The design and application of an automated microscope developed based on deep learning for fungal detection in dermatology.

BACKGROUND: Light microscopy to study the infection of fungi in skin specimens is time-consuming and requires automation. OBJECTIVE: We aimed to design and explore the application of an automated microscope for fungal detection in skin specimens. METHODS: An automated microscope was designed, and a deep learning model was selected. Skin, nail and hair samples were collected. The sensitivity and the specificity of the automated microscope for fungal detection were calculated by taking the results of human inspectors as the gold standard. RESULTS: An automated microscope was built, and an image processing model based on the ResNet-50 was trained. A total of 292 samples were collected including 236 skin samples, 50 nail samples and six hair samples. The sensitivities of the automated microscope for fungal detection in skin, nails and hair were 99.5%, 95.2% and 60%, respectively, and the specificities were 91.4%, 100% and 100%, respectively. CONCLUSION: The automated microscope we developed is as skilful as human inspectors for fungal detection in skin and nail samples; however, its performance in hair samples needs to be improved.