As COVID-19 strains healthcare systems around the world, other diseases aren’t going away. As Gabriella Antici points out, “cancer doesn’t stay in quarantine.”
Antici, the founder of the Protea Institute, will share her experience implementing a machine-learning model to detect breast cancer in Brazil at the panel “Open Problems in COVID Patient Care and New Opportunities for AI Solutions,” part of the AI Cures Conference on September 29th. The Protea Institute is a non-governmental organization in Brazil that provides access to screenings and breast cancer treatment for underserved women.
Breast cancer is eminently treatable, but the five-year mortality rate in Brazil is about 25%, much higher than in other countries. (In the United States, the five-year mortality rate ranges between 5 and 10%.)
As a two-time breast cancer survivor herself, Antici was motivated to form Protea after realizing many women in Brazil lacked access to expeditious treatment. “Not only are they diagnosed later...they wait sometimes months, sometimes even a year to start treatment,” she says.
Three-quarters of women in Brazil rely on Sistema Único de Saúde (SUS), the public health system in Brazil, for care. The system relies on philanthropic and private hospitals, which typically put a cap on patients from the public system, making it difficult or impossible for women to avoid waiting. Last year, the Protea Institute fundraised 2.4 million Reais, the equivalent of almost half a million USD, to begin filling the shortfall in hospitals’ budgets.
In 2019, The Protea Institute partnered with MIT Professor and Jameel Clinic AI lead Regina Barzilay and Dr. Constance Lehman from MGH to select hospitals to introduce a machine-learning model that detects breast cancer. Initially, Barzilay trained the model using tens of thousands of mammograms and the known outcomes of patients from Massachusetts General Hospital. This phase of the project finetuned the algorithm, which identifies in the images otherwise inscrutable patterns that correlate with cancer.
Before implementing the model, it needed to be validated in a population that matched the demographics of Brazil. One hurdle was technology, many hospitals in Brazil use printed mammograms, which don’t work for the model. They ended up partnering with two institutions, a diagnostic imaging center and a hospital in a neighborhood several hours from São Paulo, to begin validating.
Their work is promising, and they hope to validate the model in Brazil in the coming months.
Two years ago, the Protea Institute began making an impact by covering costs for women admitted to Hospital Santa Marcelina, also in the São Paulo region, where 90% of patients receive treatment through the public system. The hospital can now accept 25% more breast cancer patients each month.
Between biopsies, treatments and breast exams, “we impacted 1,417 women last year, in our first year of operation,” Antici says. But their work is still in the early stages. “The problem is so enormous in Brazil that we have a long, long way to go,” she added.
Antici would like to expand to other regions in Brazil, forming partnerships with more hospitals. She envisions taking the program to neighboring countries in South America.
In the future, if a machine-learning model helps diagnose patients sooner, it will not only save those lives, it may make treatment less costly, stretching funds to serve even more women.