CUTIS-AI: the Brazilian artificial intelligence built for skin diseases the world tends to forget

⏱️ Reading time: ~5 minutes

Introduction

When people talk about artificial intelligence in dermatology, the usual example is skin cancer detection in wealthy countries. In November 2026, Brazil presented a different path: CUTIS-AI, an app developed by Hospital Israelita Albert Einstein together with Brazil's Ministry of Health to help public health workers identify infectious and neglected skin diseases — conditions that mostly affect poor populations and rarely appear in the datasets used to train AI systems.

The context: who built it and for whom

For readers outside Brazil, two names help explain the project.

SUS (Sistema Único de Saúde) is Brazil's public, universal health system, free for the entire population of more than 200 million people. Its entry point is primary care — health posts and basic health units, many of them in small towns and remote areas, where care is provided by general practitioners, nurses and community health workers, with no dermatologist nearby.

Hospital Israelita Albert Einstein, in São Paulo, is a private, non-profit institution and a reference in medicine and research in Latin America, running projects in partnership with the public system.

CUTIS-AI connects both ends: technology developed at a research center, built to be used in basic care across the public network.

What CUTIS-AI does

The app supports the screening and identification of skin diseases. It combines a photo of the lesion with the patient's clinical, epidemiological and sociodemographic information, adapted to the reality of the Brazilian population. One detail matters for use in the country's interior: it works without an internet connection.

The diseases covered are:

Leprosy (Hansen's disease)
Cutaneous leishmaniasis
Sporotrichosis
Cutaneous tuberculosis
Chromoblastomycosis
Lobomycosis
Brown recluse spider bite lesions (loxoscelism)

Several of them belong to the group the World Health Organization classifies as neglected tropical diseases. Leprosy is the best-known example: Brazil has the second-highest number of new cases in the world, behind only India, according to the WHO. Others, such as lobomycosis, are typical of the Amazon region and almost unknown outside Latin America. Early diagnosis is decisive, because some of these diseases can leave lasting damage and disabilities when treated late.

The data behind the model

The system was trained on about 1,400 patients and more than 4,000 images of lesions. The Ministry of Health highlights the diversity of the dataset: different skin tones, regions and epidemiological profiles.

That point carries international weight. AI tools in dermatology have historically been trained largely on images of light skin, which can reduce their performance on people with darker skin. For Fabiano Geraldo Pimenta Junior, the Ministry's secretary of Health and Environmental Surveillance, "data quality is fundamental," and a dataset covering different skin types, regions and living conditions is essential for the system to work across diverse groups.

Data scientist Isabelle Carvalho, from Einstein, said the goal is to expand the ability to identify diseases that challenge health professionals.

Performance and current stage

According to the Ministry, CUTIS-AI has an accuracy rate of about 80%. The information released does not detail which metric was used or cite a published scientific study.

The tool is still in the study phase. No timeline for reaching health units and no investment figure have been disclosed. The Ministry stresses that the app will act exclusively as support for medical decisions, without replacing the clinical evaluation carried out by health professionals.

What is still unknown

When and in which regions the app will begin to be used in the public network.
How large-scale validation will be carried out, with new patients and in different regions.
Technical details of its performance, such as accuracy per disease and false positive rates.
Whether there will be a scientific publication, or whether the tool will be made available to other countries with similar diseases.

Conclusion

CUTIS-AI shows an application of artificial intelligence designed around a local problem: bringing diagnostic support for skin diseases that affect vulnerable populations, in places without specialists, using a dataset built on the diversity of the country's own population.

For Brazil, it is a tool aimed at primary care within SUS. For readers in other countries — especially tropical regions of Latin America, Africa and Asia that live with the same diseases — it is a case to watch: an AI system trained on the diseases and skin tones that usually get left out of large datasets.