CT CONTRAST PHASE IDENTIFICATION BY PREDICTING THE TEMPORAL ANGLE USING CIRCULAR REGRESSION.

Area of Science:

• Medical Imaging
• Radiology
• Machine Learning

Background:

• Contrast enhancement in computed tomography (CT) is crucial for visualizing vasculature.
• Accurate timing of contrast agent circulation significantly impacts image quality and diagnostic interpretation.
• Existing methods often categorize contrast timing into discrete phases, limiting precision.

Purpose of the Study:

• To develop and evaluate a novel technique for predicting contrast timing in CT scans.
• To treat contrast timing as a continuous variable using circular regression for enhanced granularity.
• To explore the influence of anatomical slice position on contrast timing prediction accuracy.

Main Methods:

• Utilized circular regression models to represent contrast timing as unit vectors on a circle.
• Employed 2D convolutional neural networks for predicting contrast timing based on anchor time points.
• Trained the model on 877 CT scans and validated on 112 independent scans.

Main Results:

• Achieved a classification accuracy of 93.8% on the test dataset, comparable to state-of-the-art.
• Demonstrated superior performance of the proposed regression model compared to existing 2D and 3D classification approaches.
• Identified a promising relationship between contrast timing and anatomical slice position for future improvements.

Conclusions:

• The circular regression approach provides a more refined method for contrast timing prediction in CT scans.
• This continuous variable model offers advantages over discrete phase methods, especially for patient-specific vascular analysis.
• Incorporating anatomical positional information presents a novel avenue for enhancing prediction accuracy in medical imaging.