Machine learning-driven prediction model of adult body weight based on integrated bioelectrical impedance and infrared imaging.

Traditional body weight measurement methods face challenges, particularly in critically ill patients, necessitating the development of non-invasive and accurate alternatives. We aimed to develop an artificial intelligence (AI) system for weight prediction. A total of 1416 participants were recruited to gather data on their bioelectrical impedance analysis (BIA), whole-body infrared images, height, weight, age, and gender. The BIA-derived, image-derived, and fused AI models were developed and validated using data from 1365 participants, with training and testing ratios of 6:4, 7:3, and 8:2, respectively. The CatBoost model demonstrated superior performance compared to other models evaluated. The BIA-derived model exhibited a mean absolute error (MAE) of 2.58-2.78 kg in the training and testing sets. The image-derived model showed an MAE of 2.40-3.01 kg. Notably, the fused model exhibited the best predictive capability for body weight evaluation when compared to two models, achieving an MAE of 1.42-1.68 kg. All models revealed an exceptional correlation between predicted and actual weight values (r ≥ 0.96, P < 0.05). SHAP analysis indicated that height, gender, and BIA-derived features exhibit a stronger correlation with body weight compared to image-derived features. The fused AI model demonstrated superior and robust predictive capabilities for body weight assessment, establishing a solid foundation for clinical body weight prediction, particularly in critically ill patients. The web application of those weight prediction models is accessible online at https://frp-off.com:59815/.