Facial Imagery BMI Algorithm correlates with Normal and Overweight Measured BMI

Abstract

Efficient procedures to determine health status are in high demand in today's generation. Body Mass Index (BMI) is a global method for determining an individual's health status. This research exploration is to identify if a facial image (photograph) can identify participant's BMI correctly. Sample of 1,210 young adults with a facial image and objective height and weight were used for analysis. Facial BMI (fBMI) was measured through the Local Binary Pattern (LBP)/Active Shape Model (ASM) was used to detect 76 points on each face and measured BMI (mBMI) was calculated by weight in kilograms divided by height in meters squared. Correlation analysis of fBMI to mBMI showed significant correlation between BMIs in the normal and overweight categories (p<.0001). Further analysis indicated the measure to be less efficacious in underweight and obese participants. Matched pairs data for each individual, there was a range of 14.73–49.74 for mBMI and a narrowed spread for fBMI (range = 16.29–28.85) indicating that fBMI detected participant BMI 0.4212 less than mBMI (p<.0004). BMI categories used included: less than 18.5 BMI (0), 18.5–24.9 (1), 25.0–29.9 (2), and equal to or above 30 (3). mBMI had representation of categories across 0–3 BMI, while fBMI representation included just 0–2. This shows less sensitivity of the algorithm to obese individuals. Contingency table analysis indicted 109 participants in the 3rd category of mBMI were placed into a lower category for fBMI. Agreement test for symmetry shows significant disagreement 95% CI [0.0618,0.1446] (p<.0001). Facial imagery identification of health status is a useful measure in human research however; more sensitive measures to identify underweight and obese individuals are warranted.Support or Funding InformationApproval to use the data set was granted by the University of Tennessee Institutional Review Board prior to study implementation. This material is based upon work that is supported by the National Institute of Food and Agriculture, U.S. Department of Agriculture, under award number 2014‐67001‐21851.