Deep Learning Electronic Cleansing for Single- and Dual-Energy CT Colonography

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Electronic cleansing (EC) computationally removes residual material in CT colonography for polyp detection.
• Current EC methods, developed for high-dose CT with cathartic preparation, produce artifacts with low-dose, noncathartic protocols.
• These artifacts can mislead readers during virtual colonoscopy interpretation.

Purpose of the Study:

• Review causes of artifacts in conventional EC schemes.
• Describe applications of deep learning and dual-energy CT for EC.
• Demonstrate image quality improvements with deep learning EC at single- and dual-energy CT colonography.

Main Methods:

• Utilized deep learning algorithms, including transfer learning, for EC.
• Employed dual-energy CT colonography to leverage spectral information.
• Evaluated EC performance with noncathartic bowel preparation at low radiation doses.

Main Results:

• Deep learning EC significantly reduces cleansing artifacts compared to conventional methods.
• Dual-energy CT, combined with deep learning EC, yields fewer artifacts than single-energy CT.
• Improved image quality facilitates more reliable polyp detection in virtual colonoscopy.

Conclusions:

• Deep learning offers a promising solution for improving EC in low-dose, noncathartic CT colonography.
• Dual-energy CT enhances EC by providing more specific material characterization.
• These advancements are crucial for accurate polyp detection and patient management.