2.1.

Clinical Protocols

All clinical investigations were conducted according to the declaration of Helsinki principles. In the first study, five healthy individuals (male and female, 18 to 60 years of age, fair complexioned, skin types II to IV) without an active skin disease including eczema, utricaria, and history of atopic dermatitis and with positive history to rhus dermatitis volunteered to participate after giving informed consent. Rhus dermatitis was induced with an oleoresin extract.¹⁸ The oleoresin urushiol is the cause for the allergic reaction when sensitized individuals come in contact with the poison ivy plant and its relatives. Briefly, $10\mathrm{\mu }\mathrm{l}$ of 1:50 dilution of poison oak/poison ivy urushiol were applied with disposable micropipettes to $0.6\text{-}\mathrm{cm}\text{-}\mathrm{diam}$ disks of filter paper. The disks were left open to the air for $30\min$ and carefully applied with forceps to each test site. The loaded filter paper disks were then covered with occlusive tape (3M Blenderm, St. Paul, Minnesota) and left in place for $2\mathrm{h}$ . After the designated time duration, the disks and tape were removed, and the sites were washed with liquid soap and water. Before each evaluation the volunteers were left for $20\min$ to equilibrate to ambient conditions of the test facility. Clinical evaluation by a dermatologist and spectral imaging were performed on days 1, 3, and 7 after initiation of the inflammation. The clinical evaluation was based on a five-grade scale.

In the second study, five healthy individuals (male and female, 18 to 40 years of age, fair complexioned, skin types II to IV) with mild inflammatory acne participated after giving informed consent. The qualified volunteers had a history of experiencing at least two new inflammatory lesions per week (data obtained through self-report). No treatment usage or change in existing skincare regimens was allowed for the duration of the study. The lesions were examined at baseline and at days 1, 2, 3, 4, and 7.

Finally, a patient (male, 63 years, Fitzpatrick skin type II) with inflammatory skin lesions in the jaw and neck area diagnosed with herpes zoster by a trained dermatologist gave his informed consent to be imaged with spectral imaging.

2.2.

Instrumentation

We used a custom-made spectral imaging camera (MuSIS-HS, FORTH-Photonics, Athens, Greece) with 18 narrow-band filters (full width at half maximum of $10\mathrm{nm}$ ) in the wavelength range of 400 to $970\mathrm{nm}$ and three broadband filters in the red, green, and blue spectral regions. The red, green, and blue filtered images were combined to compose a color image. The imaging detector was an 8-bit CCD camera with $1024\times 768$ pixels. The field of view was $16\times 12\mathrm{cm}$ , resulting in a final image resolution of $0.156\mathrm{mm}∕\mathrm{pixel}$ for both the $x$ and $y$ axes. We used a filtered incandescent lamp equipped with a linear polarizer (model v600, Syris Scientific, Gray, Maine) as the light source. To avoid specular reflections we placed a linear polarizer in front of the camera lens with its plane orthogonal to the plane of polarization of the illumination light. We acquired a series of spectral images at a variety of wavelengths in the visible and NIR spectral range. Acquisition took approximately $5\mathrm{s}$ per image (a series of a minimum of six images required for chromophore calculations took $30\mathrm{s}$ ). Each series comprised a hyperspectral image stack, a 3-D cube with one spectral and two spatial dimensions. The images in the stack were aligned using a modified algorithm based on minimization of distance in the Fourier domain.¹⁹ This step was enough to align the images within approximately $0.5\mathrm{mm}$ , at which level the skin tissue can be assumed to be homogeneous for the purposes of macroimaging. Each pixel in the hyperspectral image stack represented a remittance spectrum. Spectral images acquired using narrow-band filters contain information about the concentrations of chromophores that absorb light at the corresponding spectral bands and can therefore be qualitatively described based on absorption and scattering parameters. Note, however, that this information is not quantitative, partly due to the extensive overlap of the chromophore absorption profiles. To extract quantitative information about the concentrations of chromophores, the reflectance images must first be converted to optical density (absorbance) maps. Then spectral analysis algorithms can be applied on a pixel-by-pixel basis for the calculation of apparent concentrations of the chromophores. To this end, absorption spectra were calculated for each pixel as the negative logarithm of the ratio of image of interest to the image at $850\mathrm{nm}$ , where water, melanin, and hemoglobin absorptions are negligible. This normalization to the image at $850\mathrm{nm}$ also helps in minimizing artifacts due to contours. After these calculations, the resulting series of images constitutes the hyperspectral absorption image stack, where each pixel corresponds to an absorbance spectrum.

2.3.

