Using a deep neural network for four-dimensional CT artifact reduction in image-guided radiotherapy

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Breathing artifacts degrade the quality of four-dimensional computed tomography (4DCT) images.
• Deep neural networks (DNNs) offer potential for artifact reduction in medical imaging.

Purpose of the Study:

• To develop and evaluate a DNN-based method for reducing breathing artifacts in 4DCT images.
• To improve the geometrical accuracy and diagnostic quality of 4DCT scans.

Main Methods:

• Trained three DNNs (DNN1, DNN2, DNN3) using 857 thoracoabdominal 4DCT datasets.
• Simulated artifacts by interposing CT images from different breathing phases.
• Developed DNN3 with a classifier for artifact detection and a generator for artifact reduction using coronal images and an artifact map.

Main Results:

• DNN3 demonstrated superior performance in reducing artifacts and improving geometrical accuracy, particularly in sagittal sections.
• Diaphragm position accuracy was highest when using DNN3.
• DNN2 corrected artifacts but also affected artifact-free regions, unlike DNN3.

Conclusions:

• DNN-based artifact reduction significantly enhances 4DCT image quality.
• DNN3, utilizing an artifact map, effectively reduces interposition-related artifacts and improves accuracy.
• Incorporating information from other respiratory phases and artifact regions leads to substantial improvements in 4DCT analysis.