Chronic disease such as heart disease, diabetes, and obesity are known to be strongly linked with diet and may be rooted in the environmental context where they are prevalent. This proposal aims to develop imaging-based techniques to investigate the link between eating environment and dietary quality and satisfaction which are not known. The project will use images from the food environment to address the fundamental question of where, how and when food should be consumed to maximize health and prevent disease. Monitoring the personal dietary environment and determination of environmental patterns related to dietary intake can empower both health care providers and patients to optimize evidence-based decisions. This information can help individuals recognize less healthful behaviors that may be occurring in their lives. Health professionals will also have better information to advise behavioral strategies within the context of the patient's environment. The results may also be used to help guide the development of programs to reduce the prevalence of obesity and diet-related chronic diseases in the US population and advise US dietary policy.
This highly interdisciplinary investigation explores image processing and computer vision techniques to extract and quantify dietary environmental factors and study their connections with dietary intake. The project plans to build informative models of behavioral health profiles that can take advantage of a large set of observed data, including food images and contextual information that the PI has access to. The team will develop computational methods that leverage the use of contextual information for image-based dietary data which is highly individualized, temporal, and contextualized. The benefits of including contextual information are twofold: it provides a more complete composite of a person's health influencers of dietary behavior; and can improve the accuracy of food recognition and nutrient intake estimation using computer vision techniques. The proposed work will develop 1) new image analysis techniques that leverage contextual cues such as eating time, location type, co-occurrence patterns of objects, personalized learning models from image-based dietary record; 2) novel machine learning and statistical analysis tools for dietary pattern discovery and prediction by exploring relationships among the environmental factors and their association with dietary quality; 3) experimental validation of the proposed methods using existing image-based dietary data.
