Summary: AI Models for Hepatic Steatosis Detection
Recent advancements in AI have led to the development of candidate models for diagnosing hepatic steatosis using convolutional neural networks (CNNs). These deep learning models, trained on the well-regarded ImageNet dataset, optimize parameters to classify X-ray images for steatosis detection. By maximizing the Youden index, researchers established an effective threshold to balance sensitivity and specificity. The trained models showcased strong performance, achieving an area under the curve (AUC) of 0.83 for internal and 0.82 for external test sets, with accuracy, sensitivity, and specificity rates of 77%, 68%, and 82%, respectively. Notably, when each patient underwent a single exam, AUC scores improved to 0.86 and 0.83. Saliency maps identified critical areas consistent with hepatic regions, reinforcing the utility of leveraging existing chest X-rays for opportunistic screening, potentially streamlining triage to dedicated liver assessments. This innovation enhances the role of radiology in managing metabolic liver diseases effectively.
