Melanoma Screening in the AI Era: Why Dermatologists Remain the Clinical Standard and AI Should Stay an Adjunct

Introduction

Artificial intelligence has entered melanoma detection at a moment of legitimate clinical need and persistent evidentiary uncertainty. Melanoma is potentially lethal, and prognosis differs substantially by stage at diagnosis. In 2026, the Surveillance, Epidemiology, and End Results program estimated 112,000 new U.S. melanoma cases and 8,510 deaths. Five-year relative survival is substantially higher for localized melanoma than for distant-stage disease. (SEER)

These facts create a strong intuitive case for earlier detection. They do not, by themselves, prove that routine screening of asymptomatic, average-risk adults reduces melanoma mortality. That distinction is central to any evidence-based discussion of AI. A tool may classify selected lesions accurately, but a screening program must improve meaningful health outcomes, limit harms, and perform reliably in the population in which it is implemented.

The question “Are dermatologists still the gold standard?” requires refinement. Histopathologic examination remains the diagnostic reference standard when melanoma is suspected. Dermatologists are better described as the clinical standard for integrating lesion morphology, dermoscopy, total-body context, patient risk, longitudinal change, biopsy selection, and follow-up. AI can assist with selected components of this process, but it does not reproduce the full clinical act of melanoma evaluation.

Current Screening Guidance

Skin cancer remains the most common malignancy worldwide, with the incidence of both melanoma and nonmelanoma skin cancers continuing to increase in many countries. Early detection is widely recognized as an important strategy for improving outcomes, particularly for melanoma, where prognosis is strongly associated with tumor thickness and stage at diagnosis. Nevertheless, despite the intuitive appeal of routine skin cancer screening, the evidence supporting population based visual screening remains limited. This distinction between the evaluation of asymptomatic individuals and the assessment of suspicious lesions is fundamental to evidence based clinical practice.

The 2023 U.S. Preventive Services Task Force (USPSTF) concluded that the current evidence is insufficient to determine the balance of benefits and harms of routine visual skin examination performed by a clinician for the purpose of screening asymptomatic adolescents and adults. This conclusion, designated as an “I statement,” reflects uncertainty in the available evidence rather than a recommendation against skin examinations. Specifically, it indicates that existing studies do not adequately demonstrate whether routine clinician performed skin examinations improve clinically meaningful outcomes such as reduced melanoma specific mortality or decreased morbidity at the population level. (USPSTF recommendation)

Importantly, this recommendation applies only to asymptomatic adolescents and adults who do not have signs or symptoms suggestive of skin cancer and who are not known to be at increased risk. It does not apply to individuals presenting with suspicious skin lesions, patients reporting new or evolving cutaneous symptoms, those with a personal or family history of melanoma or other skin cancers, individuals with hereditary cancer syndromes, or patients undergoing regular dermatologic surveillance because of elevated risk factors such as extensive ultraviolet exposure, numerous atypical nevi, previous organ transplantation, or chronic immunosuppression.

For practicing clinicians, distinguishing between screening and diagnostic evaluation has important implications for patient management. An asymptomatic individual requesting a routine total body skin examination during a preventive health visit represents a screening scenario. In contrast, a patient presenting with a new pigmented lesion, rapid lesion growth, asymmetry, border irregularity, color variation, increasing diameter, evolving morphology, ulceration, spontaneous bleeding, persistent itching, pain, nodularity, or the appearance of an “ugly duckling” lesion requires diagnostic evaluation rather than screening. Lesions occurring in anatomically challenging locations such as acral surfaces, beneath the nails, mucosal sites, or areas with previous scarring also warrant careful assessment because these locations may harbor less common but clinically significant melanoma subtypes.

Diagnostic evaluation should include a thorough history and focused dermatologic examination, supported when appropriate by dermoscopy, clinical photography, sequential digital monitoring, or biopsy. Referral to a dermatologist or teledermatology consultation may be appropriate when diagnostic uncertainty exists or when specialized expertise is required. Early lesion directed assessment remains essential because prompt diagnosis continues to be one of the most effective means of improving outcomes in patients with melanoma and other malignant skin tumors.

