Artificial Intelligence-Assisted Diagnosis of Fungal Infections: Advances in Imaging and Beyond

Abstract

Fungal infections are a major and growing global health burden that cause an estimated 6.5 million invasive infections and approximately 3.8 million deaths annually, although they continue to be substantially underdiagnosed. Conventional diagnostic methods including microscopy, culture, serological biomarkers and molecular assays are limited by delayed turnaround times, variable sensitivity, operator dependence, and restricted accessibility. Artificial intelligence (AI) and machine learning (ML) are promising approaches for the improvement of fungal diagnostics through automated image analysis, the interpretation of spectral and genomic data, and the integration of multimodal clinical information. This narrative review summarizes recent advances in AI-assisted fungal diagnosis across key domains, including microscopic and culture-plate image analysis, computed tomography radiomics for the examination of pulmonary mycoses, ophthalmic and dermatological imaging, matrix-assisted laser desorption/ionization timeof-flight mass spectrometry, single-cell Raman spectroscopy, electronic-nose biosensors, metagenomic nextgeneration sequencing, and the prediction of antifungal susceptibility. The performance of these approaches has been encouraging: microscopy-based systems have achieved sensitivities of approximately 95-96% in selected studies, computed tomography (CT) radiomics and deep-learning models have demonstrated areas under the curve of 0.808-0.92 for pulmonary fungal infection differentiation, Raman spectroscopy has enabled rapid specieslevel identification, and electronic-nose platforms have achieved accuracies exceeding 95% in the classification of Candida species. Multimodal AI systems have also shown improved performance in the context of complementary clinical and imaging information. Despite these advances, clinical implementation remains limited by the small and geographically biased size of the relevant datasets, inadequate external validation, limited model interpretability, and evolving regulatory frameworks. Future progress will depend on large multicenter datasets, explainable AI methodologies, and prospective clinical validation to facilitate the integration of AI-assisted fungal diagnostics in routine practice