Using semantic indexing and thesauri in processing dermoscopic images

This article presents an innovative algorithm for automatic semantic indexing of dermoscopic images using clinical metadata (age, gender, anatomical location). The proposed approach integrates advanced natural language processing (NLP) techniques with medical thesauri and ontologies to enhance the efficiency of medical data analysis and structuring. The BioBERT neural language model was employed to generate clinical descriptions for each image, transforming natural language expressions into a vector space. These vector representations were then compared with selected semantic concepts from medical thesauri (e.g., SNOMED CT) using cosine similarity, enabling the automatic selection of the most relevant semantic tag. In the experimental part of the study, automatic semantic labeling was performed on 100 images from the ISIC 2020 GroundTruth dataset. Key concepts reflecting visual features of skin tumors. The results demonstrated a high level of agreement between automatically assigned tags and actual diagnoses. This approach significantly accelerates and standardizes the image labeling process, reduces subjectivity, and improves the reproducibility of results. The developed algorithm can be integrated into dermatology information systems to automate diagnostics, enable intelligent data sorting, and support the development of recommendation systems.