The 2023 U.S. Preventive Services Task Force concluded that evidence is insufficient to assess the balance of benefits and harms of visual skin examination by a clinician to screen for skin cancer in asymptomatic adolescents and adults. This “I statement” is not a recommendation against evaluating suspicious lesions.

The recommendation applies to adolescents and adults without signs or symptoms of skin cancer. It does not apply to patients with suspicious lesions, symptoms suggesting skin cancer, a personal or family history of skin cancer, or patients already under surveillance because of elevated risk.

For clinicians, the distinction is practical. An asymptomatic, average-risk patient asking about a routine total-body skin examination presents a screening question. A patient with a new, changing, bleeding, ulcerated, symptomatic, irregular, nodular, acral, subungual, or “ugly duckling” lesion presents a diagnostic question. That patient warrants lesion-directed evaluation, dermatology referral, teledermatology assessment, or biopsy by an appropriately trained clinician.

Why AI Does Not Resolve the Screening Evidence Gap

The rapid development of artificial intelligence has generated considerable enthusiasm regarding its potential role in skin cancer detection. Deep learning algorithms have demonstrated impressive performance in classifying dermoscopic and clinical images, with some studies reporting diagnostic accuracy comparable to experienced dermatologists under controlled conditions. These findings suggest that artificial intelligence may become a valuable clinical decision support tool capable of improving lesion assessment, facilitating triage, expanding access to specialist expertise, and supporting teledermatology services.

Despite these advances, artificial intelligence should not be viewed as a solution to the current evidence gap surrounding routine skin cancer screening. High diagnostic accuracy for image classification does not necessarily translate into improved outcomes when applied to population screening programs. Screening interventions must ultimately demonstrate measurable clinical benefits, including reductions in melanoma related morbidity and mortality, earlier diagnosis of clinically significant disease, and improved long term patient outcomes. They must also be evaluated for potential harms, including unnecessary biopsies, false positive results, false reassurance following false negative assessments, overdiagnosis of biologically indolent lesions, increased patient anxiety, healthcare costs, resource utilization, and equitable access across diverse populations.

Interpreting favorable stage distribution data also requires caution. An apparent increase in early stage melanoma detection does not automatically indicate that screening improves survival. Earlier diagnosis may simply reflect lead time bias, whereby diagnosis occurs earlier without altering the natural history of disease. Similarly, increased diagnostic intensity may identify slow growing or clinically insignificant lesions that would never have become symptomatic during the patient’s lifetime, resulting in overdiagnosis and potential overtreatment.

The USPSTF evidence review concluded that currently available studies provide inadequate evidence that clinician performed visual skin examinations reduce melanoma specific morbidity or mortality. Population based investigations reviewed by the Task Force have not consistently demonstrated a survival advantage attributable to routine skin screening. Consequently, although artificial intelligence may enhance diagnostic accuracy within existing clinical workflows, its incorporation alone does not establish that population screening improves patient centered outcomes. (USPSTF evidence summary)

An additional limitation concerns the datasets used to develop and validate many artificial intelligence algorithms. Numerous studies rely on curated image repositories, referral center databases, or highly selected clinical populations. These datasets frequently contain disproportionately large numbers of suspicious lesions, high quality standardized photographs, lighter skin phototypes, and lesions that have already undergone clinical selection by experienced healthcare professionals. Such conditions differ substantially from the complexity and variability encountered in routine primary care and community based screening.

As a result, algorithm performance observed in research settings may not generalize to broader clinical populations. Challenges remain in accurately evaluating lesions in individuals with darker skin tones, identifying amelanotic melanoma that lacks characteristic pigmentation, recognizing acral lentiginous melanoma occurring on the palms and soles, detecting subungual melanoma beneath the nail plate, assessing mucosal melanoma, and evaluating lesions in immunocompromised patients or those with uncommon clinical presentations. Differences in image quality, lighting conditions, lesion morphology, and anatomical location further complicate the real world application of artificial intelligence systems.

