Einstein and the Health Ministry Test CUTIS-AI, an Offline Tool to Help SUS Teams Spot Neglected Skin Diseases
Hospital Israelita Albert Einstein and Brazil’s Ministry of Health are testing an artificial intelligence tool meant to help health workers in the public system, SUS, identify neglected skin diseases, Estadão reported on October 30. The tool, called CUTIS-AI, was developed through Proadi-SUS, the programme under which leading philanthropic hospitals carry out projects for the public network.
CUTIS-AI combines photos of a lesion with clinical information. One model looks at colour, shape and texture in the image, while another weighs the answers to a short questionnaire, and the results are combined to support the professional. It covers suspected sporotrichosis, cutaneous leishmaniasis, leprosy, chromoblastomycosis, lobomycosis and cutaneous tuberculosis, as well as brown spider bites. It was built from more than 4,000 images of lesions from 1,448 patients at 10 reference centres in Brazil’s five regions and reached 80% accuracy in evaluations.
The original questionnaire had 254 questions, which the team cut to the 12 that best separate the seven conditions, so the tool can fit into a routine consultation. It works offline, which matters where these diseases are common: 19.1% of the participants whose photos trained the model had no treated water at home. Lívia Maria de Oliveira, Einstein’s innovation manager for digital products, told Estadão the tool is meant to support the investigation, not to close the diagnosis.
What it means in practice
Francisco Edilson, who directs the ministry’s communicable diseases department, said the tool prioritises socially determined diseases and the most vulnerable populations, in line with the SUS principle of equity. It joins other efforts to put AI to work on public health data, as when Unifesp used AI on SUS hospital records to map domestic violence against women.
Accuracy per skin tone has not yet been measured, and the team says the model has not seen every condition. Building the dataset from Brazilian patients rather than importing one from the Global North echoes the push for home-grown AI that Lula made a theme of his re-election, and the anonymised dataset is due to be published for other researchers.
What to watch next is the result of field testing at partner centres in October and November, the planned transfer to the ministry in December, how referral rules are set so a suspected case leads to treatment, and whether regulators clear the tool for SUS use in 2027.