Skin lesion triage in organ transplant recipients using line‐field confocal optical coherence tomography: a retrospective classification study

Summary

Background/objectives: Organ transplant recipients (OTRs) face an elevated risk of keratinocyte carcinomas (KC), leading to a lower threshold for biopsy. Line‐field confocal optical coherence tomography (LC‐OCT) is a non‐invasive imaging technique capable of visualizing malignant skin changes, yet its role in the management of OTRs remains unexplored. The objective was to investigate the potential of LC‐OCT for triage of skin lesions in OTRs.

Patients and methods: Clinically equivocal lesions in OTRs (n = 75) were scanned using LC‐OCT before biopsy. Scans were assessed retrospectively for predefined LC‐OCT criteria (i.e., image‐markers) for KC and premalignant/in‐situ lesions (Bowen's disease and actinic keratosis). The most high‐yield criteria were identified and combined into a decision‐tree.

Results: A 4‐step decision‐tree for lesion triage with five LC‐OCT criteria was developed, with a sensitivity of 87.5 % for squamous cell carcinoma (n = 8), 94.4 % for basal cell carcinoma (n = 18), 83.3 % for Bowen's disease (n = 12), and 72.7 % for actinic keratosis (n = 22). This triage is based on severe dysplasia, broad strands and keratin pearls for SCC, lobules for BCC, and severe or mild‐moderate dysplasia for premalignant lesions.

Conclusions: This study designed an LC‐OCT decision‐tree for use in immunocompromised patients to assist in triage of equivocal skin lesions with potential for improving clinical‐decision making.

INTRODUCTION

Keratinocyte carcinomas (KCs), including squamous cell carcinoma (SCC) and basal cell carcinoma (BCC), are the most common cancers worldwide. ¹ , ² Organ transplant recipients (OTRs) are at higher risk of developing KCs compared to the general population. This is primarily due to a life‐long adherence to immunosuppressive medicine to preserve graft function. ³ , ⁴ The risk of skin cancer in OTRs is estimated to be between 20–65 fold higher for SCC and 6–10 fold higher for BCC compared to an immunocompetent population. ⁵ , ⁶ , ⁷ , ⁸ Moreover, in contrast to the general population, OTRs tend to develop SCC with more aggressive growth patterns. Recent data shows that SCC recurrence and metastasis can be as high as 13 % and 8 %, respectively, in OTRs. ⁹ , ¹⁰ , ¹¹ In case of SCC metastasis, the five‐year survival rate can be as low as 30.5 %. ¹¹

Given these risks, OTRs undergo regular screenings at specialized dermatology centers to ensure timely and appropriate treatment of KCs. ¹² Nonetheless, OTRs have a high burden of KC mimickers, including premalignant/in‐situ Bowen's disease (BD) and actinic keratosis (AK), and other benign growths, such as viral warts and seborrheic keratosis (SK), ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ complicating the diagnostic process. Hence, clinicians likely have a lower threshold for diagnostic biopsies to minimize the risk of overlooking aggressive KCs.

Noninvasive imaging devices could improve real‐time detection of skin cancer, thereby reducing unnecessary biopsies and minimizing treatment delays. Such devices include line‐field confocal optical coherence tomography (LC‐OCT) that can visualize malignant changes in the epidermal and superficial dermal layers of the skin at high resolution. ²⁰ A recent meta‐analysis of five LC‐OCT studies involving 904 lesions found that LC‐OCT has a high sensitivity (pooled 87 %) for detecting most common skin cancer lesions and Aks. ²¹

To support clinical decision‐making for immunocompromised patients, this study examined LC‐OCT for triage of equivocal skin lesions in OTRs.

PATIENTS AND METHODS

Study design and patient population

This retrospective classification study applied LC‐OCT to clinically equivocal lesions among OTRs from February 2022 to October 2024. The site of recruitment was a specialized transplant clinic located at the Department of Dermatology, Bispebjerg Hospital. Eligible patients were ≥ 18 years old, with a solid organ transplant (i.e., kidney, liver, lung, and heart) and stable immunosuppressive treatment for ≥ 3 months. All included lesions had a histology‐verified diagnosis. The Regional Scientific Ethical Committee [H‐21038387] approved the study, and the study was carried out in accordance with the Helsinki Declaration (2002).

LC‐OCT image acquisition

We employed the commercially available handheld LC‐OCT imaging device (deepLive™, DAMAE Medical, Paris, France) to capture optical images of included lesions. ²⁰ The device operates with a laser at approximately 800 nm and produces high‐resolution images with an axial and lateral resolution of about 1.3 µm and a maximum imaging depth of around 0.5 mm. It combines time‐domain OCT technology with a scanning speed of eight frames/second and confocal microscopy using a 20× objective with a numerical aperture of 0.5. The acquisition method included recording videos within the tumor borders of the demarcated lesions and capturing three‐dimensional LC‐OCT images of targeted areas for analysis.

