Patient posture estimation using super-resolution reconstruction of pressure distribution image for pressure ulcer prevention

In this study, to improve the prediction of pressure ulcer spots, we have developed super-resolution (SR) techniques to reconstruct a high-resolution (HR) pressure image from a low-resolution (LR) body pressure image to overcome the limitations of sensor resolution. We implemented a super-resolution generative adversarial network (SRGAN) to reconstruct pressure images and a convolution neural network (CNN) to predict posture. To evaluate the similarity between the original pressure image and the 4× rescaled LR body pressure image restored using SR technology, we used image quality assessment (IQA) technology, peak signal-to-noise ratio (PSNR), and structural similarity (SSIM). The reconstructed pressure images were classified into four patient postures (supine, right side, left side, and others) with 98.37% accuracy showing the feasibility of practical implementation. © ICROS 2021.