Performance of a Chest Radiograph-based Deep Learning Model for Detecting Hepatic Steatosis

Purpose To develop and evaluate a deep learning model for detecting hepatic steatosis using chest radiographs. Materials and Methods This retrospective study included consecutively collected chest radiographs from patients who underwent controlled attenuation parameter (CAP) examinations at two institutions from November 2013 to May 2023. All patients were diagnosed as having or not having hepatic steatosis based on CAP value. Patients from one institution were randomly divided into training, tuning, and internal test sets using an 8:1:1 ratio. Patients from the other institution comprised an external test set. A deep learning-based model to classify hepatic steatosis using chest radiographs was trained, tuned, and evaluated. Model performance on the internal and external test sets was assessed using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. Results In total, 6599 radiographs associated with 6599 CAP examinations obtained in 4414 patients were included. The internal test set included 529 radiographs from 363 patients (mean age, 56 years ± 11 [SD]; 344 male patients). The external test set included 1100 radiographs from 783 patients (mean age, 58 years ± 16; 604 male patients). The AUC, accuracy, sensitivity, and specificity (with 95% CIs) for the internal test set were 0.83 (0.79, 0.86), 77% (74, 81), 68% (61, 75), and 82% (77, 85), respectively. For the external test set, the values were 0.82 (0.79, 0.85), 76% (73, 78), 76% (69, 81), and 76% (73, 79), respectively. Conclusion The developed deep learning model showed good performance for detecting hepatic steatosis using chest radiographs. Keywords: Liver, Hepatic Steatosis, Chest Radiography, Controlled Attenuation Parameter Supplemental material is available for this article. © RSNA, 2025.