Line-Field Confocal Optical Coherence Tomography of Plaque Psoriasis Under IL-17 Inhibitor Therapy: Artificial Intelligence-Supported Analysis

Abstract

To date, therapeutic responses in plaque psoriasis are evaluated with clinical scores. No objective examination has been established. A recently developed non-invasive imaging tool, line-field confocal optical coherence tomography (LC-OCT), enables the in vivo live imaging of skin changes in psoriasis under therapy. The aim of this study was to measure therapeutic response clinically and with LC-OCT, comparing the subjectively scored epidermal changes with an AI-supported analysis. This prospective, observational study included 12 patients with psoriasis starting a systemic treatment with IL-17 inhibitors (secukinumab, ixekizumab, and bimekizumab). LC-OCT and clinical assessment with a local psoriasis and severity index of the study plaque and a control area were performed before the initiation of therapy as well as after 4 and 12 weeks of treatment. A manual and AI-supported measurement of the thickness of epidermis, stratum corneum, and undulation of the dermo-epidermal junction was carried out. Acanthosis and hyperkeratosis showed a significant reduction under treatment. AI-supported calculations were compared to subjective measurements showing good reliability with high correlation. AI-supported analysis of vascular changes may serve as a prognostic and therapeutic response marker in the future.

1. Introduction

Psoriasis vulgaris is a systemic, inflammatory disease of high chronicity; its most frequent subtype, plaque psoriasis (PP), clinically presents with sharply demarcated, erythematous, scaly plaques [1,2]. Its predilection sites involve the scalp, auditory canal, extensor sites of the extremities, and the lumbosacral region. The degree of severity can be assessed with scores like the body surface area (BSA) [1,2] or the psoriasis area and severity index (PASI) [1,2,3]. Psoriasis can cause serious deterioration in quality of life in affected patients [1]. This can be evaluated, for example, with the dermatology life quality index (DLQI) and influence treatment decisions [2]. The monitoring of therapeutic responses in plaque psoriasis is, to date, mainly realized using clinical scores, such as the BSA or PASI [1,2,3]; both are subject to an inter-rater variability to a certain extent [4,5]. None of the existing scores can grade the severity of psoriasis or its therapeutic response objectively.

Different non-invasive imaging tools have been developed and tested to examine therapeutic response. In dermoscopy, the psoriasis’ typical pattern comprises dotted or globular vessels, distributed uniformly in the lesion, background erythema, and white scales [6,7,8,9,10]. Effective treatment can change the dermoscopic pattern; the appearance of hemorrhagic dots and globules accounts as a marker for good treatment response [6,9,11,12] but does not allow for objective quantification. Non-invasive imaging by high-frequency ultrasound (HFUS) can be used to monitor therapeutic response in PP [9,11,13] but is not capable of differentiating the different layers of the epidermis or single cells [11]. The therapeutic response of PP in vivo can be examined by reflectance confocal microscopy (RCM), but the removal of scales before examination impedes the measurement of the stratum corneum [14]. Furthermore, RCM allows for the visualization of up to a depth of only 250–300 µm; in PP, clarity starts to decrease at a depth of 180 µm [14]. Optical coherence tomography (OCT) was originally developed for examinations of the retina [15]. With OCT, the evaluation of the skin to a depth of 1–2 mm is possible [16]. It can be used for treatment monitoring in psoriasis, showing increased epidermal thickness (ET) as well as dilated vessels in PP and its reduction and normalization under therapy [17,18]. The visualization of scaling, parakeratosis, acanthosis, and dilated blood vessels is seemingly possible [17], but it does not allow for the detection of single cells [16].

Line-field confocal optical coherence tomography (LC-OCT) is a recently developed, non-invasive imaging tool that can be used to obtain live, in vivo insights into skin changes. It enables the measurement of the epidermis and dermis up to a depth of 500 µm with a very high resolution of 1–2 µm. Due to single cell depiction, histologic skin changes in psoriasis, typically detectable by light microscopy, like hyperkeratosis, acanthosis, elongated rete ridges, and dilated vessels of the papillae can be distinguished live, three-dimensionally, and in vivo by LC-OCT [19]; additionally, parakeratosis and bright inflammatory cells can be differentiated [20,21]. However, despite its high resolution, LC-OCT is limited by a relatively low penetration depth, making it impossible for imaging deeper dermal skin layers. This limitation is compounded in conditions like severe plaque psoriasis, where hyperkeratosis and acanthosis can obscure deeper structures. The limited penetration depth contrasts with its high resolution and necessitates the application of immersion oil to optimize image quality. Enhancing this depth without compromising resolution remains a challenge for future developments in LC-OCT technology.