Future research should therefore extend beyond measures of diagnostic accuracy and evaluate whether artificial intelligence improves clinically meaningful outcomes within prospective screening programs. Large, population based studies should assess melanoma specific mortality, advanced disease incidence, biopsy rates, healthcare costs, patient satisfaction, accessibility, and performance across diverse demographic and clinical populations. Particular attention should be given to ensuring equitable performance across different skin phototypes, geographic regions, healthcare settings, and socioeconomic groups.

In summary, current evidence does not support routine clinician performed visual skin examinations as a proven population screening strategy for asymptomatic, average risk individuals, although suspicious lesions always require prompt diagnostic evaluation. Artificial intelligence represents a promising adjunct for lesion assessment and clinical decision support, but its impressive image classification capabilities should not be conflated with evidence supporting effective population screening. Until robust prospective data demonstrate improvements in patient centered outcomes, artificial intelligence should be viewed as a complementary tool that augments, rather than replaces, clinical judgment, careful risk assessment, and evidence based dermatologic practice.

Dermatology, Dermoscopy, and the Clinical Standard

Dermoscopy improves diagnostic accuracy for melanoma compared with unaided visual inspection when used by trained clinicians. Cochrane evidence supports dermoscopy as more accurate than visual inspection alone for suspicious pigmented lesions, although performance depends on training, clinical context, lesion selection, and image quality. (PubMed)

The dermatologist’s value is not limited to visual pattern recognition. It includes determining which lesions matter, which require biopsy, which may be monitored, and how an individual lesion fits within the patient’s overall nevus phenotype, risk profile, and history of change.

Current AAD guidance addresses appropriate biopsy techniques for lesions clinically suggestive of melanoma and emphasizes histopathologic interpretation as central to diagnosis and management. ESMO guidance similarly places melanoma diagnosis, staging, treatment, and follow-up within a broader clinical pathway. (AAD melanoma guideline)

For selected high-risk patients, dermatologists may use total-body photography, sequential digital dermoscopy, or structured surveillance. These approaches are particularly relevant for patients with numerous nevi, atypical mole phenotypes, prior melanoma, or familial or genetic risk.

AI may eventually strengthen longitudinal workflows. The clinically consequential decision, however, often remains whether a lesion is new, evolving, discordant with the patient’s background pattern, or concerning despite a reassuring adjunctive result.

What Current AI Evidence Supports

AI systems for skin cancer classification have shown promising diagnostic performance. A 2024 systematic review and meta-analysis found that AI algorithms had higher pooled sensitivity and specificity than clinicians overall and were clinically comparable with expert dermatologists in the included studies.

The authors emphasized that limitations in clinical practice should be considered and that future studies should focus on real-world settings and AI assistance rather than isolated algorithm performance. The studies were heterogeneous, and many were not conducted in representative screening populations. (PubMed)

A separate 2024 systematic review and meta-analysis of human-AI interaction found that AI assistance was associated with higher pooled clinician sensitivity and specificity, with the largest improvement among nondermatologists. However, most included studies were conducted in experimental settings. The review therefore described potential benefit rather than established improvement in routine clinical outcomes. (PubMed)

Human-computer collaboration studies also show that the quality of AI advice matters. High-quality decision support may improve clinician performance, particularly among less-experienced users, whereas faulty AI recommendations can mislead clinicians across experience levels. (PubMed)

Prospective evidence is increasing. A 2026 systematic review and meta-analysis of prospective studies found broadly comparable melanoma diagnostic performance between AI systems and dermatologists, with similar pooled sensitivity and specificity. The review included more than 2,500 patients but found frequent risk of bias, particularly from preselection of lesions suspected of melanoma and reliance on binary classification tasks. Generalizability to unselected populations therefore remains limited. (PubMed)

A prospective Swedish primary care trial also reported high diagnostic performance for a specific AI-based clinical decision-support application used by primary care physicians evaluating lesions of concern. The findings apply to the studied tool, users, clinical pathway, and health system. They should not be generalized automatically to other algorithms, consumer applications, devices, or screening populations. (PubMed)

The appropriate conclusion is neither dismissive nor promotional. AI may be useful as a second reader, triage aid, workflow tool, or diagnostic adjunct. Current evidence does not support presenting it as an autonomous melanoma screening strategy or a replacement for dermatologic evaluation.

