2020/03/05 by Joshua Levy, Levy, Joshua J., Rebecca M. Lebeaux +9
Environmental Science · Social Sciences · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Health disparities and outcomes #Impact of Light on Environment and Health #Insurance, Mortality, Demography, Risk Management #Machine Learning (cs.LG) #Machine Learning (stat.ML)
paper · pdf · doi:10.48550/arxiv.2003.08750
openalex publication_date 2020/03/05 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Importance: Following a century of increase, life expectancy in the United\nStates has stagnated and begun to decline in recent decades. Using satellite\nimages and street view images prior work has demonstrated associations of the\nbuilt environment with income, education, access to care and health factors\nsuch as obesity. However, assessment of learned image feature relationships\nwith variation in crude mortality rate across the United States has been\nlacking.\n Objective: Investigate prediction of county-level mortality rates in the U.S.\nusing satellite images.\n Design: Satellite images were extracted with the Google Static Maps\napplication programming interface for 430 counties representing approximately\n68.9% of the US population. A convolutional neural network was trained using\ncrude mortality rates for each county in 2015 to predict mortality. Learned\nimage features were interpreted using Shapley Additive Feature Explanations,\nclustered, and compared to mortality and its associated covariate predictors.\n Main Outcomes and Measures: County mortality was predicted using satellite\nimages.\n Results: Predicted mortality from satellite images in a held-out test set of\ncounties was strongly correlated to the true crude mortality rate (Pearson\nr=0.72). Learned image features were clustered, and we identified 10 clusters\nthat were associated with education, income, geographical region, race and age.\n Conclusion and Relevance: The application of deep learning techniques to\nremotely-sensed features of the built environment can serve as a useful\npredictor of mortality in the United States. Tools that are able to identify\nimage features associated with health-related outcomes can inform targeted\npublic health interventions.\n