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

Area of Science:

• Radiology
• Artificial Intelligence
• Hepatology

Background:

• Hepatic steatosis, or fatty liver disease, is a growing health concern.
• Current diagnostic methods can be invasive or costly.
• There is a need for non-invasive, accessible screening tools for hepatic steatosis.

Purpose of the Study:

• To develop and evaluate a deep learning model for detecting hepatic steatosis.
• To utilize chest radiographs as the imaging modality for this AI model.

Main Methods:

• A retrospective study included 6,599 chest radiographs from patients who underwent controlled attenuation parameter (CAP) examinations.
• A deep learning model was trained, tuned, and validated on data from two institutions.
• Performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity on internal and external test sets.

Main Results:

• The deep learning model achieved an AUC of 0.83 on the internal test set and 0.82 on the external test set.
• Accuracy for the internal and external test sets was 77% and 76%, respectively.
• Sensitivity and specificity were also reported, demonstrating good performance in detecting hepatic steatosis.

Conclusions:

• The developed deep learning model demonstrates effective performance in identifying hepatic steatosis from chest radiographs.
• This AI-driven approach shows potential as a non-invasive screening tool for liver conditions.