To date, there are few data regarding the therapeutic monitoring of PP under systemic treatment with LC-OCT and artificial intelligence (AI)-enhanced evaluation of epidermal changes under therapy [22]. No published study has systematically compared subjectively and AI-supported assessments of epidermal changes during therapy to evaluate the accuracy of the AI-generated results. There has also not been a comparison of the skin changes in psoriatic and healthy skin of the same patients. With the segmentation of vessels in the LC-OCT scans of PP, an AI-supported analysis could be trained for the future, analyzing and measuring their density, which could be implemented as a therapeutic response marker.

2. Materials and Methods

Non-invasive live imaging was conducted with the deepLive^TM LC-OCT device (DAMAE Medical, Paris, France), which uses an 800 nm wavelength laser. It can generate vertical (1.2 mm × 0.4 mm) and horizontal pictures (1.2 mm × 0.5 mm) as well as 3D stacks (1.2 × 0.5 × 0.5 mm³), simultaneously capturing dermoscopic pictures. The application of immersion oil (paraffin oil) is necessary [23,24,25]. Because of the high resolution, it generates visuals similar to histologic slides. A live video sequence can be captured, which enables the visualization of blood flow. A second probe, containing a videodermoscope, is connected to the device and can be used separately. It allows for the colocalization of the LC-OCT pictures within the previously taken dermoscopic images.

For this prospective observational study, 12 patients with clinically or histologically confirmed PP, newly receiving treatment with an IL-17 inhibitor, were examined [26]. We focused on one group of therapeutic agents in order to minimize therapeutic outcome variations between different biologic subgroups and because IL-17 inhibitors have emerged as a highly effective class of biologics for treating moderate-to-severe plaque psoriasis. These drugs have demonstrated superior efficacy compared to many other treatment options, with some studies showing they can achieve “clear skin” for a significant proportion of patients [26]. The robust clinical response and rapid onset of action observed with anti-IL-17 drugs thus presumably also suggested a rapid improvement in psoriasis changes in LC-OCT, which is why this subgroup of drugs was selected for this study. Aforehand, approval from the ethics committee was obtained. Treatment decisions, screening before the therapy, and monitoring during therapy were carried out at our outpatient clinic, and were not influenced by participation in the study. Patients were included after obtaining their written, informed consent. They were eligible if the study lesion was not treated with topical corticosteroids 2 weeks prior to study initiation and if the biologic treatment was newly administered. Exclusion criteria were the regular use of topical corticosteroids in the last 2 weeks prior to inclusion on the chosen psoriatic plaque, the continuation of another, already prescribed systemic treatment, and a lesional psoriasis severity index (LPSI) lower than 5 in all psoriasis plaques.

A study plaque (SP) and control area (CoA) were chosen before the initiation of therapy. The SP’s localization was not restricted; the CoA had to be clinically healthy skin and, if possible, located in a body region similar to that of the SP. Therapeutic response in the SP was evaluated by changes as observed by LC-OCT and, clinically, by determining the LPSI. The LPSI was designed in accordance with the LPSI already established by Errichetti [27] as a five-point scale (scaling (0–4), erythema (0–4), and infiltration (0–4)), ranging from 0 to the highest score of 12. The overall therapeutic response was assessed using the PASI score and DLQI. The probability of psoriatic arthritis was screened for with the German psoriasis arthritis diagnostic questionnaire (GEPARD). Patients underwent the first screening before starting the new treatment (V0), as well as 4 weeks (V1) and 12 weeks (V2) after the initiation of therapy. Use of topical corticosteroids on the SP was omitted during this period in accordance with the patients. The LC-OCT scanning of the SP and CoA, dermoscopy, clinical photography, and the assessment of the LPSI, PASI, DLQI, and GEPARD were conducted at all study visits by our investigator team. For one patient, the DLQI and GEPARD were missing for V2. LC-OCT scanning was performed by capturing a 3D stack at three different areas in the SP and CoA after the application of paraffin oil. The best scan of each lesion and time point was chosen for analysis, including a manual measurement of hyperkeratosis/stratum corneum (SC) and ET (see Figure 1a–c). If the dermo-epidermal junction (DEJ) could not be detected by LC-OCT because of excessive acanthosis or hyperkeratosis, the highest measurable value was used for statistical analysis (Figure 1a).

