Multimodal Imaging of Basal Cell Carcinoma: A Usability Study

Background:

Basal cell carcinoma (BCC) is the most common type of skin cancer, accounting for approximately 80% of all keratinocyte carcinomas. Although BCC rarely spreads to other parts of the body, it can cause considerable damage to surrounding tissue if left untreated, and both its diagnosis and management place a substantial economic burden on healthcare systems.

The current gold standard for diagnosing BCC is a biopsy with histopathological examination, in which a small sample of skin is removed and examined under a microscope by a pathologist. While highly accurate, this procedure is invasive, causes discomfort, leaves a scar, and requires laboratory processing time before a diagnosis can be confirmed. In clinical practice, dermoscopy (a non-invasive handheld magnification tool) is widely used as a first-line technique to examine surface and near-surface skin structures and guide clinical decision-making. However, dermoscopy does not provide information about tumor depth, subcutaneous invasion, or detailed tissue architecture at the cellular level.

Over the past decade, new non-invasive imaging technologies have emerged that allow detailed visualization of skin lesions at varying depths and resolutions, without the need for tissue removal. These include optical coherence tomography (OCT), line-field confocal optical coherence tomography (LC-OCT), and high-frequency ultrasound (HFUS). Each of these modalities offers complementary information about the skin and underlying tissue. In recent years, artificial intelligence (AI) software modules have been integrated into these imaging systems to assist clinicians in interpreting images and detecting BCC-specific features automatically in real time.

Despite growing evidence that these imaging modalities can detect and characterize BCC with high accuracy, their clinical usability has not been systematically evaluated in a prospective, real-world clinical setting comparing all three modalities head-to-head. This knowledge gap highlights the need for dedicated usability studies to generate essential real-world data that can inform clinical implementation.

Rationale:

There is a recognized need to explore the clinical usability of AI-assisted multimodal imaging in the assessment of BCC, encompassing time consumption and clinical efficiency, technical feasibility across anatomical regions, and patient-reported discomfort during imaging. Comparing AI-assisted imaging findings with histopathological outcome will further allow an exploratory assessment of agreement between imaging and histology. Together, these data will provide a structured basis for evaluating the clinical usability of OCT, LC-OCT, and HFUS in this setting, and for determining which factors most affect their feasibility in routine clinical workflows.

Study Design:

This is a prospective, observational, single-centre usability study. The study will enroll up to 50 participants with histologically confirmed or clinically suspected basal cell carcinoma. Each participant will undergo a structured, multimodal lesion assessment at a single study visit. All imaging procedures are strictly observational and non-interventional. Subsequent clinical management, including any biopsy or surgical excision, will proceed according to standard clinical practice and is not influenced by the study protocol. Treating physicians may review imaging findings if requested. The study is conducted at the Department of Dermatology, Bispebjerg Hospital, Copenhagen, Denmark, and is funded by the Danish Research Center of Skin Cancer (Videncenter for Hudkræft).

Imaging Modalities:

Three non-invasive, AI-assisted imaging modalities are evaluated in this study. All three devices are CE-marked as Class IIa medical devices under EU MDR 2017/745 and are used within their approved intended purpose.

OCT uses low-coherence near-infrared light at approximately 1,305 nm to produce cross-sectional images of skin microstructure based on the interference of backscattered light. The VivoSight Dx (Michelson Diagnostics, England) uses multi-beam technology to achieve a lateral resolution of less than 7.5 micrometres and an axial resolution of approximately 5 to 10 micrometres, with a penetration depth of up to 1 to 2 mm and a scan area of up to 6 by 6 mm. The system incorporates VivoAid, an AI software module for AI-assisted BCC detection and image interpretation support. VivoAid processes scan data to detect image biomarkers indicative of non-melanoma skin cancer and presents results as a 3D volumetric markup and per-slice segmentation overlay. OCT provides fast, real-time imaging and has an established clinical evidence base for BCC assessment.

LC-OCT combines line-field illumination with confocal detection using a broadband supercontinuum laser light source in the 600 to 900 nm range. This enables both vertical (B-scan) and en face imaging, as well as 3D reconstruction, achieving a lateral resolution of approximately 1.3 micrometres and an axial resolution of approximately 1.1 micrometres, which is the highest resolution of the three modalities evaluated in this study, while maintaining a penetration depth of up to 500 micrometres. The deepLive system (model OSP12, DAMAE Medical, Paris, France) is a portable, hand-held system with an integrated dermoscopy camera for precise lesion targeting. It incorporates the deepLive AI module, which provides a real-time BCC probability score from 0 to 100% accompanied by an attention heatmap that highlights regions contributing to the AI prediction. LC-OCT enables near-histological visualization of skin microstructure, including cellular morphology, epidermal architecture, and dermal features associated with BCC subtypes.

