Noninvasive Smartphone Identification of Atopic Dermatitis Using Autofluorescence Imaging and Machine Learning

The skin microbiome is a diverse environment of bacteria, fungi, viruses, and other microorganisms that inhabit the skin, which plays a crucial role in the body's defense against foreign invaders, in thermoregulation, and is involved with both endocrine and exocrine activity. Usually, these microorganisms have a symbiotic relationship with the body, but dysbiosis can lead to diseases like atopic dermatitis (eczema), often linked to an excess of Staphylococcus aureus. This work aims to create an alternative to culture-based or gene sequencing methods for detecting and monitoring eczema. It is based on multiple autofluorescence images of the bacteria species populated in the microbiome obtained with a portable, inexpensive, smartphonebased imaging platform. The device consists of a smartphone, a pocket microscope attachment, LEDs for excitation, and acrylic films for filtering fluorescent emissions. The collected images are then analyzed through a convolutional neural network (CNN) model, which was able to classify bacteria mixtures of typical skin microbiota with an accuracy of 84 %, while the human subject models had an average accuracy of 59 % with the best-performing split showing 7 1% accuracy.