FDA-Authorized Adjunctive Devices: Boundaries Matter

The FDA’s De Novo classification documents for DermaSensor illustrate the regulatory boundaries clinicians should understand.

DermaSensor is a prescription, Class II, software-aided adjunctive diagnostic device intended for use by physicians who are not dermatologists. It is indicated for evaluating lesions suggestive of melanoma, basal cell carcinoma, or squamous cell carcinoma in patients 40 years of age or older, to assist in deciding whether to refer the patient to a dermatologist. (FDA De Novo database)

The device should be used with the totality of clinically relevant information, including visual assessment of the lesion. It is intended for lesions already assessed as suspicious for skin cancer and is not intended as a screening tool.

It should not be:

• Used as the sole diagnostic criterion
• Used to confirm a clinical diagnosis
• Used as a stand-alone diagnostic device
• Used to replace biopsy
• Used to replace clinical decision-making

The FDA documents also identify important limitations. Performance has not been specifically evaluated in some patients at increased risk for skin cancer, including those with inherited or drug-induced photosensitivity, genetic predisposition to melanoma or basal cell carcinoma, immune compromise, or other medical conditions that increase skin cancer or metastatic risk.

The device is intended for primary lesions. It has not been tested on previously biopsied, recurrent, or metastatic lesions; scars; tattoos; sunburned skin; hairy areas; palms; soles; mucosal surfaces; genitals; ears; lesions within 1 cm of the eye; or lesions under the nails.

The FDA also notes that less sensitivity data are available for melanoma in patients with Fitzpatrick skin phototypes IV through VI. Referral decisions for suspicious pigmented lesions in these groups should be based primarily on clinical concern. (FDA De Novo summary)

These limitations are not technical footnotes. They define safe use. An adjunctive device used outside its intended population, lesion type, body site, user group, or clinical setting can create false reassurance or unnecessary escalation.

Practical Implications for Internal Medicine Specialists and Other Clinicians

Most internal medicine subspecialists are not routinely tasked with formal total-body skin cancer screening. They do, however, encounter exposed skin during clinical care.

Cardiologists may examine the chest wall and extremities. Nephrologists and transplant clinicians manage patients receiving long-term immunosuppression. Rheumatologists, gastroenterologists, pulmonologists, neurologists, endocrinologists, oncologists, and pharmacists may encounter patients using immunomodulating or photosensitizing therapies. Hospitalists and advanced practice clinicians may notice lesions during general examinations.

The practical task is not to become a dermatologist. It is to recognize when a lesion warrants escalation.

Clinicians should consider whether a lesion is:

• New or changing
• Bleeding, painful, pruritic, or ulcerated
• Asymmetric or irregularly bordered
• Variably pigmented
• Nodular or rapidly enlarging
• Acral or subungual
• Different from the patient’s other lesions

The ABCDE criteria and the “ugly duckling” sign remain useful clinical heuristics, but they are not definitive diagnostic tests. Some melanomas do not meet classic criteria, and some benign lesions do.

When concern exists, a reassuring AI result should not override the clinical history or examination. Appropriate next steps may include expedited dermatology referral, biopsy by an appropriately trained clinician when within scope, or teledermatology triage when available.

Documentation should include lesion duration, evolution, symptoms, relevant risk factors, examination findings, and the follow-up plan. Clinical photography may be useful when appropriate, with patient consent and compliance with organizational privacy, security, and image-retention policies.

Patient Selection and Risk Stratification

Average-risk, asymptomatic adults remain the group with the greatest screening uncertainty. Current USPSTF guidance does not establish a net benefit for routine clinician visual screening in this population.

