Probabilistic Context-Free Grammar for Food Intake

摘要

The present invention relates to monitoring and correcting eating behavior through continuous food weight measurements to treat obesity and eating disorders and has been successfully applied in clinical practice. To this end, Mandometer®, dishes and scales have been used along with video recordings of subjects during a single meal course to assist clinicians in measuring related food intake parameters. In this task, we provide a novel algorithm for automatically constructing a food intake curve for a subject using only MandoMeter® gravimetric values. This significantly reduces manual labor for analysis by eliminating the need for direct observation or video recording by the clinician. The proposed algorithm detects an event based on a parametric probabilistic context-free grammar of the parameters of the recorded sequence and then performs an adaptive preprocessing step using a delta coefficient (Such as bites, food additions, artifacts) by applying them. Experimental results of 114 meal data sets from individuals suffering from obesity or eating disorders as well as individuals with normal BMI show the effectiveness of the proposed approach.