Erythema Maps

Erythema was evaluated based on the apparent concentrations of oxy-Hb. We calculated apparent concentrations of oxy-Hb and deoxy-Hb from the absorbance spectra after correction for melanin and scattering contributions according to a previously described algorithm. ^{6, 15, 20, 21} We refer to the calculated concentrations as “apparent” because they are based on light absorption curves and they are given in relative units. They are expected to be linearly related to “absolute” concentrations of hemoglobins in units of mass per volume of tissue. Due to the inhomogeneous nature of the tissue, such absolute units do not represent the reality and may be confusing. Using the absorbance hyperspectral image stack described in the previous section we calculated oxy-Hb, deoxy-Hb, melanin, and scattering values by applying the already mentioned algorithm to the absorption spectrum corresponding to each pixel. Briefly, melanin and scattering contribution were calculated as the slope and the intercept of a straight line fitted in the 630 to $730\text{-}\mathrm{nm}$ region. Then these contributions were subtracted from the original absorption spectra and the corrected spectra were fitted for the contributions of oxv-Hb and deoxy-Hb. Finally, we created distribution maps for each chromophore, in which the gray-scale intensity corresponded to the calculated chromophore values. Each chromophore map is a quantitative representation of the special distribution of the contribution to radiation absorption by the corresponding chromophore. Assuming that the contribution to light absorption is proportional to the chromophore concentration, the chromophore maps can be considered as quantitative representations of the chromophore concentrations. We considered the oxy-Hb map to represent a quantitative map of erythema involvement.⁵ To quantitate erythema we used the average gray-scale value over a region of interest defined by the size of the lesion to be measured. A neighboring area of uninvolved skin of similar size was chosen for reference. The apparent concentration of oxy-Hb corresponding to neighboring uninvolved skin sites was subtracted.

2.4.

Edema Maps

NIR radiation is absorbed by water with a local maximum at $970\mathrm{nm}$ . Water absorbance maps were calculated as the negative logarithm of the pixel intensity values of the image at $970\mathrm{nm}$ , where water has an absorption maximum, normalized to the corresponding pixel intensity values of the image at $850\mathrm{nm}$ (negligible NIR absorption by water). The normalization provides a baseline for the measurement of NIR absorption and minimizes artifacts due to contours. The amount of NIR absorption at $970\mathrm{nm}$ was assumed to be proportional to water content in the tissue. The absorbance map at $970\mathrm{nm}$ can be considered as a quantitative map of water accumulation in the tissue and therefore of edema. At these wavelength range light penetrates 1.2 to $1.6\mathrm{mm}$ into the skin,⁷ and therefore the information about cutaneous edema is limited to the tissue volume corresponding to this depth. Similarly to erythema, apparent water concentration values were averaged over a region of interest based on the lesion size. The apparent water concentration corresponding to neighboring uninvolved skin sites was subtracted.

3.1.

Allergic Dermatitis (Poison Ivy)

The volar forearms of five volunteers with positive history to rhus dermatitis were induced by urushiol for $2\mathrm{h}$ under occlusion, and the reaction was followed for 1 week. On day 1 only, two subjects developed a mild rash, while on day 3 all of the volunteers had developed a reaction including erythema and edema. Erythema was intense and limited to the area where urushiol came in contact with the skin. Edema was mild and present as perifollicular swellings within the area of induction. On day 7, the reactions persisted on all volunteers and they appeared to be more uniform than on day 3.

An example of a series of spectral images, along with the corresponding color image, acquired 3 days after initiation of Rhus dermatitis are shown in Fig. 1 . Hemoglobin absorption is maximum for the images at 560 and $580\mathrm{nm}$ and appears as dark areas in the corresponding images. At this wavelength range, light penetration is limited to a few hundred micrometers,^{4, 7} indicating that the observable features are limited to the superficial capillaries. Deeper vessels are evident at longer wavelengths, where both hemoglobins and melanin do not absorb strongly and light can travel deeper in the tissue up to a few-millimeters without being significantly attenuated. Although, in the range between 630 and $850\mathrm{nm}$ , melanin and hemoglobin absorption is minimal, the high concentration of hemoglobin at the inflamed areas provides the contrast to give rise to gray shadows. Water absorbs with a maximum at $970\mathrm{nm}$ and provides the contrast (appears dark) in the corresponding image due to local edema. Although we can describe the absorptions at each wavelength qualitatively, calculation of the chromophore values through spectral analysis is required to arrive at quantitative results.

Fig. 1

Images filtered at selected narrow bands demonstrate the optical properties of the tissue. An example of a color image and the corresponding series of narrow-band images acquired 1 week after initiation of rhus dermatitis. Hemoglobin absorption is maximum for the images at 560 and $580\mathrm{nm}$ . Deep vessels are evident at longer wavelengths. Water absorbs NIR radiation with a local maximum at $970\mathrm{nm}$ . The inflamed area looks dark at $970\mathrm{nm}$ due to edema.

In the chromophore maps shown in Fig. 2 , it is evident that the urushiol-induced inflammatory reaction manifested significantly elevated concentrations of oxy-Hb compared to surrounding skin. Deoxy-Hb concentration was elevated moderately. The increased water concentration relates to edema formation. As expected, the inflammation did not affect the local melanin concentration. In all the maps, the intensity scale is linear to the apparent concentration.