Patients with suspicious lesions or elevated risk require individualized evaluation outside the average-risk screening framework. Relevant risk factors may include:

• Prior melanoma
• Multiple atypical nevi
• Strong family history
• Familial atypical mole and melanoma syndrome
• Immunosuppression or transplant history
• Known genetic predisposition
• Extensive ultraviolet exposure
• Lesions with concerning evolution

These groups do not represent a single uniform risk category. Surveillance intensity, examination frequency, imaging strategies, and biopsy thresholds should be individualized according to the patient’s history, phenotype, prior pathology, and specialist assessment.

Some higher-risk populations fall outside the scope of average-risk screening recommendations. Some may also fall outside the validation or intended-use population of a particular AI device.

Equity and Underrepresented Presentations

Melanoma incidence is lower among populations with darker skin, but diagnosis may occur at a later stage. Differences in access, awareness, clinical presentation, risk factors, and diagnostic pathways may contribute to this disparity. (USPSTF recommendation)

Melanoma in patients with darker skin may occur at acral, subungual, or mucosal sites that are poorly represented in many training datasets and device-validation studies. Amelanotic lesions, rare melanoma subtypes, complex nevus phenotypes, and lesions in immunosuppressed patients may also be underrepresented.

Lower population incidence should not lower clinical concern when a lesion is changing, symptomatic, acral, subungual, amelanotic, ulcerated, or otherwise atypical.

AI performance should not be assumed to be equivalent across skin phototypes, anatomic sites, image-acquisition methods, or clinical settings unless subgroup validation supports that conclusion.

Diagnostic and Safety Considerations

The most consequential safety failure in melanoma detection is false reassurance. A false-negative AI output can delay biopsy, referral, or follow-up. A false-positive result can increase unnecessary referrals, biopsies, costs, anxiety, and scarring. Both harms matter.

Clinicians should avoid treating AI output as a rule-out test. This is particularly important when:

• The lesion falls outside the device’s intended use
• Image quality is poor
• The anatomic site is underrepresented
• The patient has elevated risk
• The patient’s skin phototype is underrepresented in validation data
• The lesion history or clinical examination is concerning
• The AI result conflicts with the clinician’s assessment

When clinical concern and AI output conflict, clinical concern should govern referral or biopsy decisions.

Overdiagnosis also deserves attention. Rising melanoma incidence and favorable survival after early-stage diagnosis do not, by themselves, prove a screening benefit. Earlier detection may benefit an individual patient with clinically important melanoma, but screening programs may also identify biologically indolent lesions that would not have caused morbidity during the patient’s lifetime.

Clinician-facing communication should therefore distinguish lesion-directed diagnosis from population screening and should avoid implying that more detection always produces better outcomes.

Table 1. Screening, Diagnosis, and AI

Clinical scenario	Evidence-based approach and safety boundary
Average-risk, asymptomatic adult	USPSTF evidence is insufficient for routine clinician visual screening. Individualize discussion according to risk, uncertainty, access, anxiety, and patient preferences.
New or changing lesion	This is diagnostic evaluation, not screening. Examine, document, refer, use teledermatology, or biopsy when appropriate and within scope.
High-risk surveillance	Outside the average-risk screening statement. Dermatology-directed risk assessment and follow-up are often appropriate.
AI image classification	Promising in selected settings. Treat output as decision support rather than a diagnosis.
FDA-authorized adjunctive device	Confirm intended user, age, lesion type, site, patient population, clinical context, and labeling restrictions.
Negative AI output with persistent concern	False reassurance is possible. Do not defer referral, follow-up, or biopsy solely because of the AI result.

