Differentiating solar lentigines from post‐inflammatory hyperpigmentation with hyperspectral imaging: A pilot study

Abstract

Objective

Solar lentigines and post‐inflammatory hyperpigmentation (PIH) are common pigmentation disorders, and they are usually differentiated by a trained dermatologist. Accurate identification of the spot type is essential for providing patients with the most appropriate skin treatment. The aim of this study was to evaluate the capacity of hyperspectral imaging (HSI) in characterizing and discriminating solar lentigines from PIH, that is, to identify the colorimetric features and chromophore concentrations associated with each pigmentation disorder.

Methods

Acquisition of HSI was performed using a liquid‐crystal tuneable filter spectral scanning spectrometer. The recruited panel consisted of female Japanese subjects (n = 11) with one solar lentigo on both the left and right cheeks. Five of these subjects also had one PIH spot on both the left and right cheeks.

Results

Significant differences were found in the mean difference relative to the surrounding skin between PIH spots and solar lentigines for the following features (all p < 0.05): ΔOxyhaemoglobin (PIH spots (2.3 ± 1.1) × 10⁻² g·cm·L⁻¹ vs. solar lentigines (−0.1 ± 0.6) × 10⁻² g·cm·L⁻¹), ΔDeoxyhaemoglobin (PIH spots (6.3 ± 2.6) × 10⁻² g·cm·L⁻¹ vs. solar lentigines (0.9 ± 1.7) × 10⁻² g·cm·L⁻¹), Δa* values (PIH spots 5.3 ± 2.0 vs. solar lentigines 1.6 ± 1.3) and Δb* values (PIH spots −1.2 ± 0.7 vs. solar lentigines −0.1 ± 0.4). No significant difference was observed in ΔL* values and ΔMelanin content between the two hyperpigmentation types (all p > 0.05).

Conclusion

Compared with solar lentigines, PIH spots displayed higher haemoglobin content and Δa* values and lower Δb* values. To strengthen these preliminary findings, further clinical studies need to be conducted on a larger cohort, aiming at confirming the benefit of HSI in accurately differentiating these two skin pigmentation disorders to support clinical assessment.

This preliminary study attempted to characterize solar lentigines and post‐inflammatory hyperpigmentation (PIH) observed on Japanese women. Colorimetric features and chromophore concentrations were measured using hyperspectral imaging for each hyperpigmentation type. PIH spots displayed higher haemoglobin content and Δa* values and lower Δb* values compared to solar lentigines.

INTRODUCTION

Skin hyperpigmentation results from an excess in the production of melanin by melanocytes, causing an overload of melanin pigments within epidermal or dermal cells. It is a common skin concern, manifesting as benign dark spots or patches on the skin. Hyperpigmentation can lead to loss of self‐confidence and decreased quality of life [1, 2, 3]. Various types of hyperpigmentation can occur, such as solar/actinic lentigines, post‐inflammatory hyperpigmentation (PIH), melasma and ephelides (freckles) [4]. Solar lentigines and PIH are the most prevalent types of hyperpigmentation disorders [3, 5].

Solar lentigines are related to cumulative UV exposure, with incidence increasing with age and being more prevalent in lighter skin, especially in Asian people [6]. PIH results from excess melanin production and accumulation following inflammation, such as acne or injury, and is more common in darker skin tones [7, 8, 9, 10, 11, 12]. Among the African American population, 20% of dermatological diagnoses are related to PIH, whereas only 5.7% of dermatological problems in fair skin types I to III are linked to PIH [13]. For example, dark spots rank among the top three skin concerns for Asian women over 30 years old [3]. Management of hyperpigmentation involves either physical or topical therapy. Although PIH can resolve spontaneously, it takes several months or years to fully disappear. In contrast, solar lentigines are improved by physical treatment, such as laser therapy, or by topical treatment [14].

The cosmetics industry offers various products to address hyperpigmentation, using biologically active ingredients to reduce melanin production and visibly fade dark spots, possibly associated with camouflaging makeup. However, the efficacy of these ingredients depends on the hyperpigmentation type [15, 16, 17]. Therefore, accurately identifying the underlying hyperpigmentation type can improve the understanding of product benefits during clinical testing and ensure that consumers receive the appropriate treatment.

The identification of hyperpigmentation types in cosmetic clinical trials relies primarily on visual diagnosis by dermatologists, based on colour and morphological characteristics conditioned by the capabilities of the human eye, the dermatologist's experience and variations in lighting conditions.

