Systematic screening using line-field confocal optical coherence tomography (LC-OCT) combined with AI–assisted image recognition enabled the detection of subclinical basal cell carcinomas (BCCs) in high-risk individuals, according to findings from a cross-sectional feasibility study published in JAMA Dermatology. 

Background and Study Methods

LC-OCT is a noninvasive imaging technique that has demonstrated greater sensitivity and specificity for diagnosing BCC than clinical examination and dermoscopy. The technique can also differentiate among BCC subtypes to help guide treatment decisions, but it is primarily used to evaluate individual lesions rather than screen large areas of skin.  

Patients were recruited from the Department of Dermatology at Uniklinikum Erlangen between April and October 2025. Eligible patients were required to have at least two established risk factors for BCC, including age 65 years or older, Fitzpatrick skin phototype I or II, a personal history of BCC, chronic sun exposure (≥ 20 years of occupational exposure or ≥ 30 years of recreational exposure), a genetic syndrome associated with an increased risk of BCC (eg, Gorlin-Goltz syndrome, Bazex-Dupré-Christol syndrome, or Rombo syndrome), or prior exposure to ionizing radiation involving the head or neck.  

Most patients were recruited during routine inpatient care. Facial areas covered by wound dressings or dense facial hair were excluded from screening, as were lesions with macroscopic features suggestive of cancer. The investigators systematically examined clinically inconspicuous facial skin to identify carcinomas that were not apparent on clinical examination.  

Imaging was performed by trained personnel using a deepLive (Damae Medical) LC-OCT device. 

The researchers used an integrated AI-based image recognition application to screen facial skin for regions of interest. The application generated a real-time probability score indicating the likelihood of BCC. When the algorithm identified an area with a high probability of BCC, the display changed from blue to yellow to alert the examiner. The research team termed this approach SUBSCAN (subclinical BCC screening by AI-assisted noninvasive imaging).  

The diagnosis of BCC was made independently of the algorithm’s probability score, and suspicious areas were subsequently biopsied for histologic confirmation. Investigators were not blinded to the AI probability scores.  

Because only lesions identified as suspicious for BCC underwent biopsy, the study was subject to verification bias, and only positive predictive value could be determined. 

Study Findings

Of the 150 participants, 62% were male, and the mean age was 72.9 years. More than half (52%) could not undergo complete facial screening because some areas were inaccessible. Patients had a mean of 3.3 BCC risk factors, which were not mutually exclusive and often co-occurred. Phototype I or II was the most common risk factor, present in 95.3% of patients.  

Seventeen subclinical BCCs were identified in 14 patients (9.3%). These included 13 superficial BCCs (76.5%), 3 nodular BCCs (17.6%), and 1 infiltrative BCC (5.9%). Lesions were most commonly located on the forehead (35.3%), followed by the temples (29.4%).  

Of 18 lesions initially diagnosed as BCC using AI-assisted LC-OCT, 15 were histologically confirmed, corresponding to a positive predictive value of 83.3% (95% confidence interval [CI] = 58.1%–96.4%). One lesion was identified as actinic keratosis, and two patients declined biopsy; however, expert analysis supported the diagnosis of BCC in both nonbiopsied lesions. In a sensitivity analysis that included these two lesions, the positive predictive value increased to 94.4% (95% CI = 72.7%–99.9%). BCC subtype classification was accurate in 86.7% of cases.  

The study authors, including lead and corresponding author Moritz Ronicke, MD, of the Department of Dermatology, Uniklinikum Erlangen, Friedrich-Alexander-Universität (FAU) Erlangen-Nürnberg, Erlangen, Germany, noted that clinically inapparent lesions, such as the subclinical BCCs identified in the study, would not typically undergo evaluation with dermoscopy or LC-OCT. “The use of advanced techniques for the detection of BCC, such as LC-OCT, sharpens the eye for very early lesions visible to the naked eye by providing immediate feedback and could therefore even improve the clinical identification of BCC by physicians using this approach,” they wrote.  

However, the natural history of subclinical BCCs remains unclear, including whether they would progress to nodular BCC requiring surgery or could be eliminated by the immune system without treatment. “The finding that superficial BCC accounted for the majority of lesions is relevant, as nonsurgical treatments for superficial BCC have a favorable adverse effect profile. Whether the potential risk of overdiagnosis is acceptable will require formal cost/risk-benefit analysis in future prospective studies,” the study authors wrote.  

Further studies are needed to assess the sensitivity and specificity of SUBSCAN, as well as other measures of diagnostic performance, such as the false-negative rate, to determine its usefulness for screening and follow-up care.  

“Broad implementation in clinical practice further requires assessment of cost-effectiveness, as the identification of lesions of uncertain malignant significance carries the risk of inducing unnecessary procedures,” the study authors noted.  

Insights

In an accompanying editorial, Veronica Rotemberg, MD, PhD, of Memorial Sloan Kettering Cancer Center, commented that “this study highlights both the promise and the complexity of AI-enabled early detection. AI-assisted line-field confocal optical coherence tomography enables visualization of disease at a stage previously beyond routine clinical reach, yet the earlier we detect disease, the less certain we may be about its clinical meaning.” 

She concluded that further studies are needed to define the natural history of subclinical lesions, establish treatment thresholds, and determine whether earlier detection improves patient outcomes. 

DISCLOSURES: Funding was granted by the German Research Foundation. Dr. Ronicke reported receiving personal fees from AbbVie, Almirall, Fidia, Johnson & Johnson, Novartis, Pfizer, Recordati Rare Disease, UCB, and URGO outside the submitted work. Dr. Erdmann reported receiving personal fees from Pierre Fabre, Immunocore, and Novartis outside the submitted work. No other disclosures were reported. 

Dr. Rotemberg reported in-kind support from Kaggle, personal fees from Atria Institute, stock from Inhabit Brands Inc, and data storage with AWS outside the submitted work; Dr. Rotemberg also serves as Chair of the Augmented Intelligence Committee and Board of Directors of the Society for Imaging Informatics in Medicine. 

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.