An artificial intelligence system for predicting body weight from chest radiographs

Abstract

Background In the field of diagnostic radiology, radiation dose management, determination of the contrast-medium dose, or estimation of the specific absorption rate level require patients’ body weight. However, accurate body weight is not necessarily available in routine clinical practice. In this study, we investigated whether body weight can be predicted from chest radiographs using deep learning. Methods Our Institutional Review Board approved this retrospective study, and a total of 85,849 chest radiographs obtained for medical checkups between July 2019 and July 2021 were included. A two-stage hierarchical approach composed of a sex-classification model and body-weight prediction model was used. The deep-learning models were trained with chest radiographs from 68,679 training cases and 8585 validation cases, and the remaining 8585 cases were used as test data. The sex-classification model was evaluated for accuracy. The body-weight prediction model was evaluated by calculating the mean absolute error (MAE) and Spearman’s rank correlation coefficient (ρ). Results The overall accuracy of the sex-classification model was 0.992. The MAEs of the body-weight prediction model were 2.62 kg and 3.34 kg for females and males, respectively. The predicted body weight was significantly correlated with the actual body weight (ρ = 0.917, p < 0.001 for females; ρ = 0.914, p < 0.001 for males). Conclusion Body weight was predicted from chest radiographs by applying deep learning. Our method is potentially useful for radiation dose management, determination of the contrast-medium dose, or estimation of the specific absorption rate level in patients with unknown body weights.