The Feasibility of Using a Deep Learning-Based Model to Determine Cardiac Computed Tomographic Contrast Dose

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Cardiac computed tomography (CT) requires precise contrast agent administration for optimal image quality.
• Predicting the necessary iodine dose for adequate contrast enhancement is crucial for diagnostic accuracy and patient safety.
• Current methods often rely on patient physical size, which may not precisely correlate with individual contrast needs.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) model for predicting contrast effects in cardiac CT using only CT localizer radiographs.
• To compare the predictive performance of the DL model against conventional methods based on patient body size.

Main Methods:

• A retrospective analysis of 473 cardiac CT scans was performed.
• Deep learning models were developed to predict iodine dose per contrast effect (IDCE) from CT localizer radiographs.
• Model performance was assessed by comparing Pearson correlation coefficients (r) between actual and predicted IDCE, using body weight, lean body weight, and body surface area as comparators.

Main Results:

• The DL model demonstrated higher correlation coefficients for predicting IDCE compared to body weight, lean body weight, and body surface area in both male (r=0.607) and female (r=0.412) groups.
• These findings indicate superior predictive accuracy of the DL model.
• The model's performance was robust, requiring only readily available CT localizer radiographs.

Conclusions:

• Deep learning analysis of CT localizer radiographs provides a powerful tool for predicting cardiac CT contrast effects.
• The DL model's performance is comparable, if not superior, to conventional methods relying on patient physical size.
• This approach offers a non-invasive and efficient way to optimize contrast dosing in cardiac CT.