The chosen 3D stacks were analyzed with artificial intelligence models developed by Damae Medical (Paris, France). One model was trained for skin layer segmentation, the other model for follicle detection. The segmentation of the skin layers and identification of stratum corneum, epidermis, and dermis were realized through the application of a 2D U-Net model to the LC-OCT stacks. The boundaries of the different skin layers were defined using the interface between two segmented areas; for example, the detection of the DEJ was realized using the pixel boundary between the epidermis and dermis. The AI model was trained on a big and diverse dataset, including acquisitions of healthy skin from different body parts and different extent in the undulation of the DEJ, actinic keratoses, and inflammatory dermatoses like atopic dermatitis and psoriasis. To ensure the highest level of accuracy, all segmentations underwent a meticulous validation process. This involved a comprehensive manual review conducted by a skilled operator, who carefully examined and verified each segmentation output by the model. We used the same U-Net model as described in detail elsewhere [22,28,29,30,31,32,33,34,35,36], especially that by Chauvel-Picard et al. A few studies [22,28,29,30,31,32,33,34,35,36] have already shown the applicability of this U-Net AI model in the evaluation of healthy skin, actinic keratoses, atopic dermatitis, and also in the evaluation of plaque psoriasis.

The segmentation of the skin layers and detection of hair follicles was visually supervised by an expert team to ensure validated results. All 3D stacks with a variation of more than 30% between manually and AI model-generated measurements were reviewed by an expert team to detect possible reasons for differences or wrong measurements. The number of pixels between the blue and red line (see Figure 2 and Figure 3) divided by the number of pixels corresponding to the region of interest was used for calculating the SC thickness (SCT). The result in microns was obtained with the pixel–micron equivalence. The same procedure was used for the ET, calculating the number of pixels between the red and green line and summing up the SCT. DEJ undulation was quantified as the ratio of the DEJ contour length to the straight-line length of the analyzed area in percent. In acquisitions without visible DEJ, because of excessive acanthosis, the AI model placed the green line at the bottom of the stack. For each parameter, the area detected as a follicle by the follicle detection model was excluded from the calculation in order to prevent a bias of the skin layer measurements.

One 3D stack had to be excluded from the AI-supported calculation because of a severe mismeasurement by the AI model caused by scales and dust in between the skin surface and probe, which led to a misrecognition of great parts of the skin surface. Subjectively scored calculations in this 3D stack were possible.

For the statistical analysis, nonparametric tests were used due to the non-normal distribution of the data. Clinical scores (PASI, DLQI, GEPARD, and LPSI) as well as measurements (SCT, ET, and DEJ undulation) were compared between all visits with the Friedman test. Further, differences between the SP and CoA were calculated and compared between all visits. The distribution of all measurements at each visit is depicted with boxplots. A statistical comparison between the AI-supported and manual measurements was carried out together with specialists of the DAMAE Medical team. Pearson’s and Spearmen’s coefficients were computed for the correlation between AI-supported and subjective measurements of ET and SCT. Additionally, the percentage deviation between both methods was provided. A linear regression analysis was performed to judge the agreement of both methods, where an intercept of 0 and a slope of 1 indicate perfect accordance. A p-value less than 0.05 was considered significant. Statistical analysis was performed using R version 4.3.1.

3. Results

In total, 12 patients (6 male and 6 female) were included in this study: 4 patients received secukinumab (IL-17A); 7 patients, ixekizumab (IL-17A); and 1 patient, bimekizumab (IL-17A and -F inhibitor) as biologic treatment [1]. The patients’ age ranged from 25 to 84 years (mean age: 52.9 years; sd: 16.5 years). One patient (bimekizumab) had to interrupt therapy due to candida esophagitis after the completion of the study. There were no serious adverse effects observed among the other patients. The location of the SP was in the lower extremities of six patients and upper extremities including the scapular region in the other six patients. The mean time between the first study visit (V0) and second study visit (V1) was 5.4 (sd: 1.6) weeks, and that between V1 and the third study visit (V2) was 8.1 (sd: 3.0) weeks.

