Automatically Measuring Individual Consumption Events During Natural Eating Using a Table Embedded Scale

This thesis is motivated to improve the tools available for tracking energy intake. The goal of this work is to develop a table-embedded scale capable of measuring the weight of individual consumption events during unrestricted eating. The method was tested on a dataset gathered from 276 subjects eating 518 courses consisting of 22,383 marked individual consumption events in a cafeteria environment. Approximately 30% of the consumption events can be detected and weighed. The remaining 70% of the events occur without participants interacting with the scale or when noisy interactions with the scale prevent weight measurement. The relationship between bite size in grams and BMI was analyzed across all 7,501 measurable bites found using ground truth bite times and across 10,240 automatically detected events. The relationship between bite weight and BMI was found to be 0.28 g/BMI. Without using ground truth bite times, a relationship of 0.21 g/BMI was found for automatically detected events. The trend is diminished but still clearly present even with the presence of false alarms. In addition, when each bin is broken into quartiles, the results indicate that g/bite vs BMI is nearly constant for the smallest 25% of bites, but increases in each quartile. When the largest 25% of bites are analyzed, a relationship of 0.58 g/BMI is found. While these amounts may seem small, the cumulative effect over hundreds or thousands of bites suggests new opportunities for behavior change based on bite size.