AI in Dermatology: Accuracy & Cost-Effectiveness of Skin Cancer Screening Tools

The increasing incidence of skin cancer, coupled with a high volume of referrals for suspected lesions, is placing a significant strain on dermatology services. Recent research explores the potential of artificial intelligence (AI) systems as diagnostic screening tools, specifically evaluating two technologies: DERM (Deep Ensemble for Recognition of Malignancy) and MoleAnalyzer Pro (FotoFinder). The study, published in Health technology assessment, investigated the diagnostic accuracy of these tools, their potential clinical impact, and economic viability within the healthcare workflow, as an aid to decision-making after primary care referral.

Methodology and Findings

Researchers conducted a systematic review, including a narrative synthesis and meta-analysis of diagnostic accuracy studies, alongside a cost-effectiveness conceptual assessment. The review aimed not only to compare diagnostic performance but also to understand how these technologies could alter the care pathway, particularly by reducing unnecessary referrals to specialists.

The results showed that the DERM system demonstrated high sensitivity for detecting malignant lesions (approximately 96%), although with moderate specificity. The algorithm also showed reasonable performance in identifying benign lesions, suggesting its potential as an initial screening tool to reduce in-person evaluations by dermatologists. Some modeling within the study indicated that roughly half of referred patients could be cleared without specialist consultation, albeit with a risk of missing a small number of malignant tumors.

MoleAnalyzer Pro exhibited lower overall sensitivity but higher specificity for melanoma compared to remote clinical assessments by dermatologists, suggesting complementary utility in specific scenarios. However, the available evidence base for this technology was considered limited, with a small number of studies and significant methodological heterogeneity.

Diagnostic Accuracy Isn’t the Whole Picture

One of the most pertinent aspects discussed by the authors is that isolated diagnostic accuracy doesn’t necessarily translate to real clinical benefit. It remains uncertain how the introduction of AI could impact outcomes such as time to diagnosis, detected tumor stages, or patient safety in real-world practice. Interviews and acceptability assessments revealed significant resistance from both patients and healthcare professionals to the autonomous apply of AI without medical validation, highlighting important cultural and ethical barriers.

From an economic perspective, robust cost-effectiveness studies specific to the evaluated technologies were not identified. Existing models presented structural limitations, particularly in how diagnostic accuracy was translated into long-term clinical and economic benefits.

The study reinforces that AI holds promise for optimizing diagnostic workflows in dermatology, particularly in identifying benign lesions and streamlining referrals. However, the field remains in a phase of scientific maturation, with key gaps including the lack of robust comparative studies, a scarcity of prospective data in real-world settings, and uncertainties regarding clinical and economic impact at scale.

Implications for Clinical Practice

For physicians and residents, the central message is that diagnostic algorithms represent promising decision-support tools, but they do not yet replace specialized clinical judgment. Their adoption should be guided by solid evidence of patient benefit. This isn’t about replacing dermatologists, but about equipping them with tools to manage increasing caseloads and prioritize patients who truly need specialized care.

Skin cancer is a significant public health concern. According to the Brazilian Ministry of Health, skin cancer is the most frequent type of cancer in Brazil, and worldwide. Even as more common in individuals over 40, it is rare in children and people with darker skin tones. The primary cause is excessive exposure to the sun. There are two main types: melanoma, which originates in melanin-producing cells and is more common in white adults, and non-melanoma, which is more prevalent in Brazil, accounting for 30% of all malignant tumors registered in the country.

The potential of AI to improve early detection is particularly relevant given that the prognosis for skin cancer is generally good when detected in its initial stages. Research published in Nature demonstrates that AI can achieve diagnostic accuracy comparable to board-certified dermatologists in classifying skin lesions. This suggests a future where AI-powered tools could be used to triage patients, helping to ensure that those with the highest risk receive prompt attention.

What’s Next: Refining AI’s Role in Dermatology

The path forward involves several key steps. Further research is needed to address the limitations identified in the current study, including the need for larger, more diverse datasets and prospective trials in real-world clinical settings. It’s also crucial to develop standardized protocols for integrating AI tools into existing workflows and to establish clear guidelines for their use.

Beyond technical improvements, addressing the ethical and cultural barriers to AI adoption is essential. This requires open communication with patients and healthcare professionals, as well as ongoing education about the benefits and limitations of these technologies.

Finally, ongoing surveillance and evaluation are needed to monitor the impact of AI on patient outcomes and to ensure that these tools are used safely and effectively. The Brazilian National Cancer Institute (INCA) estimates 8,450 new cases of melanoma will be diagnosed in Brazil in 2020, highlighting the urgent need for innovative approaches to improve early detection and treatment.

the goal is not to replace dermatologists with AI, but to empower them with tools that can enhance their diagnostic capabilities, improve patient care, and address the growing burden of skin cancer.