Image assessment

To obtain the building blocks for the decision‐tree, we qualitatively assessed the presence or absence of predefined LC‐OCT criteria for KC, BD or AK in images of the included equivocal lesions. ²² For BCC, these LC‐OCT criteria were lobules, clefting, and collagen alterations; for SCC, broad strands, keratin pearls, and interrupted DEJ; for BD, severe dysplasia and well‐defined dermal‐epidermal‐junction (DEJ); and for AK, mild‐moderate dysplasia, tumor budding, and well‐defined DEJ.

Developing the LC‐OCT decision‐tree

We employed a tree‐building algorithm, Exhaustive Chi‐Squared Automatic Interaction Detection (Exhaustive CHAID), ¹⁰ , ¹¹ in SPSS, using the results of the image assessment and histology as the reference. The decision‐tree's performance was reported as sensitivity (true positive/true positive + false negative). The algorithm explored all possible decisions based on the most significant associations (p < 0.05) between an LC‐OCT criteria (n = 10) and the five histology‐verified categories (SCC, BCC, BD, AK, and benign lesions). Thus, it identified the most high‐yield LC‐OCT criteria for a diagnosis and combined them into a decision‐tree. To increase generalizability, we employed 10‐fold cross‐validation, meaning that algorithm was trained on 90 % of the data and tested on 10 % of the data, 10 times. Furthermore, we limited the maximum depth of the decision‐tree to five levels, reducing the risk of overfitting (i.e., fitting data to statistical noise).

Number needed to biopsy (NNB)

To provide a frame of reference for the performance of the decision‐tree, we also calculated the studies NNB, which in this study represents the number of biopsies taken before a diagnosis was confirmed with histology, corresponding to the mathematical inverse of the positive predictive value. ²³ , ²⁴

RESULTS

Patient demographics and diagnosis

Table 1 provides a summary of patient demographics and included lesions. We included 51 OTRs with 75 equivocal skin lesions over a 2.5‐year period, of which 34 were men and 17 were women (mean age 63.8 years, range 43‐85). The majority of patients were kidney recipients (68.7 %), followed by liver (21.5 %), lung (5.9 %), and heart (3.9 %) recipients.

TABLE 1.

Patient demographics and diagnosis.

Characteristics	All OTRs (n = 51)
Gender, n (%)
Male	34 (66.7)
Female	17 (33.3)
Mean age at inclusion, years (range)	63.8 (43‐85)
Transplant type, n (%)
Heart	2 (3.9)
Lung	3 (5.9)
Kidney	35 (68.7)
Liver	11 (21.5)
Number of lesions, n (%)
1	31 (60.8)
2	18 (35.3)
3	1 (2.0)
5	1 (2.0)
Histological diagnosis, n (%)
SCC	8 (10.7)
mSCC	4 (5.3)
hSCC	4 (5.3)
BCC	18 (24.0)
sBCC	4 (5.3)
nBCC	13 (17.3)
iBCC	1 (1.3)
BD	12 (16.0)
AK	22 (29.3)
Benign lesions *	15 (20.0)

Abbr.: OTR, organ transplant recipient; SCC, squamous cell carcinoma; mSCC, moderately differentiated squamous cell carcinoma; hSCC, highly differentiated squamous cell carcinoma; BCC, basal cell carcinoma; sBCC, superficial basal cell carcinoma; nBCC, nodular basal cell carcinoma; iBCC, infiltrative basal cell carcinoma; BD, Bowen's disease; AK, actinic keratosis

^*Benign lesions include papilloma/verruca/condyloma (n = 5), seborrheic keratosis (n = 4), dermatitis/folliculitis/inflammation/demodex (n = 4), sebaceous adenoma (n = 1), solar elastosis (n = 1). Percentages are rounded up and therefore the sum of the individual numbers do not always add up to 100.

Approximately one‐third of the patients examined had a KC, including BCC (n = 18) and SCC (n = 8). Premalignant/in‐situ lesions included BD (n = 12) and AK (n = 22), while the remaining lesions were benign (n = 15), among which warts were the most common (n = 5). In terms of clinical appearance, lesions ranged from markedly hyperkeratotic, elevated, wart‐like, to flat erythematous patches (online supplementary Figure S2).

Assessment of LC‐OCT criteria

Figure 1 graphically summarizes the qualitative image assessment of ten predefined LC‐OCT criteria for KC, BD, and AK in included skin lesions (n = 75), grouped by diagnosis. The assessment revealed that certain LC‐OCT criteria were frequent in BCC, such as lobules (17/18), while other criteria were more commonly observed in SCC, including broad strands (5/8), and in BD, severe dysplasia (11/12). Signs of mild‐moderate dysplasia (16/22) were common in AK.

