Development of an Image-to-Image Methodology for Customized Prediction of Body Shape Changes

We developed an algorithm to predict changes in a person’s body shape when they reach the target body mass index (BMI) by using the current body shape image and target BMI as inputs. This algorithm is the first image-based generation methodology to predict body shape changes according to the desired BMI level in a single photographic image. Frontal and lateral images, and height and weight data, were collected from 230 women who visited an obesity hospital. Any insufficient data were reinforced using a CcGAN. The superiority of this algorithm was proved through qualitative and quantitative evaluations. As a representative evaluation result using a lateral image, Fréchet inception distance (FID), learned perceptual image patch similarity (LPIPS), peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and BMI error values of 106.4913, 0.0090, 60.4438, 0.5612, and 0.0052, respectively, were recorded, proving the superiority of the developed algorithm over other algorithms. The algorithm can be used not only as a weight management, but also as an important tool for managing and predicting postoperative recovery processes and body shape changes, and is expected to have a positive impact on individual body shape management and health promotion.