Artificial Intelligence in Dermatology: Exploring New Diagnostic Strategies

With artificial intelligence (AI) technologies being integrated into clinical practice, studies have been conducted on how they can benefit dermatologists. Recent research has demonstrated that AI tools are capable of classifying skin lesions and detecting malignancies like melanoma, but questions remain about the accuracy and reliability of these tools in clinical practice.

A Patient-Focused Approach to AI Training

While traditional clinical examination often involves evaluating individual risk factors, phenotypic characteristics, and the patient’s overall lesion pattern, early AI models have not been trained with this context in mind. Their ability to analyze individual skin lesions has shown promise, but the lack of patient-specific context has raised concerns.

A recent study in the Journal of the European Academy of Dermatology and Venereology aimed to address this issue. Specifically, the researchers evaluated whether including multiple images from the same patient would enhance their ability to detect malignant melanoma.

As part of the 2020 SIIM-ISIC Melanoma AI Classification Challenge, thousands of AI algorithms were tested to determine their accuracy in discriminating between melanoma and benign lesions. The diagnoses from the 50 most accurate algorithms were compared to diagnoses from 176 dermatology professionals.

The study found that the top 50 algorithms outperformed the clinicians in accurately diagnosing the lesions. Notably, providing contextual images for individual patients did not improve the accuracy of either the AI algorithms or the clinicians. In fact, the addition of contextual images negatively impacted the dermatologists’ ability to diagnose benign lesions.

Future Directions

While AI cannot fully replicate a patient-focused approach to diagnosis, it has the potential to support us by identifying suspicious lesions early and prompting further examination. Further research is needed to better understand the value of incorporating patient information into AI models.

As AI tools become more common in clinical practice, concerns about overreliance, misdiagnosis, liability, and bias arise. While some clinicians are comfortable and familiar with AI technologies, many others still have unanswered questions about their limitations.

Lastly, while research on the capabilities of these tools continues to grow, implementation into clinical practice in dermatology is rare. Determining how these technologies can be applied to real-world scenarios is a key question that remains unanswered. However, recent findings suggest that they can enhance the future of dermatology practice through their accuracy and efficiency.

To learn more about how we can apply AI to dermatological practice, check out Dr. Veronica Rotemberg’s sessions at the 2025 Dermatology Foundation Clinical Symposium.

References:

Kurtansky, NR, Primiero, CA, Betz-Stablein, B, et al. Effect of patient-contextual skin images in human- and artificial intelligence-based diagnosis of melanoma: Results from the 2020 SIIM-ISIC melanoma classification challenge. Journal of the European Academy of Dermatology and Venereology. 2024;00:1-11.doi.org/10.1111/jdv.20479

Li, Y, & Rotemberg, V. From promise to practice: Artificial intelligence in skin cancer screenings. Journal of the European Academy of Dermatology and Venereology. 2024;38(12):2203-2204. doi.org/10.1111/jdv.20376