Fig. 2

Apparent concentration maps for the skin chromophores demonstrate the intensity and spatial distribution of erythema and edema. The maps shown here were calculated from the images shown in Fig. 1. The urushiol-induced inflammatory reaction demonstrates significantly elevated concentration of oxy-Hb compared to surrounding skin. Deoxy-Hb concentration is elevated moderately and the increased water concentration relates to edema formation. For comparison, the melanin map is also shown. Inflammation has not affected the local melanin concentration. In all the maps the scale is linear. The maximum gray-scale value (255) corresponds to apparent concentration of 10 for oxy-Hb, deoxy-Hb, and melanin, and 0.5 for water. The minimum gray-scale intensity (0) corresponds to zero concentration for all the maps.

We can extract quantitative information from the concentration maps using image analysis. For example, we can calculate the average value for the apparent erythema and edema after baseline subtraction. The erythema values calculated from the oxy-Hb concentration maps in this way were in good agreement with clinical evaluation by the physician as demonstrated by the correlation coefficient $R=0.979$ (data shown in Fig. 3 ).

Fig. 3

Erythema values obtained by spectral image analysis correlate with the clinical evaluation. Clinical evaluation of erythema was based on five-scale grading (0 to 4) with 0 being no detectable reaction and 4 being the most severe reaction. Erythema calculated from spectral analysis is expressed in oxy-hemoglobin values. Data are given as mean $\pm 1$ standard deviation. The correlation coefficient of the two methods is $R=0.979$ .

3.2.

Acne

Spectral imaging was used to evaluate the intensity and extent of erythema and edema in acne lesions. The lesions were monitored over a period of 1 week. Similarly to the poison ivy study, erythema maps were constructed based on the oxy-Hb apparent concentration values calculated from the spectral images.

Based on spectral imaging of the whole face we created erythema and edema maps (Fig. 4 ). Lesion count, an important parameter in acne clinical trials, is easier to perform on the erythema image rather than the traditional way on the RGB image, due to the improved image contrast. Furthermore, this procedure can be automated using appropriate image-thresholding limits. It is interesting to note that in the erythema map, the lips appear to be brighter than the surrounding skin, indicating the high degree of vascularization of the vermillion. In parallel to the erythema map where all the inflammatory lesions can be clearly seen, the edema map shows only the ones that are at the stage of papules. Interestingly, these are the lesions that are most painful due to activation of the pain receptors by the local buildup of pressure.

Fig. 4

Erythema and edema maps can be used to extract parameters relevant to acne. Here, the visible image and the corresponding pseudocolored erythema and edema maps are shown. The color bar indicates the relative intensity for the erythema and edema. A lesion count can be performed easily on the oxy-Hb (erythema) map. The intensity and extent of erythema are also evident. Note the high values of oxy-Hb on the lips, indicating higher perfusion compared to surrounding skin. The intensity and extent of edema can be documented on the water distribution map. Not all lesions present edema.

The gray-level intensity of the erythema map has a linear relationship with the apparent concentration of oxy-Hb. This means that regions of interest (ROIs) can be drawn around each lesion and the apparent amount of oxy-Hb can be quantified and monitored over time. In this procedure, we used a circular ROI centered at the geometrical center of the lesion. The same ROI is used on the same lesion for all the images in the subsequent time points. The difference of the lesional erythema intensity from the erythema level of neighboring uninvolved skin can be recorded and plotted over time [Fig. 5a ]. Another way of analysis is to threshold the erythema maps of the time sequence to the same limit and count the number of the non-zero-value pixels within the ROI of the lesion. This procedure gives an estimate of the lesion area. The percentage change in lesion area from baseline over time is shown in Fig. 5b. Both the erythema intensity and the lesion size are relevant parameters that can be used to monitor the evolution of an acne inflammatory lesion over time.

Fig. 5

Evolution of skin inflammation can be monitored over time. (a) The difference of the apparent oxy-Hb value of the lesion compared to the surrounding uninvolved skin can be used as a measure of the erythema intensity of the lesion, and (b) the average lesion size as percentage change from day 1. Data are shown as mean $\pm 1$ standard deviation for a total of 17 lesions of the five volunteers examined.

3.3.

Herpes Zoster

Spectral imaging was used to document the inflammatory reaction on a patient diagnosed with herpes zoster on the cheek and jaw area (Fig. 6 ). The oxy-Hb apparent concentration map provides better contrast than the conventional RGB image for erythema evaluation. Individual papules are easily detected as well as their agglomerations. Deoxy-Hb is slightly elevated in the inflamed area, though not to the levels of oxy-Hb, indicating the high demand for oxygen compared to surrounding tissue. Accumulation of extracellular fluid in the edematous areas can be demonstrated in the water apparent concentration map. These reactions are difficult or impossible to be identified in the visible image.

Fig. 6

Documentation of cutaneous inflammation induced by a viral infection (herpes zoster). (a) Visible image, (b) oxy-Hb apparent concentration map relates to erythema, (c) deoxy-Hb apparent concentration map, and (d) water apparent concentration map relates to edema. In all the maps, the gray-scale depends linearly to the apparent concentration of the chromophore. The maximum gray-scale value (255) corresponds to apparent concentration of 10 for oxy-Hb, 10 for deoxy-Hb, and 0.5 for water. The minimum gray-scale intensity (0) corresponds to zero concentration for all the maps.