Harvard computer scientists are using publicly available satellite data and AI to predict nutrient deficiencies, a process that could prove a real breakthrough in treating malnutrition.
In a study published by the Association for the Advancement of Artificial Intelligence, computer scientist Elizabeth Bondi and colleagues at Harvard University used satellite data, such as forest cover, weather, and presence of water, to predict deficiency of micronutrients, including iron, Vitamin B12, and Vitamin A, directly from their biomarkers. The researchers used real-world data collected from four different regions across Madagascar. They commented, “Our method could be broadly applied to other countries where satellite data are available, and potentially create high societal impact if these predictions are used by policy makers, public health officials, or healthcare providers.”
More than 2 billion people worldwide, including 340 million children, are affected by micronutrient deficiencies (MND). These deficiencies are difficult to diagnose because the effects often become visible only when the deficiency is severe. And current diagnoses generally require time-consuming and expensive blood draws and laboratory tests. And quantifying micronutrient levels in a blood sample requires limited, specialized laboratory equipment.
The Harvard researchers decided to look for a more efficient and cost-effective way to collect information about MND. Their solution? Satellite data.
But predicting an indirect feature such as MND prevalence raises technical challenges, including choosing relevant satellite data, linking a limited amount of ground truth data from individuals to satellite data to train machine learning models, and supporting interpretability for public health experts.
The Harvard team examined raw satellite measurements, using AI to sift through the data and pinpoint key features. The team was able to combine data such as vegetation cover, weather, and water presence to indicate where populations will lack iron, vitamin B12 or vitamin A.
In their study of four regions of Madagascar, the researchers found the model’s predictions of regional-level micronutrient deficiency in populations outside the training datasets were as accurate, and sometimes more accurate, than estimates based on surveys by local public health officials.
The Harvard researchers concluded that their method could promote early public health interventions. The researchers plan to develop a software application that extends this analysis to other countries that have public satellite data. “We hope that this application could allow public health officials to interact with the insights our system can provide and help to inform interventions,” Bondi concluded.