Recent advancements in skin imaging, particularly hyperspectral imaging (HSI), have shown promise in improving skin evaluation. HSI has increasingly gained attention for skin evaluation in dermatology [18] and for medical diagnosis [19] due to the non‐invasive and non‐contact nature of the measurement, allowing evaluations free from any skin constriction or blood flow modification. In contrast to conventional RGB cameras, HSI captures the entire spectrum of light reflected from the skin, offering insights into various chromophores, including melanin and haemoglobin, and is also independent of the light source. In addition, HSIs can be designed to acquire full‐face images, providing extended spatial information compared with a spectroradiometer [20, 21]. Although HSI has been used to evaluate the efficacy of brightening products, the classification of erythema and the quantification of skin chromophores [22, 23, 24], to our knowledge, no studies have explored its potential in differentiating hyperpigmentation types.

The present study investigated the use of HSI as a supporting tool for diagnosing and differentiating PIH from solar lentigines instrumentally. A pilot study was conducted with a trained dermatologist who visually diagnosed solar lentigines and PIH lesions. Based on this clinical assessment, the specific spectral attributes and image features for each hyperpigmentation type were extracted. The attributes were the relative colorimetric ΔL*Δa*Δb* values and the relative chromophore concentrations of the spots (ΔMelanin, ΔOxyhaemoglobin [ΔOH] and ΔDeoxyhaemoglobin [ΔDOH]) compared with the natural skin complexion of the subject.

MATERIALS AND METHODS

Hyperspectral imager

An 80 cm diameter integrating sphere combined with a fibre‐coupled light source was used to ensure diffuse illumination during HSI acquisition [25]. The HSI system had a spectral‐scanning spectrometer based on a liquid‐crystal tuneable filter with a bandwidth of 32 nm and a spectral range between 420 and 730 nm. The imaging sensor of the HSI used a CMOS monochromatic camera with 2016 × 2016 active pixels, imaging a working field of view approximately 30 cm in diameter. Further details, including the complete design and technical performance of the instrument, are given in previous publications, such as [26], on the instrument design.

Test panel

A test panel was recruited and comprised 11 Japanese women aged 42–61 years (mean age ± standard deviation: 52.3 ± 6.0) of any face skin type (normal, oily, dry or combination) and Fitzpatrick skin phototypes ranging from II to IV. Exclusion criteria included sensitive skin, active acne in the cheek area, acne‐prone skin and irritated/damaged skin on the face. Subjects were also excluded if they used treatments such as whitening products (e.g. hydroquinone, kojic acid and tretinoin), acne treatments (e.g. adapalene, tazarotene and azelaic acid), vitamin C and its derivatives, vitamin A and its derivatives, local retinoid‐based treatment or peeling/exfoliation treatment up to 3 months prior to the study and retinoid oral treatment or professional beauty routines (e.g. laser, energy devices and hyaluronic acid/Botox injections) up to 6 months prior to the study.

All subjects showed at least one mild to moderate solar lentigo on both the right and left cheeks, and five of them also showed at least one PIH spot on both cheeks. Subjects were asked to avoid extreme UV exposure as much as possible during the 2 weeks before the study. Solar lentigines and PIH spots were identified by a dermatologist during preliminary clinical grading, and the position and type of each spot were marked. Pictures of the spots were taken and saved for the image analysis step. For each subject, one hyperspectral datacube was acquired on both left and right views after spot identification. This study was carried out under the Declaration of Helsinki and in the spirit of the general principles of Good Clinical Practice as published by the International Council for Harmonization of Technical Requirements for Pharmaceuticals for Human Use. All subjects received detailed information about the study, written in Japanese and provided a signed informed consent before enrolment.

Image analysis and region of interest selection

An interactive Python 3 framework (Python.org) for the image analysis sequence was written by our team. This analysis software computes colour and spectral parameters from the acquired camera data and allows manual selection of regions of interest (ROIs) within a given image. A 2‐pixel‐radius circle was defined for the manual selection of each spot ROI, which was manually selected on HSI acquisition as close as possible to the clinical assessment pictures identified by the dermatologist. To measure the baseline of the general skin condition, natural skin tone features (skin complexion) were also evaluated using 20‐pixel‐radius circles as skin tone ROIs and areas free from visible spots or PIH were selected on the cheeks. The skin tone is referred to as the reference hereafter. The CIE XYZ values for D65 illumination were calculated from the measured spectra, and L*a*b* values were calculated by using the transformation equation from the XYZ values [26]. As the working range (420–730 nm) of this instrument does not cover the full range of the XYZ definitions (380–800 nm), it introduces a spectrum‐dependent systematic error into the determination of absolute L*a*b* values from the measured spectra. For this reason, and to reduce the dependence of our results on the absolute calibration of the instrument, only relative L*a*b* differences between the spot or PIH areas and the skin reference were pair‐compared.

