Kinect body composition: assessing clinical metabolic health indices using 3D optical images

Presenting Author Senior Author
Name: Bennett Ng Name: John Shepherd
Email: Email:
Presenting Author’s RIG/SRG: Musculoskeletal  
Presenting Author's Lab Location: Parnassus   

Abstract Information
Imaging Modality: Optical DXA
Disease Application: Obesity
Complete author list: Bennett K. Ng, Benjamin J. Hinton, Jesus I. Avila, Bo Fan, Leila Kazemi, Eboni Stephens, Viva W. Tai, Caitlin Sheets, Alka Kanaya, John A. Shepherd
Abstract highlights: Towards an accessible platform for body composition assessment, we aim to derive accurate models of body composition using body shape information from 3D optical images. Such models can be used in clinical and consumer environments for assessment and management of metabolic health.
Worldwide prevalence of obesity has skyrocketed in recent decades. Affecting both adults and children, obesity is directly associated with chronic health conditions such as diabetes, cardiovascular disease, and cancer. The WHO defines obesity using body mass index (BMI), however this metric is insensitive to the relative amounts and distribution of fat and lean tissues throughout the body. As part of comprehensive strategies to counter the obesity epidemic, there is a clear need for accurate, accessible, and affordable technologies for body composition assessment. Existing methods, however, are sub-optimal due to high cost, limited accuracy, and/or use of ionizing radiation. We are investigating the emerging use of 3D optical imaging devices for whole-body and regional body composition assessment.
We are conducting a cross-sectional study of 54 healthy adults, stratified by age, sex, and BMI. This study is approved by the UCSF CHR and informed consent was obtained prior to all measurements. Each participant is imaged twice using a 3D surface scanner (Proscanner, Fit3D Inc., Redwood City, CA). Participants receive whole-body dual energy X-ray absorptiometry (DXA) (Discovery/W and Horizon/A, Hologic Inc., Bedford, MA) and air displacement plethysmography (ADP) (BodPod, COSMED, Rome, Italy) scans as criterion measurements. An example of matched 3D and DXA images is shown in Figure 1. The data collected were used to derive predictive models for whole-body and regional lean tissue mass as well as subcutaneous and visceral adipose tissue mass from optical measurements. Specifically, landmark points are algorithmically placed on 3D images. These landmarks are used to calculate anthropometric lengths, areas, and volumes. Generalized linear regression models are then derived to predict various DXA body composition indices as a function of anthropometric measures from the 3D scans. Two external validation sets of bioelectrical impedance analysis (BIA) and DXA data acquired at fitness centers were used to validate selected model performance.
The 3D scanner exhibited high precision on common anthropometric measurements (chest, hip, thigh, and waist circumferences: %CV 0.25-1.23). Body volume measurements from the 3D scanner showed high agreement with ADP (R2 = 0.986, RMSE = 2.08 liters). The predictive models validated well to the external datasets (whole body %fat [adjusted R2 = 0.72, RMSE = 3.62], fat mass [adjusted R2 = 0.76, RMSE = 3.62 kg], fat free mass [adjusted R2 = 0.86, RMSE = 2.20 kg], visceral fat mass [adjusted R2 = 0.65, RMSE = 0.12 kg]) and required only 2-4 volumes and length/volume ratio features derived from the 3D scans.
3D optical surface scanning shows promise as an accessible modality for accurate clinical assessment of holistic and regional body composition. Falling technology costs and lack of ionizing radiation make 3D surface imaging a particularly appealing solution for regular use in various environments. Further development is needed to validate 3D optical body composition across populations, and to derive more advanced shape features indicative of metabolic health.