Neighbourhood looking glass: 360º automated characterisation of the built environment for neighbourhood effects research

Abstract

BackgroundNeighbourhood quality has been connected with an array of health issues, but neighbourhood research has been limited by the lack of methods to characterise large geographical areas. This study uses innovative computer vision methods and a new big data source of street view images to automatically characterise neighbourhood built environments.MethodsA total of 430 000 images were obtained using Google’s Street View Image API for Salt Lake City, Chicago and Charleston. Convolutional neural networks were used to create indicators of street greenness, crosswalks and building type. We implemented log Poisson regression models to estimate associations between built environment features and individual prevalence of obesity and diabetes in Salt Lake City, controlling for individual-level and zip code-level predisposing characteristics.ResultsComputer vision models had an accuracy of 86%–93% compared with manual annotations. Charleston had the highest percentage of green streets (79%), while Chicago had the highest percentage of crosswalks (23%) and commercial buildings/apartments (59%). Built environment characteristics were categorised into tertiles, with the highest tertile serving as the referent group. Individuals living in zip codes with the most green streets, crosswalks and commercial buildings/apartments had relative obesity prevalences that were 25%–28% lower and relative diabetes prevalences that were 12%–18% lower than individuals living in zip codes with the least abundance of these neighbourhood features.ConclusionNeighbourhood conditions may influence chronic disease outcomes. Google Street View images represent an underused data resource for the construction of built environment features.

Associations of Residential Greenness with Diabetes Mellitus in Chinese Uyghur Adults

“…A study of 249,405 USA elderly (≥65 years) also found greenness significantly reduced diabetes mellitus risk; an increase in NDVI from 1 SD less to 1 SD more than mean was associated with a reduced risk of 14% for diabetes mellitus [14]. Likewise, another three cross-sectional studies conducted in USA [15], the United Kingdom [18], and Canada [20] observed increased neighborhood or street green space was significantly associated with lower individual diabetes mellitus prevalence. Additionally, several cohort studies have also documented a significant and inverse relationship between greenness levels and diabetes mellitus incidence [9,16].…”

“…Several studies have explored the associations of greenness exposure with glucose levels, insulin resistance, and diabetes mellitus risk, but showed inconsistent results [8][9][10][11][12][13][14][15][16][17][18][19][20]. For example, cross-sectional studies conducted in Netherlands, Australia, the United States, the United Kingdom, and Canada have found that green space had a protective effect on diabetes mellitus [13][14][15]17,18,20], while a cross-sectional study among the Norwegian population found no significant association [8]. Two prospective cohort studies from England and Canada observed a significant association between increased greenness exposure and lower relative hazard of developing diabetes mellitus [9,16], but the Framingham study did not find such a longitudinal association [11].…”

Health and the built environment in United States cities: measuring associations using Google Street View-derived indicators of the built environment

“…Methods for identifying street intersections and retrieving and labeling GSV images have been published previously [14,15]. Briefly, latitude and longitude data coordinates for all U.S. street intersections were obtained from the 2017 Census Topologically Integrated Geographic Encoding and Referencing (TIGER) data.…”

“…Image resolution was 640 × 640 pixels. Images were processed using trained Visual Geometry Group (VGG-16 model) deep convolutional networks [30,31] (previously detailed by Nguyen et al [15]) to identify the five built environment features of interest (one network per feature). Accuracy of the recognition tasks (comparing the images labeled using this machine learning approach compared with assessment by a human reviewer) ranged from 85 to 93%, and these figures were consistent with a separate, semi-supervised learning approach.…”

“…Researchers have demonstrated the reliability of using GSV to derive data on physical features, finding high-level agreement with field assessments [11][12][13]. GSV images have been used to determine the presence of features such as crosswalks, commercial buildings, highways, and grasslands, which were in turn used to assess their association with chronic health outcomes at the county [14] and zip code level [15]. Globally, GSV image coverage is more complete for some regions than others, with cities in developed nations having near-complete coverage while many low-and middle-income countries in Africa, Southeast Asia, and South America have no GSV imagery at all [10].…”