Advancing diagnostic precision in dermatology: A new standardized lexicon for skin neoplasms

We read with great interest the article by Scope et al.¹ The study, performed by experts from the International Skin Imaging Collaboration (ISIC), addresses a critical need in dermatology: the development of a standardized terminology for skin neoplasms. As diagnostic challenges increase with advances in artificial intelligence (AI) and molecular pathology, a common lexicon is essential for clinical communication, research and AI model training. By using a modified Delphi consensus approach, the authors have created a comprehensive, hierarchically organized system of diagnostic terms for skin neoplasms, which offers substantial implications for clinical practice and future AI applications.

Historically, dermatology has lacked a unified terminology for skin neoplasms, which complicates diagnoses, especially when benign, malignant and indeterminate lesions share overlapping features. With the increasing use of AI in dermatology, precise and consistent terminology is more important than ever. Structured data is essential for training AI algorithms, and imprecise terms could hinder their effectiveness. A standardized lexicon enhances clinical communication, facilitates research and underpins the development of accurate AI diagnostic tools.^{2, 3}

The authors employed a modified Delphi process, gathering input from 18 experts across three rounds to refine a comprehensive set of proposed terms: during this process, the authors could suggest modifying, deleting or adding terms. This iterative approach ensures broad agreement and flexibility in incorporating expert insights. The hierarchical mapping of terms into 3 super-categories (i.e. ‘benign’, ‘malignant’ and ‘indeterminate’) and cellular/tissue-differentiation categories (e.g. ‘melanocytic’ and ‘keratinocytic’) increases the utility of the system for clinical and research settings, providing a framework for AI systems and clinical decision support.

Overall, 94% of the 379 proposed terms reached agreement in the first round, which demonstrates the reliability of the process. Most terms requiring further refinement belonged to the ‘indeterminate’ super-category (which displayed by far the lower agreement among the experts), signalling the complexity of certain diagnoses and the need for continued refinement. Importantly, this process underscores the need for a dynamic, adaptable system that can evolve alongside new scientific findings and clinical practices.⁴ The final taxonomy includes 362 terms, mapped to the 3 super-categories and 41 cellular/tissue-differentiation categories. The structure offers a comprehensive classification of skin neoplasms, ranging from benign conditions like seborrheic keratosis to malignant ones such as melanoma. We feel that one of the advantages of the study was the use of the ‘intermediate’ super-category, contrary to many previous investigations on skin neoplasm diagnosis that employed a simpler dichotomic classification (‘benign versus malignant’).⁵

A key strength of this work is its potential to inform AI-based diagnostic systems. AI models require large, annotated datasets to learn and improve. The standardized taxonomy developed here can serve as a reference point for training AI algorithms, ensuring they operate on a consistent framework. Incorporating these terms into training datasets can improve the accuracy of AI models in identifying and classifying skin neoplasms, ultimately enhancing diagnostic precision and patient outcomes.^{2, 4} Furthermore, the hierarchical nature of the taxonomy will support AI systems in making more nuanced diagnostic decisions, particularly in complex or ambiguous cases.

In conclusion, the creation of a standardized taxonomy for skin neoplasms is an important milestone for dermatology, with wide-reaching applications in clinical practice, research and AI development. The ISIC's consensus-driven approach provides a structured, expert-backed framework that addresses the need for consistent terminology, while paving the way for more accurate diagnostic tools. Though further validation and periodic updates are necessary, this lexicon is poised to streamline communication, support AI innovations and enhance global collaboration in dermatologic care. We, therefore, congratulate the authors for this brilliant effort, which will be undoubtfully beneficial to the whole dermatologic community in the future.

None.

None to declare.