle. The signal contributed by such glare was stronger than the Pol image of interest. Without the optical coupling, the desired Pol image was masked by this surface glare signal.

Figure 2 (left) shows the images for the normal skin with two pigmented nevi. The texture of the dermis is revealed in the Pol image as striations. The pigmentation of the two superficial nevi are largely canceled in the Pol image.

Figure 2 (right) shows the images for the skin burn scar. The I _per and I _par images show the burn scar as a white region due to the strong backscatter from the collagen fibers induced by the burn trauma. However, the Pol image shows this region as a darker signal relative to normal skin. The collagen fibers in the burn scar are randomizing the polarized illumination faster than they are backscattering the polarized illumination. The balance between the rate of randomization and the rate of backscattering may be influenced by the size of collagen fiber bundles in the scar. While the I _per and I _par images show a white scar, the Pol image reveals some structure in the scar, suggesting that Pol images may prove useful in evaluating the topography of scarring.

3.2.

Estimating the Point Spread Function

Figure 3 shows the experiment where a razor blade cast a shadow onto the skin surface. The experiment illustrates how light diffuses from an illuminated area into a shaded area on a forearm skin site. Figure 3(a) shows I _par and I _per as functions of position relative to the edge of the shadow. Also shown is the intensity reflected from a white paper card which identified the shadow edge. The intensities begin to fall within the illuminated area near the shadow edge due to the effect of the boundary condition on diffusion of light. Since the shadow does not contribute to light diffusion, the concentration of photons in the tissue drops in the illuminated region near the edge of the shadow.

Figure 3

Shadow of razor blade on human skin in vivo. (a) Average intensity as function of position, x, along axis perpendicular to shadow of razor blade for I _par , I _per , and intensity from a white card. (b) Pol values of skin as function of x. Dashed vertical line indicates edge of shadow.

Figure 3(b) shows the calculated Pol values as a function of position. The Pol values are about 0.08 in the illuminated area but increase significantly at positions closer to the shadow edge. The Pol values drop to zero within the shadow distant from the shadow edge. The peak of Pol at the edge drops off with a 390 μm half-width-half-max extending into the shadow and the same half-width extending into the illuminated area. The Pol values do not fall exactly to zero deep within the shadow. Far into the shadow distant from the shadow edge the I _per was 1.6 higher (76 counts) than the I _par rather than being equal as expected for light that has been randomized by diffusion. This difference was thought to be an artifact of the optical system.

The point spread function for Pol in Figure 3(b) showed a peak value near the shadow edge. Of course the deeply penetrating photons that contribute to I _par and I _par will cancel due to the subtraction in the numerator during calculation of Pol [see Eq. (2)]. Consider the role of early scattered photons from superficial layers that contribute to both I _par and I _par . Let there be two neighboring tissue sites A and B associated with camera pixels A and B, respectively. The Pol value for pixel A depends on the superficial I _par that reflects from tissue site A and on the superficially scattered I _per that has scattered from tissue site B into tissue site A and escaped from tissue site A to also contribute to pixel A. If tissue site A is at the shadow edge and tissue site B is in the shadow, there are no I _per photons contributed from tissue site B to tissue site A and pixel A. Therefore, pixel A at the shadow edge is depleted of superficial I _per photons that normally might have been scattered from tissue site B into tissue site A and escaped from tissue site A. Depletion of I _per photons causes the Pol value for pixel A to increase. The extent of the Pol peak into the illuminated regions characterizes this depletion of superficially scattered I _per photons due to the shadow. Some incident photons can also scatter into neighboring pixels and escape as I _par photons. The extent of the Pol peak into the shaded region characterizes this scattered contribution of superficial I _par photons into neighboring pixels.

3.3.

Pilot Clinical Study

Figure 4 shows a series of images that compare normal light images with polarized light images. On the left of each pair of images is the normal light image, represented by I _per . The choice of I _per provides an optimal diffuse light image by avoiding specular glare. As evident in Figure 2, the I _per and (I _par +I _per ) images are essentially equivalent. On the right of each pair of images is the polarized light image, Pol.

Figure 4

Images of skin sites representing various skin pathologies. For each pair of images: (Left) normal light image represented by I _par image; (Right) polarized light image represented by Pol image.

Figure 4(a) shows a ruler to specify the field of view. Figure 4(b) shows a freckle. The Pol images successfully removed the superficial epidermal pigmentation of the freckle to reveal normal skin underneath.

Figure 4(c) shows a benign pigmented nevus. The Pol image did not cancel the melanin pigmentation of the nevus but rather showed an enhanced Pol signal. One possible explanation was that the melanosomes of the nevus backscatter the incident polarized light, as they do during confocal reflectance microscopy,²² although the superficial melanosomes of freckles did not behave this way. A second possible explanation is that when melanin pigmentation is located below the surface the melanin no longer acts as a surface filter and the cancelation of T _mel in Eq. (2) is incomplete and an apparent Pol signal is generated. A third possible explanation is that the nevus had structure that backscattered polarized illumination. The Pol image showed a dark area corresponding to the epidermal infolding around each hair follicle, which was not seen with normal light. A possible explanation is that the polarized illumination light is not strongly reflected by the epidermis relative to the dermis and so the Pol signal is decreased around the hair follicles. In the skin areas between hair follicles, the Pol signal is stronger and may originate from the initial backscatter from the collagen fibers of the dermis before the collagen fibers can depolarize the illumination. In the Pol figure, the texture of the dermis seems evident as if the Pol image is faithfully imaging the papillary and upper reticular dermis. Figures 4(d)–4(f) show compound pigmented nevi illustrating the variation in images that can be encountered clinically.

Figure 4(g) shows a black tattoo. The tattoo particles reflect a strong Pol signal which presents as a white coloration in the Pol image. Apparently polarized light is penetrating sufficiently deep to encounter these particles and be reflected as still polarized I _par . A working hypothesis is that the tattoo particles present a strong refractive index difference that causes strong backscattered reflectance of the polarized illumination light.

Figures 4(h)–4(o) are nonpigmented lesions: a nonpigmented intradermal nevus, neurofibroma, actinic keratosis, malignant basal cell carcinoma, squamous cell carcinoma, and vascular abnormality (venous lake). In all cases, the Pol image provides an image that captures the texture and structure of the lesion as it invades or modifies the surrounding normal skin.

During acquisition of the I _par and I _per images, any movement by the patient resulted in the misregistration of the two images. One common example is that hairs sometimes showed as a pair of black and white lines. Such misalignment was clearly evident in the tattoo of Figure 4(g). However, the skin pressed against the glass was more stationary than the hairs, especially thicker hairs that can behave like little springs and move while the underlying skin remains fixed by the surface contact with the glass plate. A new camera system, now being completed, overcomes this problem by automating the rotation of the analyzing polarizer to provide rapid image acquisition that minimizes movement artifacts.

Another common artifact in these pilot study images occurs when the water droplet used for optically coupling the skin to the glass plate captures some air bubbles and traps a glass/air/skin interface within the field of view. The air/water or air/skin interfaces strongly reflect polarized illumination light yielding a high Pol value that is not representative of the tissue. For example, Figures 4(b), 4(h), 4(j), 4(l), and 4(o) all show such air bubbles which appear as bright white spots in the image. A more viscous coupling agent is needed to ensure more reliable wetting of the tissue surface and coupling to the glass plate.