3.1. Clinical Response

3.1.1. PASI, DLQI, and GEPARD

For the exact values of the PASI, DLQI, and GEPARD, see

Table 1. Psoriasis area and severity index (PASI), dermatology life quality index (DLQI), and local psoriasis severity index (LPSI) median values, as well as the median values of subjective measurements of the epidermal thickness (ET) and stratum corneum thickness (SCT), the respective p-values in the study plaque (SP) and control area (CoA), and their differences in the course of treatment; µm = micrometer, IQR = interquartile range, 25–75% percentile.

	V0 Median (IQR)	V1 Median (IQR)	V2 Median (IQR)	p-Value (Friedman Test)
PASI	18.3 (8.4–30.0)	8.6 (2.0–15.0)	3.8 (1.7–7.6)	p < 0.001
DLQI	13.5 (7.8–15.3)	3.0 (1.8–6.3)	1.0 (0.0–4.0)	p < 0.001
GEPARD	6.0 (3.8–9.3)	7.0 (4.8–9.3)	5.0 (4.0–10.0)	p = 0.519
LPSI (SP)	6.0 (6.0–9.0)	3.0 (2.8–5.0)	1.0 (1.0–2.3)	p < 0.001
ET (SP)	290 µm (249–328 µm)	213 µm (139–256 µm)	121 µm (90–168 µm)	p < 0.001
SCT (SP)	84 µm (56–94 µm)	41 µm (29–63 µm)	22 µm (13–31 µm)	p = 0.001
LPSI (CoA)	0	0	0
ET (CoA)	87 µm (73–90 µm)	76 µm (66–87 µm)	77 µm (72–92 µm)	p = 0.563
SCT (CoA)	9 µm (9–13 µm)	13 µm (11–14 µm)	13 µm (10–16 µm)	p = 0.052
Difference ET (SP) and ET (CoA)	202 µm (161–243 µm)	133 µm (76–173 µm)	46 µm (23–57 µm)	p< 0.001
Difference SCT (SP) and SCT (CoA)	72 µm (42–84 µm)	29 µm (16–48 µm)	8 µm (2–20 µm)	p < 0.001

. The reduction in the PASI under therapy was clinically significant when comparing all study visits (p < 0.001) (Figure 4a).

The improvement of the DLQI in the course of treatment was also statistically significant (p < 0.001) (see Figure 4b).

The GEPARD remained relatively equal under therapy. There was no difference during the study period.

3.1.2. LPSI and Manually Measured Epidermal Thicknesses

The LPSI in the SP showed a significant reduction in the course of treatment (p < 0.001) (see Figure 5) (for exact values, see Table 1).

The presence of hyperkeratosis and acanthosis was observed in all patients before treatment (12/12), acanthosis was present at V1 in 9 patients (9/12), and hyperkeratosis in 9 of the 12 patients. At V2, acanthosis was still present in three patients (3/12), and hyperkeratosis, in three patients (3/12). Before treatment, the exact thickness of acanthosis was measurable in 4 of 12 patients (in the other patients, acanthosis exceeded the penetration depth of the LC-OCT device); at V1 and V2, acanthosis was measurable in 11 of 12 patients. The reduction in acanthosis during treatment showed statistical significance (p < 0.001) (see Figure 6a). For one patient, acanthosis continued to exceed the device’s penetration depth, thus remaining non-measurable.

Exact hyperkeratosis was measurable at all times in all patients; its reduction was statistically significant (p = 0.001) (for exact values, see Table 1).

In the CoA, the LPSI was 0 for all patients at all study visits, and there was no sign for hyperkeratosis or acanthosis (for exact values, see Table 1). No statistically significant difference during the study period (see Figure 6b) could be observed.

To assess whether the thickness of the SP converged toward that of the CoA, the difference between the SP and CoA was calculated at each time point and compared between visits. For ET as well as SCT measurements, this difference significantly decreased over the course of treatment (ET: p < 0.001; SCT p < 0.001; see Table 1 for exact values).

