Dose Optimization Using a Deep Learning Tool in Various CT Protocols for Urolithiasis: A Physical Human Phantom Study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Radiation Dose Optimization

Background:

• Determining optimal radiation doses for CT scans is crucial for maintaining image quality.
• Deep learning (DL) applications show potential in enhancing medical images.
• Evaluating DL's impact on image quality and radiation dose is essential.

Purpose of the Study:

• To assess the effectiveness of a deep learning application in optimizing radiation dose for CT imaging.
• To maintain diagnostic image quality using reduced radiation exposure.
• To compare deep learning-enhanced images with traditional reconstruction methods.

Main Methods:

• Uric acid stones were placed in a physical human phantom.
• CT scans were performed using varying tube voltages (120, 100, 80 kV) and current-time products (100, 70, 30, 15 mAs).
• Images were reconstructed using filtered back projection (FBP), iterative reconstruction (IR, iDose), and knowledge-based iterative model reconstruction (IMR), with and without deep learning application.
• Objective (Hounsfield unit standard deviation) and subjective assessments by radiologists and urologists were used to evaluate image quality and diagnostic accuracy.

Main Results:

• Deep learning application reduced objective image noise across all reconstruction methods.
• Deep learning-applied FBP achieved similar noise levels to IR at approximately one-third the radiation dose (100 kV-30 mAs).
• Subjective image quality scores deteriorated at radiation doses below 100 kV-30 mAs.
• Diagnostic accuracy improved with deep learning at settings below 100 kV-30 mAs, except for 80 kV-15 mAs.

Conclusions:

• Deep learning-applied FBP demonstrates comparable image quality to IR at 100 kV-30 mAs or higher settings.
• A radiation dose reduction of approximately one-third is achievable at 100 kV-30 mAs using deep learning while preserving objective noise levels.
• Deep learning holds promise for reducing radiation exposure in CT imaging without compromising diagnostic performance.