Background: Malassezia yeasts are almost universally present on human skin worldwide. While they can cause diseases such as pityriasis versicolor, their implication in skin homeostasis and pathophysiology of other dermatoses is still unclear. Their analysis using native microscopy of skin tape strips is operator dependent and requires skill, training and significant amounts of hands-on time.

Objectives And Methods: To standardise and improve the speed and quality of diagnosis of Malassezia in skin tape strip samples, we sought to create an artificial intelligence-based algorithm for this image classification task. Three algorithms, each using different internal architectures, were trained and validated on a manually annotated dataset of 1113 images from 22 samples.

Results: The Vision Transformer-based algorithm performed the best with a validation accuracy of 94%, sensitivity of 94.0% and specificity of 93.5%. Visualisations providing insight into the reasoning of the algorithm were presented and discussed.

Conclusion: Our image classifier achieved very good performance in the diagnosis of the presence of Malassezia yeasts in tape strip samples of human skin and can therefore improve the speed and quality of, and access to this diagnostic test. By expanding data sources and explainability, the algorithm could also provide teaching points for more novice operators in future.

Download full-text PDF

Source
http://dx.doi.org/10.1111/myc.13777DOI Listing

Publication Analysis

Top Keywords

malassezia yeasts
12
tape strip
12
strip samples
12
artificial intelligence-based
8
presence malassezia
8
yeasts tape
8
human skin
8
skin tape
8
improve speed
8
speed quality
8

Similar Publications

Want AI Summaries of new PubMed Abstracts delivered to your In-box?

Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!