Synthetization of high-dose images using low-dose CT scans.

Area of Science:

• Medical Imaging
• Radiology
• Image Reconstruction

Background:

• X-ray CT dose reduction is crucial for patient safety.
• Current methods include protocol optimization, hardware improvements, and advanced reconstruction algorithms.
• Further advancements are needed to minimize radiation risks.

Purpose of the Study:

• To develop a novel approach for generating synthesized high-dose (SHD) images from low-dose CT scans.
• To overcome limitations of existing methods like model-based iterative reconstruction (MBIR) and deep learning image reconstruction (DLIR), particularly regarding image texture and data availability.
• To address the bottleneck of limited high-quality clinical training data for deep learning image reconstruction.

Main Methods:

• Image processing orthogonal to the imaging plane to create an equivalent thick-slice image (TSI).
• Utilizing the differential signal between original and processed images to identify anatomical modifications.
• Employing iterative noise reduction on the differential signal and subtracting it from TSI to generate SHD.

Main Results:

• Extensive evaluation with phantom and clinical datasets showed negligible residual structures in difference images.
• Quantitative analysis confirmed no CT number bias and consistent noise reduction across anatomical regions.
• Noise Power Spectrum (NPS) analysis demonstrated preservation of noise texture.

Conclusions:

• A new method for generating SHD datasets from low-dose CT scans has been presented.
• The proposed approach achieves excellent noise reduction with desirable noise texture.
• Clinical and phantom studies validated the method's efficacy and robustness.