Table 2. Where AI Adjuncts May Fit

Use case	Reasonable role and limitation
Nondermatologist evaluating a suspicious lesion	May support second reading or referral decisions. Not a stand-alone rule-out test.
Teledermatology workflow	May support triage or image prioritization. Requires clinician review and a reliable follow-up pathway.
Dermatology image review	May assist workflow or selected image-based tasks. Does not replace the complete clinical assessment.
High-risk surveillance	May eventually assist longitudinal comparison. Requires validated systems and dermatology oversight.
Average-risk population screening	Net benefit is not established. No proven melanoma mortality benefit.
Darker skin, acral, subungual, mucosal, or atypical lesions	Use heightened caution. Validation gaps may be clinically important.

A Practical Approach for Nondermatology Clinicians

The first step is to separate screening from diagnosis.

An asymptomatic, average-risk patient asking about routine skin examination should receive an evidence-based discussion. The benefit of routine clinician screening remains uncertain, although individualized decisions may be reasonable based on risk, access, anxiety, and patient preference.

A patient with a concerning lesion should not be managed under the screening-uncertainty framework.

For lesion-directed care, clinicians should document:

• Duration and evolution
• Symptoms
• Previous trauma or biopsy
• Personal and family history of melanoma or other skin cancer
• Immunosuppression
• Relevant medication exposure
• Asymmetry, border, color, diameter, and evolution
• Nodularity, ulceration, bleeding, or tenderness
• Acral or subungual location
• Whether the lesion differs from the patient’s other nevi

When AI is used, clinicians should confirm that the tool is authorized or otherwise appropriately validated for the intended clinical use, patient population, lesion type, anatomic site, image type, and user group.

The result should be documented as adjunctive information. It should not be the sole basis for reassurance, referral avoidance, biopsy avoidance, or termination of follow-up.

Implementation Considerations for Health Systems

Health systems adopting AI-assisted dermatology tools should treat them as clinical decision-support technologies rather than simple consumer software.

Governance should define:

• Intended use and eligible users
• Training and competency requirements
• Documentation standards
• Referral and escalation pathways
• Image-acquisition and quality standards
• Image storage and retention
• Privacy and cybersecurity safeguards
• Management of discordant AI and clinician assessments
• Software-version control and update review
• Performance monitoring and incident reporting

Local quality monitoring is essential. Systems should evaluate false-negative and false-positive results, referral rates, biopsy rates, time to dermatology assessment, follow-up completion, and performance across age, sex, race, ethnicity, skin phototype, anatomic site, user group, and immunosuppression status.

AI deployment should not widen disparities by performing best in populations already well represented in dermatology datasets while performing less reliably in populations that already experience delayed diagnosis.

Limitations of the Evidence

The AI evidence base has grown, but important limitations remain.

Many studies are retrospective and depend on curated image databases enriched for lesions already considered suspicious. Image quality is often better than that encountered in routine care. Even prospective studies may include selected lesions rather than consecutive, unselected patients.

Binary melanoma-versus-nonmelanoma classification also oversimplifies the clinical task. Real-world diagnosis requires distinguishing among benign nevi, keratinocyte carcinomas, inflammatory lesions, rare tumors, atypical melanocytic proliferations, and lesions whose significance depends on temporal change.

Dataset representativeness remains a substantial concern. Many training and validation cohorts underrepresent darker skin phototypes, acral and subungual sites, mucosal lesions, amelanotic melanoma, rare melanoma subtypes, immunosuppressed patients, and complex nevus phenotypes. These limitations restrict generalizability and may increase diagnostic error in underrepresented groups.

Device-specific evidence should not be generalized across algorithms, hardware platforms, imaging protocols, software versions, clinical settings, or patient populations. Performance established for one tool in one pathway does not establish a class effect for all dermatology AI.

A critical evidence gap also remains between improved diagnostic metrics and improved patient outcomes. Sensitivity, specificity, accuracy, and balanced accuracy are useful measures of technical or diagnostic performance. They do not inherently show that clinical decision-making improves or that patients experience better outcomes.

Melanoma-specific mortality is a patient-centered outcome. Late-stage diagnosis rates are clinically meaningful intermediate outcomes but require interpretation alongside overdiagnosis, lead time, false-positive findings, diagnostic intensity, and downstream biopsy burden.

