Artificial Intelligence in the Diagnosis of Superficial Mycoses: A Scoping Review

Artificial intelligence (AI) is increasingly proposed for diagnosing superficial mycoses, yet whether this literature covers the diagnostic pathway evenly, and is mature enough for clinical translation, remains unclear. We mapped AI applications onto an operational partition of the diagnostic pathway (D1–D4), a construct of this review: pre-analytical clinician-facing imaging (D1), analytical microscopy and histopathology (D2), analytical molecular identification (D3) and integrative post-analytical interpretation (D4). Thirty-three primary studies (2007 to 2026) were included, 85% published since 2021 and predominantly from Asia. Applications concentrated on D2 (n = 16) and D1 (n = 13), with few in D3 (n = 4) and none in D4. Reported accuracy was frequently high but rested on internal, single-centre validation: complete external validation and prospective preregistration were each present in a single study, only a handful shared both public code and public data, about half reported no demographic data and no D1 study stratified performance by Fitzpatrick phototype. A substantial minority carried conflicts of interest tied to the evaluated tool, concentrated in the molecular domain where independent verification was scarcest; where such re-evaluation was possible, reported accuracy fell markedly (one area under the curve (AUC) from 0.98 to 0.75). This corpus is skewed toward the early diagnostic steps, leaves the integrative post-analytical step unaddressed and shows methodological maturity insufficient for routine clinical translation. Institutionalised external validation, reproducibility through open code and data, declaration of dataset overlap and conflicts of interest, and clinically meaningful outcomes are the priorities for future work.