Deep Learning-Based Deblurring for Computed Tomography and a Phantom Evaluation for Coronary Artery Calcium...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computed tomography (CT) imaging can suffer from noise and blur, impacting diagnostic accuracy.
• Coronary artery calcium (CAC) scoring is crucial for cardiovascular risk assessment.
• Current CT reconstruction methods may not fully optimize image quality for detecting small calcifications.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) based deblurring algorithm for CT imaging.
• To assess the impact of DL deblurring on CT image quality and CAC detectability using phantom studies.

Main Methods:

• A two-stage deep learning model using sequential convolutional neural networks (CNNs) for denoising and deblurring was developed.
• The DL algorithm was tested on an anthropomorphic phantom with simulated calcifications under various radiation dose levels.
• Image quality and CAC detectability (Agatston score) were evaluated with and without DL deblurring.

Main Results:

• Deep learning deblurring increased CAC detectability from 39% to 48% at high radiation doses.
• At routine clinical doses, DL deblurring achieved 39% CAC detectability, a 6% improvement over standard reconstruction.
• The DL algorithm demonstrated potential for improving image texture by reintroducing estimated noise post-deblurring.

Conclusions:

• Quantitative phantom testing confirms that DL deblurring enhances CT image quality.
• The algorithm provides a clinically relevant improvement in CAC detection metrics at standard radiation doses.
• DL deblurring represents a promising advancement for improving diagnostic performance in CT imaging.