Artificial Intelligence-Based Prediction of Contrast Medium Doses for Computed Tomography Angiography Using Optimized...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Optimizing contrast medium dose in computed tomography (CT) angiography is crucial for diagnostic image quality and patient safety.
• Current methods may not fully account for individual patient variability, potentially leading to suboptimal dosing.
• Artificial intelligence offers a promising avenue for personalized dose prediction.

Purpose of the Study:

• To develop and evaluate an AI-based algorithm for predicting the optimal contrast medium dose in CT angiography of the aorta.
• To compare the performance of different machine learning models, including random decision forests (RDF), k-nearest neighbors (KNN), and regression neural networks (RNN), for contrast dose prediction.
• To assess the feasibility of implementing the AI system in routine clinical practice.

Main Methods:

• The contrast dose prediction was framed as a classification problem, utilizing image contrast as a primary feature.
• Random decision forests (RDF) and k-nearest neighbor (KNN) methods were employed for classification.
• Feature selection involved evaluating all combinations of 22 clinical parameters, with accuracy and precision as quality metrics.
• Regression neural networks (RNN) were used for feature transformation and direct classification, as well as preprocessing for KNN.

Main Results:

• For feature selection, RDF achieved 84.42% accuracy and KNN achieved 86.21% precision, with age, height, and hemoglobin identified as key parameters.
• Feature transformation using RNN significantly improved performance, reaching 90.00% accuracy and 97.62% precision with all 22 parameters.
• Using a standard clinical feature set, the RNN achieved 86.67% accuracy and 93.18% precision, balancing performance with clinical feasibility.

Conclusions:

• A reliable hybrid AI system was developed to assist radiologists in determining optimal contrast doses for CT angiography.
• The system leverages patient-specific parameters for personalized contrast administration.
• The AI approach demonstrates potential for enhancing diagnostic accuracy and efficiency in CT angiography procedures.