Mayo Clinic researchers and collaborators are training an AI tool to read smartphone photos of patch test reactions, and early data suggests it can come close to dermatologist-level accuracy. The hope is to cut down on the repeat clinic visits that allergic contact dermatitis often requires and to give people with stubborn rashes or limited access to dermatology another way to get answers.
In a paper that trained the model on roughly 28,000 expert-graded patch test images collected over about a decade, the team reported that a deep learning approach could classify reactions from photographs using a retrospective dataset, according to Mayo Clinic Proceedings: Digital Health. That work provided the imaging backbone for the pipeline researchers are now testing in real patients.
Those follow-up results come from a single-arm clinical trial of 206 participants. Investigators reported that the algorithm showed high discrimination (area under the receiver operating characteristic curve 0.86) and strong specificity (about 93 percent) but lower sensitivity (around 58 percent), according to a trial published in Dermatitis. Human readers interpreting the same photographs produced variable results, and the study found that combining multiple human interpretations sometimes outperformed the model on its own.
How an at-home test would work
In the proposed setup, patients would apply standard patch strips themselves, remove them on a set schedule, then snap photos of the test areas with a smartphone. The AI would flag likely allergic reactions for a clinician to review, Alison Bruce of the Mayo Clinic explained in a script from the Mayo Clinic News Network. The team is working to make the system more forgiving of real-world photo issues such as off-center framing, tricky lighting and varying resolution so that ordinary phone pictures can feed the algorithm reliably.
Limitations and next steps
The trial authors caution that the algorithm’s lower sensitivity and the practical challenges of capturing clear images mean the tool still needs broader validation and fine-tuning before it can be used for at-home testing, according to Dermatitis. Broader reviews of AI in dermatology highlight the need to test systems across diverse skin tones and everyday phone and lighting conditions, and point to regulatory and clinical-integration hurdles that will have to be addressed, as discussed in a review on ScienceDirect.
Patient reaction
Participants in early testing generally found the process straightforward and liked the idea of skipping extra in-person appointments. “Having AI as a potential to read these results, I think that’s really cool and innovative,” pharmacist participant Natalie Thackerdin told Boston 25 News, while another participant described the AI read as “pretty convenient.”