The concentration (in g·cm·L⁻¹) of each chromophore (melanin, OH and DOH) in each pixel was determined by fitting an absorption spectrum model to measured spectra $X$ following the method published in [27]. In our implementation, the model optical density is

$$\begin{array}{c}O{D}_{\mathit{\text{predicted}}}\left(\lambda \right)=a_{\mathit{OH}}*e_{\mathit{OH}}\left(\lambda \right)+a_{\mathit{DOH}}*e_{\mathit{DOH}}\left(\lambda \right)\hfill \\ +a_{\mathit{Mel}}*e_{\mathit{Mel}}\left(\lambda \right)+a_{\mathit{\text{Offset}}}\hfill \end{array}$$

where $\mathit{OD}$ is the optical density ($\mathit{OD}=-1{\log }_{10}\left(\mathrm{X}\right)$). $a$ and $e\left(\mathrm{\lambda }\right)$ are the extinction coefficients for OH, DOH, melanin and non‐chromophore sources, indicated by the subscripts OH, DOH, Mel and Offset, respectively. The extinction coefficients of haemoglobin used in this model were the experimentally determined molar extinction coefficients for haemoglobin in water ([28], compiled from [29, 30]) and those of melanin were similarly the measured extinction coefficients for melanin in water ([31], compiled from [32, 33]). The values of $a$ that provided the best fit to the observed spectrum were determined using the Trust Region Reflective algorithm implemented in the SciPy Optimize package curve_fit method [34].

Because of the simplicity of this optical model and the fact that this measurement setup uses a diffuse light source, this method can only measure the convolution of concentration (in g·cm·L⁻¹) and average pathlength. Therefore, some care must be taken in comparing values from skin regions with large differences in thickness, such as around the nose. In the current work, the spots and PIH areas studied were mainly in regions of the cheek where the skin thickness variation is likely low. Since the curve fitting was only performed in the measured spectral range, it did not suffer from the same problems as the determination of absolute L*a*b* values. Nonetheless, a focus was made on the relative difference in chromophore concentration between the target and reference areas to reduce any reliance on the absolute calibration of the chromophore reconstruction (in g·cm·L⁻¹).

Statistical analysis

Principal component analysis (PCA) was conducted on the relative values dataset using the Scikit‐learn library [35]. Normalization and mean centring were performed on the data with the preprocessing module of Scikit‐learn, and the PCA was run with the decomposition module of Scikit‐learn.

The differences between the targeted solar lentigo values and the reference measurements and between the targeted PIH values and reference measurements were calculated as follows: for each evaluated feature, each volunteer and each cheek (the slash ‘/’ in the subscript means either Solar Lentigo or PIH):

$$∆L^{*}=L_{\mathit{\text{Solar lentigo}}/\mathit{PIH}}^{*}-L_{\mathit{\text{Reference}}}^{*}$$

$$∆a^{*}=a_{\mathit{\text{Solar lentigo}}/\mathit{PIH}}^{*}-a_{\mathit{\text{Reference}}}^{*}$$

$$∆b^{*}=b_{\mathit{\text{Solar lentigo}}/\mathit{PIH}}^{*}-b_{\mathit{\text{Reference}}}^{*}$$

$$∆\mathit{\text{Melanin}}={\mathit{\text{Melanin}}}_{\mathit{\text{Solar lentigo}}/\mathit{PIH}}-{\mathit{\text{Melanin}}}_{\mathit{\text{Reference}}}$$

$$∆\mathit{OH}={\mathit{OH}}_{\mathit{\text{Solar lentigo}}/\mathit{PIH}}-{\mathit{OH}}_{\mathit{\text{Reference}}}$$

$$∆\mathit{DOH}={\mathit{DOH}}_{\mathit{\text{Solar lentigo}}/\mathit{PIH}}-{\mathit{DOH}}_{\mathit{\text{Reference}}}$$

Calculating the relative values between the hyperpigmentation spot and the reference eliminated the incidence of the base skin colour of each subject and reduced the data variability. Once the relative differences for each subject were calculated for all characterized features, the statistical analysis was run. Given the disparity in sample sizes between the PIH spots dataset and the solar lentigines dataset, an unpaired t‐test was conducted using the SciPy library [34] to assess the probability of the results under the null hypothesis in each group comparison. The unpaired t‐tests were all two‐sided with a significance level of α = 0.05.

