A smartphone mounted on a microscope and AI: in Chiang Mai, vaginal infections were diagnosed correctly in 100 per cent of the samples tested

The RT-DETR detector achieves 98.25 per cent accuracy per image on Gram-stained smears photographed using the Galaxy S23 Ultra. The authors call for further multicentre validation.

A team from the University of Chiang Mai, Thailand, has published in Frontiers in Artificial Intelligence a system that recognizes the main vaginal infections from photos of slides taken with a smartphone attached to an ordinary light microscope. The equipment is deliberately banal: a Samsung Galaxy S23 Ultra, a Gram stain and a microscope like those found in most laboratories.

Rather than classifying the entire image, the models identify characteristic structures (yeasts, bacteria, parasites). Three detection architectures were compared on 1,966 images with 11,069 annotated objects, divided between candidiasis, bacterial vaginosis, trichomoniasis, gonorrhea and healthy flora. RT-DETR obtained 98.25% correct responses per image and made the correct diagnosis on all 96 test samples when aggregating several fields; YOLOv26 achieved 95.72%.

The interest is to provide a rapid diagnosis where biologists are lacking. However, the authors list the limitations: a single institution, standardized imaging conditions, no mixed infection in the data. Prospective validation over several centers remains necessary before any autonomous clinical use.