Prospective evidence remains limited regarding whether AI deployment shortens time to diagnosis, improves access in underserved populations, reduces unnecessary biopsies without increasing missed cancers, or produces equitable performance in routine clinical workflows.

Published studies also provide limited evidence regarding implementation factors such as user training, workflow integration, documentation, software updates, patient follow-up, accountability, and disparities in dermatology access.

Overall, diagnostic performance in selected studies is promising, but current evidence does not establish a clear net benefit from population-wide AI-assisted melanoma screening.

Future Directions

The next phase of research should move beyond simple “AI versus dermatologist” comparisons and focus on clinically consequential questions.

Future studies should assess whether AI-assisted care:

• Reduces time to diagnosis for clinically important melanoma
• Improves access to dermatologic evaluation
• Reduces unnecessary biopsies without increasing missed cancers
• Performs equitably across skin phototypes and lesion sites
• Remains safe when used by the intended clinicians in routine practice
• Improves follow-up completion and referral efficiency
• Provides benefit relative to its financial and workflow costs

Studies should be prospective, multicenter, pragmatic, and inclusive of consecutive patients. Investigators should prespecify decision thresholds, use histopathology when clinically appropriate, follow nonbiopsied lesions, report subgroup performance, document software versions, measure workflow outcomes, and evaluate potential harms.

Claims about screening benefit require patient-centered outcomes rather than diagnostic performance alone.

References

American Academy of Dermatology. (2026). Melanoma clinical guideline. Retrieved July 25, 2026, from https://www.aad.org/member/clinical-quality/guidelines/melanoma

Amaral, T., Ottaviano, M., Arance, A., Blank, C., Chiarion-Sileni, V., Donia, M., Dummer, R., Garbe, C., Gershenwald, J. E., Gogas, H., Guckenberger, M., Haanen, J., Hamid, O., Hauschild, A., Höller, C., Lebbé, C., Lee, R. J., Long, G. V., Lorigan, P., … Ascierto, P. A. (2025). Cutaneous melanoma: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. Annals of Oncology, 36(1), 10-30. https://doi.org/10.1016/j.annonc.2024.11.006
PMID: 39550033

Dinnes, J., Deeks, J. J., Chuchu, N., Ferrante di Ruffano, L., Matin, R. N., Thomson, D. R., Wong, K. Y., Aldridge, R. B., Abbott, R., Fawzy, M., Bayliss, S. E., Grainge, M. J., Takwoingi, Y., Davenport, C., Godfrey, K., O’Sullivan, C., Gulati, A., Chan, S. A., & Cochrane Skin Cancer Diagnostic Test Accuracy Group. (2018). Dermoscopy, with and without visual inspection, for diagnosing melanoma in adults. Cochrane Database of Systematic Reviews, 12(12), CD011902. https://doi.org/10.1002/14651858.CD011902.pub2
PMID: 30521682

Ferrante di Ruffano, L., Dinnes, J., Deeks, J. J., Chuchu, N., Bayliss, S. E., Davenport, C., Takwoingi, Y., Godfrey, K., O’Sullivan, C., Matin, R. N., & Cochrane Skin Cancer Diagnostic Test Accuracy Group. (2018). Computer-assisted diagnosis techniques, including dermoscopy- and spectroscopy-based systems, for diagnosing skin cancer in adults. Cochrane Database of Systematic Reviews, 12(12), CD013186. https://doi.org/10.1002/14651858.CD013186
PMID: 30521691

Henrikson, N. B., Ivlev, I., Blasi, P. R., Nguyen, M. B., Senger, C. A., Perdue, L. A., & Lin, J. S. (2023). Skin cancer screening: Updated evidence report and systematic review for the US Preventive Services Task Force. JAMA, 329(15), 1296-1307. https://doi.org/10.1001/jama.2023.3262
PMID: 37071090