3.2. Artificial Intelligence (AI)-Supported Analysis

3.2.1. AI-Supported Measurement of Epidermal Thickness and Stratum Corneum Thickness

The reduction in the ET in the course of therapy measured by the AI-model of the SP was significant (p = 0.006), as well as that of the SCT (p < 0.001); for the exact values of the AI-supported measurements, see

Table 2. Artificial intelligence (AI)-supported measurements of epidermal thickness (ET) and stratum corneum thickness (SCT), respective p-values in the study plaque (SP) and control area (CoA), and differences in the ET and SCT comparing the SP and CoA. µm = micrometer; IQR = interquartile range, 25–75% percentile; DEJ = dermo-epidermal junction.

	V0 Median (IQR)	V1 Median (IQR)	V2 Median (IQR)	p-Value (Friedman Test)
ET (SP)	254 µm (201–272 µm)	200 µm (139–243 µm)	124 µm (92–166 µm)	p = 0.006
SCT (SP)	62 µm (40–71 µm)	36 µm (27–47 µm)	25 µm (20–39 µm)	p < 0.001
ET (CoA)	70 µm (61–81 µm)	71 µm (67–80 µm)	73 µm (67–82 µm)	p = 0.339
SCT (CoA)	14 µm (12–19 µm)	16 µm (14–19 µm)	18 µm (15–21 µm)	p = 0.105
Percentage difference of ET (SP, AI vs. subjective scoring)	−17.6% (−23.2–−13.0%)	−15.4% (−17.6–−5.0%)	−0.35% (−13.0–−9.6%)	Pearson coefficient = 0.97, Spearmen = 0.91
Percentage difference of SCT (SP, AI vs. subjective scoring)	−23.6% (−42.3–−6.8%)	−20.8% (−29.7–−6.9%)	37.4% (6.5–−84.1%)	Pearson coefficient = 0.84 Spearmen = 0.81
Difference in ET (SP) and ET (CoA)	181 µm (141–194 µm)	117 µm (77–150 µm)	56 µm (25–77 µm)	p = 0.004
Difference SCT (SP) and SCT (CoA)	39 µm (24–59 µm)	18 µm (11–30 µm)	9 µm (3–16 µm)	p = 0.001
DEJ undulation (SP)	75.9% (52.7–134.6%)	48.8% (36.4–68.0%)	27.0 (14.0–79.0%)	p = 0.148
Difference in DEJ undulation (SP and CoA)	65.5% (20.0–126.1%)	46.4% (21.9–57.1%)	19.3% (3.2–49.6%)	p = 0.103

. The difference in the thicknesses measured by the AI-model between the SP and CoA was calculated at each time point and compared between visits to assess for the convergence of the SP toward the CoA. This difference significantly decreased over the course of treatment for ET as well as SCT (ET: p = 0.004; SCT: p = 0.001) (see Figure 7 for illustration). The AI-supported measurements of the ET and SCT in the SP and CoA showed a high correlation with subjectively scored measurements (see Figure 8). The Pearson coefficient for the ET (SP and CL) was 0.97, and the Spearmen coefficient was 0.91. Regarding the SCT (SP and CoA), the Pearson coefficient was 0.84, and the Spearman coefficient was 0.81. For a better comparison of the variation between the subjective and AI model-calculated thicknesses, we also applied a linear regression. When both methods had perfect accordance, the intercept was 0 and the slope was 1. For the ET, the intercept was 14.3 (95% confidence interval (CI): 5.7–23.0), and the coefficient was 0.79 (95% CI: 0.73–0.84). The median percentage difference in the AI-calculated ET from the subjectively scored value was 13.4%.

The intercept for the SCT, respectively, was 12.9 (95% CI: 8.4–17.4), with a coefficient of 0.53 (95% CI: 0.45–0.61). The median percentage difference between the subjective and AI model measurements was 30.8%.

3.2.2. Dermo-Epidermal Junction (DEJ) Undulation

The AI model quantification of the undulation of the DEJ was calculated. The undulation of the SP decreased in the course of treatment without statistical significance (p = 0.148) (see Figure 9; for exact values, see Table 2). In the CoA, undulation was lower and quite stable in the course of time without significant differences. The difference in DEJ undulation when comparing the SP and CoA decreased under therapy without statistical significance (p = 1.03; for exact values, see Table 2).

3.2.3. Vessels in Psoriasis

One of the most intriguing findings in our study was the long-term persistence of elongated vessels, sometimes showing a psoriatic vasculature in the CoA. For better understanding, a review of the vessels within the 3D stacks was carried out, depicting the vessels better by applying minimal rendering to the LC-OCT scans, demonstrating a longer persistence under therapy than other dermo-epidermal alterations (see Figure 10). The segmentation of the vessels was still carried out manually when needed and still highly time-consuming. The training of an AI model this could be realized more rapidly in the future.

