Poster Topical Area: Obesity

Location: Hall D

Poster Board Number: 697

P23-070 - Are human bodies really shaped like fruit? Using cluster analysis of 3D body scans to challenge current body shape classification norms

Sunday, Jun 10
8:00 AM – 6:00 PM

Background: Due to the manual burden in collecting numerous anthropometric measures, to date, no systematic data analysis determining body shape classifications (and the effect on health) has gone beyond using measurements like BMI or waist and hip circumference, for example, classifying individuals as apple or pear shaped.

Objective: Use unsupervised machine learning techniques on 161 anthropometric measurements obtained by the 3D laser imaging device Human Solutions to determine classification of body shapes and body proportions.


Methods:
Over n = 20,896 soldiers (28% female) recruited for US Army basic training at Fort Jackson, SC were scanned for uniform fitting using the Human Solutions Kinect-based 3D imaging technology. Each subject image consisted of 161 body shape measurements. After removing subjects with missing measurements and separating the dataset by gender, we performed a 2-step cluster analysis on the measurements: feature selection followed by k-means clustering. We interpret the resulting clusters as archetypal body shapes. We also examined the rate of discharge due to physical injury in each cluster to determine whether certain body types are associated with higher incidence of severe injury.


Results:
Three distinct body types (clusters) were identified for males using the 2-step cluster analysis. Cluster 1 consisted of individuals with mild obesity, longer leg and arm lengths, taller, higher waist and hip circumferences, and longer torso length. Cluster 2 consisted of a slightly overweight population with average lengths and circumferences. Cluster 3 was a normal weight population but with below-average lengths and circumferences. Cluster 3 had the highest rate of injury, suggesting the other populations were "protected." Female clusters were similar to male clusters, but exhibited the highest injury rate among the intermediate group (Cluster 2).


Conclusions:
Using data-based techniques from machine learning allows us to identify body types using 3D imaging technology. Our study finds that more shapes arise than the classical "apple" and "pear" dichotomy, and these new shapes appear to give insight into risk for injury.



Body type cluster results

CoAuthors: Kevin Talty – U.S. Military Academy; Diana Thomas – U.S. Military; Patrick Kuiper – U.S. Military Academy; Steven Heymsfield – Pennington Biomedical Research Center; Michael Scioletti – U.S. Military Academy

Steven Morse

Instructor
U.S. Military Academy
West Point, New York