RESULTS

HSI data analysis was conducted to compare the solar lentigines and PIH spots identified by the ΔL*Δa*Δb* colorimetric measurements and chromophore evaluation. Figure 1 shows the boxplots of the ΔL*Δa*Δb* colorimetric relative differences between the spots and the skin reference, and the relative differences for each chromophore concentration of ΔMelanin, ΔOH and ΔDOH between the spots and the skin reference. The dataset consisted of 22 datapoints for solar lentigines (11 subjects with one spot on each cheek) and 10 datapoints for PIH spots (5 subjects with one PIH spot on each cheek). Each relative difference was calculated with the base skin tone of the specific volunteer as the reference.

FIGURE 1.

Boxplot plots for ΔL*, Δa*, Δb*, ΔMelanin concentration, ΔOH concentration and ΔDOH concentration for solar lentigines and PIH spots. The boxes extend from the first quartile (Q1) to the third quartile (Q3), and this distance represents the interquartile range (IQR). The solid line corresponds to the median (Q2). The whiskers extend from Q1–1.5 × IQR to Q3 + 1.5 × IQR. The dots represent the individual datapoints and the filled diamonds represent outliers.

Mean relative differences are shown in Figure 2 for all characterized attributes. Numerical mean values and the associated 95% confidence interval are listed in Table 1.

FIGURE 2.

Mean differences of evaluated attributes for solar lentigines and PIH spots relative to the reference for the colorimetric results (left) and chromophore concentration results (right). The mean values for each attribute are shown in the bar plot, and the error bars represent the 95% confidence interval. The green asterisks (*) and horizontal lines indicate the statistically significant differences.

TABLE 1.

Mean colorimetric relative results and mean chromophore relative concentrations (mean ± 95% confidence interval) for solar lentigines and PIH.

	ΔL*	Δa*	Δb*	ΔMelanin (g cm L−1)	ΔOH (g cm L−1)	ΔDOH (g cm L−1)
Solar lentigines	−3.8 ± 1.4	1.6 ± 1.3	−0.1 ± 0.4	(1.6 ± 0.6) × 10−2	(−0.1 ± 0.6) × 10−2	(0.9 ± 1.7) × 10−2
PIH	−4.7 ± 1.6	5.3 ± 2.0	−1.2 ± 0.7	(1.4 ± 0.5) × 10−2	(2.3 ± 1.1) × 10−2	(6.3 ± 2.6) × 10−2

The Δa* values were significantly higher for PIH spots than for solar lentigines (p = 5 × 10⁻³ < 0.05) and the Δb* values were significantly lower (p = 7 × 10⁻³ < 0.05). No statistical differences were observed for the ΔL* measurements between solar lentigines and PIH spots (p = 0.5 > 0.05). Regarding chromophores, haemoglobin concentrations were significantly higher for PIH spots than for solar lentigines for both ΔOH (p = 2 × 10⁻³ < 0.05) and ΔDOH (p = 5 × 10⁻⁴ < 0.05). Differences in ΔMelanin between PIH spots and solar lentigines were found to be non‐significant (p = 0.7 > 0.05).

Figure 3 shows the PCA results for the relative values of all variables. Compared with solar lentigo datapoints, PIH spot datapoints were mostly grouped under higher Δa*, ΔDOH and ΔOH values. There was a positive correlation between the Δa* values and haemoglobin concentrations (ΔDOH and ΔOH) and a negative correlation between ΔL* values and ΔMelanin.

FIGURE 3.

PCA biplot for colorimetric variables and chromophore variables. Datapoints are grouped and coloured by ROI (hyperpigmentation type). Green arrows represent the variables and their direction. The explained percentage variance of each principal component is displayed in parentheses.

DISCUSSION

In this pilot study, solar lentigines and PIH colorimetric features and chromophore concentrations were instrumentally measured using HSI. Compared to solar lentigines, the higher Δa* values and higher haemoglobin concentrations (ΔDOH and ΔOH) observed for PIH spots could be attributed to an increased blood flow in the PIH region, which may be due to the inflammatory response of the human body [36]. Inflammation is usually followed by redness caused by arteriolar dilation (also called vasodilation), which is responsible for the increased blood flow in inflamed regions during the healing process. The present pilot study was not meant to explain the specific biological mechanisms initiating the inflammatory aspect of PIH. However, the chromophore and colorimetric HSI measurements could reflect underlying biochemical reactions related to PIH, which can be caused by various possible factors, such as infections, burns, bug bites or rash [8]. PIH is initiated by different inflammatory mediators (cytokines, chemokines, reactive oxygen species, epidermal growth factor), which can stimulate melanocyte growth, melanin production and melanin transfer to keratinocytes [7]. The observed simultaneous increase in ΔDOH, ΔOH (haemoglobin) and the Δa* (redness) likely reflect the interaction between the underlying blood vessels and epidermal or dermal skin cells during the extended healing process.