Jones, O. T., Matin, R. N., van der Schaar, M., Prathivadi Bhayankaram, K., Ranmuthu, C. K. I., Islam, M. S., Behiyat, D., Boscott, R., Calanzani, N., Emery, J., Williams, H. C., & Walter, F. M. (2022). Artificial intelligence and machine learning algorithms for early detection of skin cancer in community and primary care settings: A systematic review. The Lancet Digital Health, 4(6), e466-e476. https://doi.org/10.1016/S2589-7500(22)00023-1
PMID: 35623799

Krakowski, I., Kim, J., Cai, Z. R., Daneshjou, R., Lapins, J., Eriksson, H., Lykou, A., & Linos, E. (2024). Human-AI interaction in skin cancer diagnosis: A systematic review and meta-analysis. npj Digital Medicine, 7, 78. https://doi.org/10.1038/s41746-024-01031-w
PMID: 38594408

Laiouar-Pedari, S., Kühn, A., Wies, C., Nogueira Garcia, C., Winterstein, J. T., Heinlein, L., Hoffsommer, A., Chanda, T., Haggenmüller, S., & Brinker, T. J. (2026). Prospective evidence on artificial intelligence-assisted melanoma diagnostics: A systematic review and meta-analysis. JAMA Dermatology, 162(5), 478-487. https://doi.org/10.1001/jamadermatol.2026.0217
PMID: 41879756

National Cancer Institute. (2026). SEER cancer stat facts: Melanoma of the skin. Retrieved July 25, 2026, from https://seer.cancer.gov/statfacts/html/melan.html

Papachristou, P., Söderholm, M., Pallon, J., Taloyan, M., Polesie, S., Paoli, J., Anderson, C. D., & Falk, M. (2024). Evaluation of an artificial intelligence-based decision support for the detection of cutaneous melanoma in primary care: A prospective real-life clinical trial. British Journal of Dermatology, 191(1), 125-133. https://doi.org/10.1093/bjd/ljae021
PMID: 38234043

Salinas, M. P., Sepúlveda, J., Hidalgo, L., Peirano, D., Morel, M., Uribe, P., Rotemberg, V., Briones, J., Mery, D., & Navarrete-Dechent, C. (2024). A systematic review and meta-analysis of artificial intelligence versus clinicians for skin cancer diagnosis. npj Digital Medicine, 7, 125. https://doi.org/10.1038/s41746-024-01103-x
PMID: 38744955

Swetter, S. M., Tsao, H., Bichakjian, C. K., Curiel-Lewandrowski, C., Elder, D. E., Gershenwald, J. E., Guild, V., Grant-Kels, J. M., Halpern, A. C., Johnson, T. M., Sober, A. J., Thompson, J. A., Wisco, O. J., Wyatt, S., Hu, S., & Lamina, T. (2019). Guidelines of care for the management of primary cutaneous melanoma. Journal of the American Academy of Dermatology, 80(1), 208-250. https://doi.org/10.1016/j.jaad.2018.08.055
PMID: 30392755

Tschandl, P., Rinner, C., Apalla, Z., Argenziano, G., Codella, N., Halpern, A., Janda, M., Lallas, A., Longo, C., Malvehy, J., Paoli, J., Puig, S., Rosendahl, C., Soyer, H. P., Zalaudek, I., & Kittler, H. (2020). Human-computer collaboration for skin cancer recognition. Nature Medicine, 26(8), 1229-1234. https://doi.org/10.1038/s41591-020-0942-0
PMID: 32572267

U.S. Food and Drug Administration. (2024). De Novo classification request for DermaSensor, DEN230008: Classification order. Retrieved July 25, 2026, from https://www.accessdata.fda.gov/cdrh_docs/pdf23/DEN230008.pdf

U.S. Food and Drug Administration. (2024). De Novo summary, DEN230008: DermaSensor. Retrieved July 25, 2026, from https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN230008.pdf

U.S. Preventive Services Task Force. (2023). Screening for skin cancer: US Preventive Services Task Force recommendation statement. JAMA, 329(15), 1290-1295. https://doi.org/10.1001/jama.2023.4342
PMID: 37071089