3.3. Further Epidermo-Dermal Alterations

We also screened for further dermo-epidermal alterations in the SP and CoA, which may occur in PP-like serum crusts, acanthosis, parakeratosis, spongiosis, interface dermatitis, bright inflammatory cells, micro- or macro-vesicles, elongated rete ridges, and Munro micro-abscesses. For the frequency of occurrence of all criteria, please see

Table 3. Further dermo-epidermal alterations in the study plaque (SP) and control area (CoA) in the course of therapy. If examination of the criterion was not possible, due to extensive acanthosis and/or hyperkeratosis, the patient was excluded for the analysis of this criterion. First number = number of patients presenting the criterion, second number = number of examinable patients.

Alteration	V0 SP	V1 SP	V2 SP	V0 CoA	V1 CoA	V2 CoA
acanthosis	12/12	9/12	3/12	0/12	0/12	0/12
hyperkeratosis	12/12	9/12	3/12	0/12	0/12	0/12
parakeratosis	12/12	10/12	7/12	0/12	0/12	0/12
bright inflammatory cell epidermis	10/12	10/12	9/12	10/12	9/12	8/12
elongated rete ridges	11/12	10/12	7/12	0/12	0/12	0/12
dilated vessels	11/12	11/12	9/12	7/12	6/12	5/12
bright inflammatory cell dermis	2/3	4/9	8/11	9/12	8/12	7/12

. Due to the small number of patients included in this study, we did not look for statistical significance among these alterations.

4. Discussion

In this study, we focused on non-invasive live imaging with the LC-OCT of PP under therapy, comparing manual and AI-supported measurements of epidermal thicknesses in lesional and healthy skin. The results of this study show a significant reduction in acanthosis and hyperkeratosis measurable by LC-OCT, correlating with the significant reduction in the clinical severity (LPSI, PASI) and improvement in quality of life (DLQI). In comparing the ET and SCT of the CoA to the corresponding values in the SP during the study period, an approximation of the SP to the CoA could be seen. The applicability of an AI model for the evaluation of epidermal alterations and the severity of actinic keratosis has already been shown [32]. In our study, an AI-supported measurement of the ET and SCT was possible, calculating reliable and highly correlating values compared to subjectively scored measurements. In the future, this is a chance for a faster and more objective evaluation of PP by LC-OCT imaging. AI-supported analysis leads to higher accuracy of the measurements, as it evaluates the thickness all over the scanned part of the lesion, calculating a mean value. Subjective quantifications of ET in different regions in one scan are time-consuming and cannot measure such exact and objective results. We verified all scans with a difference in thickness of more than 30% between both methods. If the DEJ was not visible, the AI algorithm measured until the end of the 3D stack in these parts. Careful revision is crucial when applying an AI model, since it is not capable of stating that measurements are not possible due to bad or poor visibility and always tries to segment. For the stratum corneum, AI-supported measurements showed slightly higher values, especially in CoAs and SPs after effective treatment, adding the first layers of the stratum spinosum to the stratum corneum. The AI model was trained to differentiate stratum corneum and epidermis through the detection of nuclei in the cells, thus drawing the boundary between these layers (stratum corneum: without nuclei or white nuclei for parakeratosis; epidermis: regular, signal poor nuclei). For the upper parts of the stratum granulosum, this can be difficult to detect, even for a trained expert. However, results often differed by only a few micrometers; because of the thin layer, this results in high percentage variations without clinical relevance. The median percentage difference when applying a linear regression model for the difference between manual and AI model-measured SCT was 30.8%. For even more accurate findings in the future, a re-training of the AI algorithm could be aimed for. For the ET, subjective measurements were often slightly higher than the corresponding AI model measurements. This is most likely because the more elongated rete ridges were measured manually. The median percentage difference from the AI model versus subjective measurements of the ET was 13.5%, which is, with respect to the aforementioned reasons, very good. In scans with greater undulation of the stratum corneum or wavy DEJ, the AI-supported measurements seem to calculate smaller but also more exact results because of computing a mean over the whole stack. In conclusion, the main difference between both methods, which also represents a limitation of the study, lies in the fact that subjective measurements were confined to a representative area, whereas the AI model calculated an average across the entire scanned part of the lesion. Future studies could potentially address this discrepancy by manually measuring multiple psoriasis areas and calculating a mean, which might yield results more comparable to those obtained by the AI.