FIGURE 1.

Assessment of line‐field confocal optical coherence tomography criteria in equivocal skin lesions in organ transplant recipients. Image assessment of ten predefined line‐field confocal optical coherence tomography (LC‐OCT) criteria in 75 histology‐verified equivocal lesions. LC‐OCT criteria are presented in two rows. Results are reported as frequencies. Data were processed using IBM SPSS. Presence of an LC‐OCT criterion (+) was coded as 1 and absence (–) as 0. Criterion presence was prioritized over absence, and multiple images per lesion (n = 1–3) were consolidated into a single record (n = 1) to ensure that all identified criteria were retained. Based on consensus from a previous study, ³⁶ the dataset was recoded assuming that mild/moderate and severe dysplasia cannot coexist within a single image, and likewise a well‐defined and an interrupted dermal–epidermal junction (DEJ). In cases of co‐occurrence, the higher severity grade was retained (e.g., severe dysplasia over mild/moderate dysplasia, interrupted DEJ over well‐defined DEJ). Abbr.: BCC, basal cell carcinoma; SCC, squamous cell carcinoma; BD, Bowen's disease; AK, actinic keratosis

LC‐OCT decision‐tree for triage of equivocal skin lesion

Table 2 demonstrates the decision‐tree's performance, reported as sensitivity, based on the Exhaustive CHAID analysis. For BCC, sensitivity with a single criterion (lobules, p < 0.001) was 94.4 % (1 false negative case), while for SCC, it was 87.5 % (1 false negative case) using three criteria (severe dysplasia, p = 0.008, broad strands, p = 0.001, and keratin pearls, p = 0.008). In comparison, BD could be detected with a sensitivity of 83.3 % (2 false negative cases) with one criterion (severe dysplasia, p = 0.008). For AK, the sensitivity was slightly lower at 72.7 % (6 false negative cases) using a single criterion (mild‐moderate dysplasia, p = 0.001). The absence of the above‐mentioned criteria resulted in a sensitivity of 53.3 % for benign lesions (7 false negative cases). Online supplementary Figure S1 graphically illustrates the decision‐tree, consisting of 4 steps and 5 LC‐OCT criteria. The tree should be interpreted stepwise by evaluating the presence or absence of an LC‐OCT criterion at each step, thus enabling triage of equivocal lesions in OTRs into three groups: keratinocyte carcinomas (BCC/SCC), premalignant/in‐situ lesions (BD/AK), and benign lesions.

TABLE 2.

Decision‐tree performance.

Decision‐tree
Histology		BCC	SCC	BD	AK	Benign	Sensitivity
	BCC	17	0	0	0	1	94.4%
	SCC	0	7	1	0	0	87.5%
	BD	1	0	10	1	0	83.3%
	AK	0	0	5	16	1	72.7%
	Benign	1	2	1	3	8	53.3%

Green: cases of agreement between line‐field confocal optical coherence tomography decision‐tree and histology.

Red: cases where LC‐OCT down‐staged (suggested a less severe diagnosis).

Orange: cases where LC‐OCT up‐staged (suggested a more severe diagnosis).

Abbr.: BCC, basal cell carcinoma; SCC, squamous cell carcinoma; BD, Bowen's disease; AK, actinic keratosis

Biopsy trends

Based on the calculated NNB for the study, approximately three biopsies were taken to confirm one case of SCC (NNB 2.8), BCC (NNB, 2.75), or BD (NNB, 2.5). In comparison, fewer biopsies were taken to confirm one case of AK (NNB 1.5) or a benign lesion (NNB 1.5).

DISCUSSION

In this retrospective classification study, we designed a simple 4‐step LC‐OCT decision‐tree with only five criteria to triage equivocal lesions in OTRs into KCs, premalignant/in‐situ‐, and benign lesions. The decision‐tree had only a single false negative case for KCs and only two false negative cases for BD. These findings suggest that the use of the decision‐tree could facilitate more informed real‐life decisions in the OTR clinic, particularly when triaging KCs and BD vs other KC‐mimickers. For patients, the potential advantages include added safety, as well as fewer biopsies and hospital visits.

The sensitivity of LC‐OCT for KC/BD, based on the results of the decision‐tree, is non‐inferior to other studies using LC‐OCT to detect skin cancers in an immunocompetent population. The first research groups to describe the sensitivity of LC‐OCT for KCs (n = 182–226), reported sensitivities of 86 % for SCC/BD and 97 %–98 % for BCC. ²⁴ , ²⁵ This also suggest that KCs in OTRs have similar morphological characteristics as seen in the general population and that LC‐OCT can be used in a similar fashion for these high‐risk patients. Moreover, the decision‐tree's high sensitivity has the potential of lowering the observed high NNB specific to OTRs.