HFUS uses high-frequency sound waves at 71 MHz to generate real-time greyscale images of skin and subcutaneous tissue structures. The SkinScanner M (Dermus Ltd., Budapest, Hungary) is a portable, hand-held system that provides B-mode and Doppler imaging. It offers a penetration depth of up to 10 to 15 mm, which is the greatest of the three modalities, making it particularly well-suited for assessing tumor thickness, deep invasion, and subcutaneous boundaries. Lateral resolution is approximately 50 to 200 micrometres. The system incorporates SkinAid, an AI software module that performs automated skin layer segmentation and lesion depth and width measurement. Standard ultrasound contact gel is applied to the skin surface prior to imaging to ensure adequate acoustic coupling.

AI-Assisted Analysis:

Each imaging system incorporates a device-embedded AI module for lesion characterization, segmentation, and BCC detection support. The AI output for each modality is recorded separately from the clinician's independent assessment, allowing comparison between the clinician's pre-AI diagnosis, the AI output, and the clinician's final post-AI decision. This approach enables evaluation of whether and how AI assistance influences clinical decision-making in real-world practice.

Study Procedures:

Each participant attends a single study visit at the Department of Dermatology, Bispebjerg Hospital. The estimated duration of the visit is approximately 40 minutes per lesion. Participants with more than one eligible lesion may have a longer visit. The study visit includes clinical photography, dermoscopy, sequential non-invasive imaging with all three modalities, completion of a patient questionnaire, and completion of a clinician registration form. High-resolution standardized clinical photographs of the lesion are obtained and assigned a unique anonymized ID number. Dermoscopic examination is performed using the Handyscope (DermLite, California, USA) and findings are recorded according to standard dermoscopic criteria for BCC. For HFUS, standard ultrasound gel is applied to the skin surface. For LC-OCT, paraffin viscous oil is applied to the probe tip or lesion surface to achieve optimal optical contact. For OCT, optical immersion oil is applied to the lesion surface.

Participants complete a brief structured questionnaire immediately after all imaging procedures. The questionnaire covers discomfort experienced during imaging rated on a scale from 0 to 5 (0 equals no discomfort, 5 equals discomfort equivalent to that of a biopsy), overall satisfaction with the scanning experience, position and comfort during the examination, and preference for non-invasive scanning over diagnostic biopsy. The examining clinician completes a structured registration form immediately following all imaging procedures, capturing time measurements, practical feasibility of scanning at the anatomical location of the lesion, image quality assessment per modality, AI output for each device, the clinician's diagnosis before and after seeing the AI result, and any technical challenges encountered during imaging.

Histological Follow-up:

Histological diagnoses are collected from participants' electronic medical records for all lesions that undergo biopsy or surgical excision as part of routine clinical care. Histological results serve as the reference standard for correlation with imaging findings. The following information is collected: histological diagnosis (BCC confirmed, other diagnosis, or no malignancy), BCC histological subtype (nodular, superficial, infiltrative, morphoeic, or basosquamous), tumor thickness and depth of invasion where reported, and other histological findings of clinical relevance. Collection of histological data takes place only after informed consent has been obtained.

Outcome Measures:

The primary endpoint is time consumption per patient for each imaging modality (OCT, LC-OCT, and HFUS) and in total, measured in minutes. Time is recorded from the start of device set-up to the completion of image acquisition and documentation for each modality. Secondary endpoints include patient-reported discomfort during imaging for each modality, practical feasibility of scanning with each device across different anatomical regions rated as simple, challenging, or complex, image quality of each imaging device across different anatomical regions rated as high, adequate, or low, and overall agreement between AI-assisted imaging findings and histopathological morphology. Exploratory endpoints include assessment of BCC lesion depth in millimetres derived primarily from HFUS and SkinAid measurements, and characterization of BCC histological subtype based on imaging features compared to histopathological diagnosis.

Sample Size:

Sample size estimation is based on pilot data from 14 tumors assessed at the study site, in which mean imaging time and standard deviation were calculated for each device: HFUS (mean 10.86 minutes, standard deviation 2.96), LC-OCT (mean 19.36 minutes, standard deviation 9.03), and OCT (mean 8.86 minutes, standard deviation 3.32). Power analyses were performed using a paired design at a two-sided significance level of 0.05 and a desired power of 0.80. The most conservative estimate was obtained for the comparison of HFUS versus OCT, which required 41 lesions per modality to achieve a power of 0.81. To account for potential dropout and incomplete imaging due to anatomical or practical constraints, the target sample size was set at 50 histologically verified BCC lesions with complete multimodal imaging data.

Ethics and Data Protection:

All participants provide written informed consent before any study-related procedures are initiated. Participation is entirely voluntary, and participants may withdraw their consent at any time without any consequences for their future treatment at the hospital. A minimum reflection period of 24 hours is offered before consent is obtained. The study has received a positive opinion from the Medical Research Ethics Committee of the Capital Region of Denmark (VMK case no. 16-0302-194). All study data are securely stored on the Capital Region of Denmark's servers in pseudonymized form using a unique study ID number, in accordance with the EU General Data Protection Regulation (GDPR), the Danish Data Protection Act, and the Danish Health Act. No biological samples are collected. Data will be stored for 10 years following study completion.

Publication:

Results will be submitted for publication in peer-reviewed scientific journals and presented at national and international dermatology conferences, regardless of outcome.