Moreover, the statistical insignificance of DEJ undulation reduction (p = 0.148) observed in this study may be attributed to several factors inherent to the study design and patient population. First, the relatively small sample size (12 patients) limits the statistical power necessary to detect subtle changes in DEJ undulation. Additionally, the heterogeneity of the treatment regimens, with patients receiving different biologic therapies (secukinumab, ixekizumab, and bimekizumab), and variability in the anatomical location of the psoriatic plaques (lower and upper extremities, including the scapular region) may have introduced variability in treatment response. The variation in age (25 to 84 years; mean: 52.9 years), and differences in baseline DEJ characteristics across patients might also contribute to the lack of statistical significance.

Moreover, the study duration (median of 5.4 weeks between V0 and V1, and of 8.1 weeks between V1 and V2) might have been insufficient to capture more profound changes in DEJ undulation. The persistence of elongated rete ridges and other dermo-epidermal alterations, which could diminish at a slower rate compared to hyperkeratosis and acanthosis, may have contributed to the observed results. Another consideration is the technical limitations of LC-OCT; undulation changes might be more challenging to quantify with this imaging modality, as it relies on averaging across the entire lesion, potentially underestimating localized variations.

In future studies, increasing the sample size, standardizing treatment regimens, and extending the follow-up period might help provide more definitive insights into the significance of DEJ undulation changes under therapy. Additionally, integrating imaging modalities with greater penetration depth, such as dynamic OCT, could enhance the sensitivity of detecting subtle alterations in the DEJ.

In concordance with the results obtained by Deußing, Ruini [37], inflammatory cells in the epidermis were also present in most of our patients in the SP. During the study, when the dermis became visible in most patients, inflammatory cells could also be detected in this skin layer in the majority of cases. Interestingly, bright inflammatory cells also occurred frequently in the CoA in the epidermis and dermis and persisted in most patients during the study period. This might be a sign of systemic inflammation caused by severe psoriasis. In some patients we also found further psoriatic changes in the CoA, like dilated vessels. Vascular changes at the edge of PP have already been detected by laser Doppler flowmetry as a marker for activity, preceding epidermal alterations and without histologically detectable psoriatic skin alterations [38]. They are thought to be the last psoriatic change persisting in a regressing plaque [39]. Vascular alterations have also been shown to be detectable by videocapillaroscopy in clinically improved or healed skin in PP under therapy with cyclosporine A [40]. In our study, the persistence of dilated vessels could be detected in the majority of patients and also initially in the majority of clinically healthy CoAs. A longer follow-up seems to be necessary to monitor the long-term evolution of altered vascularization. This observation suggests that vascular normalization may be a gradual process that continues beyond the initial treatment phase. A longer follow-up is indeed necessary to monitor the long-term evolution of altered vascularization, as it could provide critical information about the durability of treatment effects and potential need for maintenance therapy. By enabling the visualization of these persistent vascular changes, LC-OCT offers clinicians a powerful tool for assessing treatment response, adjusting therapeutic strategies, and making informed decisions about the continuation or modification of systemic treatments. These vascular alterations might serve as a marker for severe psoriasis and need for systemic treatment when detectable in healthy skin. For further evaluation, a comparison between patients suffering from severe psoriasis compared to patients suffering from mild or moderate psoriasis receiving only topical treatment might be of great value in the future, including an analysis of clinically healthy skin.

AI model analysis of previously segmented LC-OCT scans for the evaluation of vessels might be of interest in the evaluation of treatment response and continuation or discontinuation of therapy. Dilated vessels seem to persist longer than acanthosis und hyperkeratosis. The diminution of hypervascularization could be investigated as a marker for treatment response. AI-enhanced analysis might be of great help in objectively measuring the extent of increased vasculature.

In our study, in contrast to the study by Orsini, Trovato [22], the SP’s location was not restricted to the patients’ back, but different body regions were included without limitations. LC-OCT scans are feasible in various and diverse body parts. For evaluation, anatomical features and peculiarities like variation in the thickness of healthy epidermis and dermis must be considered for analysis. AI model segmentation and measurement of skin layers were carried out in complete 3D stacks, thus generating more reliable values. In the preceding study by Orsini, Trovato [22], segmentation and measurements were only carried out in several vertical images per patient. In their study, 3D stack segmentation was only conducted in a few patients and not used for statistical analysis, thus not generating such a reliable mean.