This study employed an algorithm to determine LC‐OCT sensitivities for KC, BD, and AK that may have meaningful clinical implications. For instance, if a clinician were to biopsy only lesions predicted to be BCC, SCC, or BD based on the algorithm, 30 biopsies could have been avoided while detecting all BD cases and all but one each of BCC and SCC cases. While there is currently no consensus whether these thresholds are clinically acceptable, our findings represent early steps towards implementing LC‐OCT in clinical practice. Moreover, decision trees, in conjunction with emerging artificial intelligence (AI)‐based decision support models for BCC, ²⁶ may be valuable tools for clinicians by offering interpretability and transparency in the analysis of LC‐OCT images to support the evolving use of AI‐based software.

To our knowledge, this is the first proposed LC‐OCT decision‐tree for triaging equivocal skin lesions in OTRs. However, a recent larger study designed a similar LC‐OCT decision‐tree for the detection of BCC among equivocal lesions in the general population cohort (n = 303 lesions). ²⁷ The authors of that study reported a sensitivity of their LC‐OCT decision‐tree for BCC vs. non‐BCC of 97 %. We add to that study by presenting an LC‐OCT decision‐tree with high sensitivity scores also for SCC, BD, and AKs. Thus, in addition to providing better clinical discrimination between equivocal AK/BD and SCC, there is potential to improve histological accuracy using LC‐OCT, as only 67 % of SCC cases are confirmed by histology. ²⁸

Exhaustive CHAID excluded five LC‐OCT criteria (collagen alterations, clefting, tumor budding, and interrupted/well‐defined DEJ). However, while these criteria were not part of our decision‐tree, they may still be important for skin cancer assessment, e.g., an interrupted DEJ has been found associated to SCC in previous LC‐OCT studies. ²⁹

In a few cases, the decision‐tree algorithm tended to error on the side of caution and suggested a more severe diagnosis than histology (orange marking in Table 2). For example, a sebaceous adenoma was misclassified as BCC (S1, 4d), likely due to glandular hyperplasia mimicking BCC lobules. ³⁰ Similarly, two cases classified as SCC were SK (S1, 3b and 8b), possibly due to similar morphological characteristics in SK and SCC. ³¹ Thus, the decision‐tree could be refined by incorporating specific LC‐OCT criteria for benign lesions, which OTRs have a propensity to develop. For example, adnexal skin conditions with glandular hyperplasia could be distinguished from BCC by the presence of (1) granular‐lobular pattern and (2) a connection between the lobular structures and the hair follicle. ³² SK specific LC‐OCT criteria also exist, such as the string‐sign in acanthotic SK and thick layers of keratin in hyperkeratotic SK. ³³

More critical misdiagnoses occurred when LC‐OCT suggested a less severe diagnosis than histology (red marking in Table 2). This included a single case of false negative BCC. One possible explanation may be that the lesion was located at a challenging angle (the alar crease) and had an ulcerated surface (S1, 15e), limiting full LC‐OCT coverage, and resulting in a potential oversight of BCC lobules. ³⁴ Additionally, a single case predicted to be BD turned out to be SCC, because neither broad strand nor keratin pearls were found. Thus, an LC‐OCT examination of skin lesions in OTRs should likely not stand alone but can complement clinical inspection and dermoscopy assessments. ²⁵

The study's limitations include confirmatory bias because the image assessment was conducted by a single evaluator. However, all scans were acquired by the same experienced clinician, which helped provide consistent evaluations and reduced the risk of observer variability. While this study is the first attempt to provide a simple LC‐OCT decision tree to triage BCC, SCC, and precursor lesions among OTR patients, the algorithm should be validated in an external cohort to determine its performance in a different sample. Moreover, given the moderate sample size of included skin lesions (< 100), different histological subtypes were not included in the decision tree, but this provides an opportunity for future studies. For reference, superficial BCC can be identified by lobules connected to the epidermis, nodular BCC by lobules not connected to the epidermis, and infiltrative BCC by branched lobules or shoal of fish patterns. ³⁵ Statistical limitations include the use of a single statistical algorithm, and the inability to perform more extensive statistical training on the dataset due to the limited sample size.

Conclusions

This study developed an LC‐OCT decision‐tree for the triage of equivocal lesions in the transplant clinic, which may support real‐time clinical decision‐making. However, given the high prevalence of KC mimickers in OTRs, further investigation into LC‐OCT criteria specific for these lesions could improve the sensitivity of the proposed decision‐tree.

FUNDING

The work was executed as a part of the Danish Research Center for Skin Cancer (www.researchcenterforskincancer.org), a public‐private research partnership between the Private Hospital Molholm, Aalborg University Hospital and Copenhagen University Hospital, Bispebjerg and Frederiksberg.

CONFLICT OF INTEREST STATEMENT

None.

Supporting information

Supporting Information

Supporting Information

Supporting Information