Translated from portuguese. Find original here.
By Thiago Sawada
Between January and May, Brazil faces the period of the year with the highest incidence of dengue cases. Now, the concern of health authorities is even greater since the mosquito ‘Aedes aegypti’ also carries the virus that causes chikungunya disease and zika. While there is no vaccine or specific treatment, prevention is done through the fight against mosquitoes.
This year, a new technology imported from Malaysia can help in this difficult task. It is an artificial intelligence based algorithm developed by startup Aime. He is able to predict, with up to three months in advance, the places where there is a greater incidence of the disease.
The new technology is beginning to be tested in January by the São Paulo state
government. The startup was chosen for the project through the Pitch Gov program selected in November, 15 startups with ideas to improve public services. Through an agreement, the government will test the new technology and if it help fight the dengue mosquito in the coming months, Aime may be hired.
Aime (Artificial Intelligence in Medical Epidemiology, its acronym in English) was born earlier this year, during the course of innovation Singularity University, located within the university campus of NASA in Silicon Valley. Visiting the campus, the NGO Viva Rio has partnered with the startup to put the solution into practice in Brazil. “The Aime comes with technology and people with the field operation to generate social impact,” says innovation coordinator of Viva Rio, Francisco Araujo.
It will be the first time the Aime will apply its technology outside of Malaysia. In tests carried out in two provinces, the system could correctly predict 88% of the places where the disease is manifested. For the technology to work here, the startup had to adjust its algorithm to Brazilian data. For this, the founders spent three weeks in Rio de Janeiro to raise the necessary information together with the Viva Rio.
The algorithm analyzes data that are already collected by local government and by satellite image recognition systems, allowing the technology to be employed in any city at a low cost.
In Rio de Janeiro, the startup team tested the system using prefecture of data covering the period from 2007 to 2013. Through machine learning, the algorithm identified patterns among the variables and localized disease outbreaks each year. After understanding the correlations, the system generates a map with dengue focus predictions can be compared to the occurrence recorded by region.
According to the Malaysian epidemiologist and founder of startup, Dhesi Raja, were not required many changes in the program, because the climate and ecosystem Brazil are similar to Malaysia. “At first we thought it would be difficult, but the experience in Brazil was wonderful because we achieved an accuracy of 84% in the diagnosis,” he says. The algorithm analyzed data from an area of 63.7 square kilometers, which registered the highest number of dengue cases this year in Rio de Janeiro.
Variables. To achieve this result, we need to analyze large data sets. The program takes into account all influences the mosquito flight, such as weather, wind speed and direction, radiation and rainfall index. Other factors have a more direct relationship with the mosquito breeding, such as proximity to lakes, forests, works and clean, standing water accumulation sites. To transmit dengue, Aedes aegypti need to poke a person infected with the virus, so the software also analyzes the health history of the people living in the area, population
density and to the income level of the population.
On the map, the system shows potential outbreaks and a radius of 400 meters around them, where there is a probability of 65% of the rise of dengue cases the size of the proposed area follows the recommendations of the World Health Organization (WHO ). In practice, the definition of risk areas allows the most effective use of antidengue mechanisms such as larvicides, “fogging” and genetically modified mosquitoes.
Currently, the Ministry of Health has no precise method for identifying regions where the mosquito will proliferate. The Index Rapid Assessment for Aedes aegypti (LIRAa) released by the agency in November, revealed that 199 municipalities are in dengue outbreak risk, chikungunya and zika. In 2015, the number of people infected by dengue increased 176% compared to last year, reaching 1.5 million reported cases.
The goal of Aime and Viva Rio is to provide governments a kind of epidemic prevention advice. From the diagnostic algorithm, the startup help devise strategies to solve the problem in partnership with public health authorities. In addition to São Paulo, there is the expectation that the city of Rio also adopt the new technology because of the proximity of the start of the Olympic Games in 2016.
For now, the algorithm only identifies outbreaks of dengue, which according to the WHO puts at risk the health of about 2.5 billion people worldwide. According to Aime, however, the technology can be adjusted soon to predict cases of chikungunya and zika. In the future, the goal is to go even further and improve the technology to predict outbreaks of tuberculosis, malaria, ebola, flu and even Aids. “Our goal is to be the Google of prevention of infectious diseases,” says Raja.
What is artificial intelligence?
The Artificial Intelligence (AI) seeks to simulate, on computers, the human capacity to reason, to perceive the environment, recognize parameters and making decisions. There are several strands of study on intelligent systems, each dedicated to a specific aspect of human behavior. From the advancement of technology, there are machines capable of “thinking” creatively, make use of the language and to learn. The latter case refers to a subfield of AI known as a machine learning. The algorithms developed based on the technology become able to optimize its performance based on experiences, observations and past analyses.
In the case of the system developed by Aime, the algorithm analyzes a large set of data and search patterns that will determine the location of dengue points. In case of forecasting failures, the system can adjust its way of thinking for increasing accuracy results in future analytics.