In our study, the exact measurement of the ET was initially not possible in 8 of 12 patients, and therefore, no analysis of the upper dermis was possible. This is due to excessive hyperkeratosis as well as acanthosis and the limited penetration depth of the LC-OCT. Similar findings have been reported by Orsini, Trovato [22]. Visualization improved during therapy in concordance with reduction in the LPSI. Statistical analysis was performed using the highest measurable ET, although a higher thickness can be presumed. Therefore, the statistical significance of the reduction would not change when considering even higher initial values. The inability of LC-OCT to measure ET in 8 out of 12 patients at baseline due to excessive hyperkeratosis and acanthosis underscores a critical limitation of the technology in clinical and research applications. The high, nearly cellular resolution of the LC-OCT contrasts the limited penetration depth of the device. It has to be taken into account that it is already offering a greater penetration depth compared to RCM, which offers a similar resolution. This issue highlights the need for further refinement in LC-OCT’s imaging capabilities to effectively penetrate and resolve highly keratinized or thickened epidermal layers. The challenge reduces the generalizability of findings, particularly in populations with conditions characterized by significant epidermal alterations. To address this, future efforts should focus on enhancing the imaging resolution depth, developing adaptive algorithms for image interpretation in hyperkeratotic tissues, and exploring adjunctive methods to preprocess or optimize tissue visualization. Another solution might be to add a second imaging tool with a greater penetration depth like OCT to provide deeper insights. Acknowledging and tackling these limitations is essential to expand the applicability of LC-OCT in diverse dermatological conditions. The technical improvements of the LC-OCT, leading to a bigger penetration depth in the future, would offer a great advantage in the examination of PP.

In one patient, the SP was located in the prepatellar region. This is an area with physiologically higher epidermal thickness, especially in psoriatic lesions in this location [41]. The patient showed clinical improvement under therapy (V0: PASI = 32.2, LPSI = 10; V2: PASI = 8.7, LSPI = 2), but due to the SP’s location, the dermis remained non-measurable during the study period. Other regions did not show such limitations among the study population.

Limitations of this study are the small number of participants, the single-center design, the relatively short study duration, and the lack of subgroup analysis. Follow-up studies with bigger study populations will be needed in the future. This would also offer the possibility of subgroup analysis, e.g., concerning mild and severe psoriasis or different therapeutic treatment modalities. Adding a multi-center design would offer further examination of comparability between different examiners. The control period of 3–4 months was calculated regarding the time needed to reach a significant clinical improvement, which was remarkable most of the time in this time period. We now note that psoriatic changes in the epidermis and dermis seem to be observable longer than clinical alterations (judged by the application of the LPSI). This is a novel insight and could therefore not be known before the initiation of the study. The persistence of vascular changes and also inflammatory patterns during the chosen study period of only three months should give rise to longer examination periods in the future. Longer control periods seem to be necessary in future studies and will be taken into account for planning future studies. An examination of the SP and CoA was carried out with clinical scores and LC-OCT; we did not compare different non-invasive imaging modalities. This is something that may be necessary in future investigations since other imaging modalities, like dynamic optical coherence tomography, can offer a higher visualization depth compared to that of LC-OCT. A comparison would also offer objective verification of the reliability of the findings in each non-invasive imaging tool.

Our study population consisted of severe cases of psoriasis in need of systemic treatment with biologics. The examination of mild to moderate psoriasis under different treatment modalities should be carried out in future research to allow for the comparison of inflammatory patterns. To date, there is no standard concerning the decision of continuation or termination of therapy with biological agents. Long-term follow-up with non-invasive imaging tools might offer the detection of possible prediction markers for long-term remission.

5. Conclusions

An objective evaluation of the therapeutic response in PP by LC-OCT using AI-supported measurements of ET is possible. Other psoriatic changes, like inflammatory patterns and vascular alterations, can be detected in lesional and even healthy skin. Further investigations might reveal clues objectifying the need for systemic treatment. Even unveiling prediction markers for therapeutic response requires the continuation of treatment in macroscopically healthy skin, or long-term remission could be possible in the future.