Diagnosing contact dermatitis using machine learning: A review

Background: Machine learning (ML) offers an opportunity in contact dermatitis (CD) research, where with full clinical picture, may support diagnosis and patch test accuracy.

Objective: This review aims to summarise the existing literature on how ML can be applied to CD in its entirety.

Methods: Embase, Medline, IEEE Xplore, and ACM Digital Library were searched from inception to February 7, 2024, for primary literature reporting on ML models in CD.

Results: 7834 articles were identified in the search, with 110 moving to full-text review, and six articles included. Two used ML to identify key biomarkers to help distinguish between allergic contact dermatitis (ACD) and irritant contact dermatitis (ICD), three used image data to distinguish between ACD and ICD, and one used clinical and demographical data to predict the risk of positive patch tests. All studies used supervision in their ML model training with a total of 49 704 patients across all data sets. There was sparse reporting of the accuracy of these models.

Conclusions: Although the available research is still limited, there is evidence to suggest that ML has potential to support diagnostic outcomes in a clinical setting. Further research on the use of ML in clinical practice is recommended.