The Δb* values were found to be lower for PIH spots than those of solar lentigines and the reference skin tone values. Dark purple‐grey spots, which are occasionally associated with PIH [7], may possibly account for these lower Δb* values. Because PIH spots mostly appear as red or purple‐grey dermal discolouration, the Δa* and Δb* results found here seem to match the visual observations in most patients. The correlation between Δa* values and haemoglobin (ΔDOH and ΔOH), and between ΔL* and ΔMelanin likely relates to the density of the chromophores in the production of skin colour and, hence, the L*a*b* values.

The relative values between the targeted spot and the natural skin tone (skin complexion) were used for the data analysis instead of the absolute values. This approach minimized the influence of varying skin tones (e.g. phototypes) across subjects.

The working spectral range of the liquid‐crystal tuneable filter limits the determination of absolute L*a*b* values from measured spectra. The use of relative differences reduces the dependence of the results on the absolute instrument calibration. Further instrument optimization, such as the inclusion of a reference colour card in all captured datacubes, may enable the reliable use of absolute values. The reference baseline is currently defined by the operator selecting reference skin complexion areas using a 20‐pixel‐radius circle ROI where the skin was visually close to the natural skin tone of the subject. However, the ROI could be chosen in areas close to the spot. Therefore, the evaluated features depend on the ROI selection and the reference choice. In further studies, the investigation of the ROI selection and its impact on the results may bring valuable insights.

Another factor that may affect the variability of the measurements is the position of the specific spot targeted by the dermatologist. Spots on the edge of the cheek should be avoided because these areas are prone to undesirable light scattering and geometrical distortion caused by the curvature of the face and cheek. Therefore, these spots may not have been fully visible when the 2D HSI of the side views was acquired and may have caused potential interference in the measurements. Spots closer to the centre of the face should be selected in future studies.

Colorimetry provided satisfactory discrimination between solar lentigines and PIH spots according to the Δa* values, but other skin conditions are also associated with high Δa* values, such as erythema [37]. For instance, post‐inflammatory erythema shares similar features with PIH and is also characterized by red discolouration from residual acne marks. However, unlike PIH, post‐inflammatory erythema does not involve melanin [38], which is a key difference that HSI can detect. Therefore, HSI supplements colorimetric evaluation and gives additional information through chromophore characterization, allowing improved discrimination of various skin disorders. Because clinical evaluation often involves subjective judgement of spots by dermatologists, HSI could support their decision‐making during the assessment. The HSI system used in this study allows non‐contact measurement, avoiding skin constriction or modification of blood flow, which commonly occurs during contact measurements.

The present exploratory study has two main limitations. First, the small number of participants restricts the generalizability of our conclusions. Future investigations need to be conducted with a larger sample size to confirm these preliminary findings. Second, since the study was only conducted on Japanese women with phototypes II to IV, further research should include other skin types with darker complexions, where PIH is more frequent and intense [10, 11, 12].

Despite these limitations, this pilot study shows the potential of HSI to objectively differentiate these two types of hyperpigmentation, enabling more accurate diagnosis and targeted treatment strategies, including medical interventions and appropriate cosmetic regimens (sunscreens, whitening agents and camouflage makeup). The characterization and identification of other pigmentation disorders using HSI, such as erythema, melasma or freckles, could provide valuable insights into the underlying physiological mechanisms behind these skin conditions.

CONCLUSION

This pilot study demonstrated the potential use of HSI to differentiate solar lentigines and PIH, providing an objective measurement and discriminative assessment of the skin conditions. The HSI results revealed that PIH spots showed higher ΔOH and ΔDOH contents, higher Δa* values and lower Δb* values than solar lentigines. HSI could be an effective tool to support accurate clinical diagnosis of skin disorders by dermatologists, which is essential for identifying the most appropriate treatment for patients.

AUTHOR CONTRIBUTIONS

VE was responsible for the ROIs selection and data extraction from the study files, the results visualization, the statistical analyses, as well as the writing of the original draft of this manuscript. CB contributed to the setup of the HSI system, collection of the HSI data, the instrumental study design and the programming of the interactive handling software. AI and ME were responsible for the study design of the clinical evaluation, the recruiting of the test panel, the spots identification with the dermatologist and the collection of HSI data with models. AN helped in the study conceptualization, acquisition of funding and the management and coordination of this research study. All authors contributed to the review of this paper and approved the submitted version.

CONFLICT OF INTEREST STATEMENT

All authors are employees and/or shareholders of L'Oréal.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.