Low-dose lung CT processing using weighted intensity averaging over large-scale neighborhoods.

Area of Science:

• Medical Imaging
• Radiology
• Image Processing

Background:

• Low-dose computed tomography (LDCT) is crucial for lung cancer screening.
• Reducing radiation dose in LDCT can compromise image quality, increasing noise and artifacts.
• Existing methods struggle to maintain diagnostic quality at significantly reduced radiation levels.

Purpose of the Study:

• To develop and evaluate a novel image processing technique, weighted intensity averaging over large-scale neighborhoods (WIA-LN), for improving LDCT image quality.
• To assess the effectiveness of WIA-LN in noise and artifact reduction while enhancing nodule visibility in reduced radiation settings.
• To validate the clinical applicability of WIA-LN for low-dose lung CT screening.

Main Methods:

• Implementation of the WIA-LN algorithm, which uses weighted averaging of pixel intensities based on surrounding texture similarity.
• Application of compute unified device architecture (CUDA) based parallelization to accelerate the WIA-LN processing.
• Acquisition of LDCT images with reduced tube current (75%) and tube voltage (33.3%) using a Siemens CT scanner, alongside standard-dose CT images for comparison.
• Evaluation using both an anthropomorphic lung phantom and clinical patient data, with qualitative and quantitative assessments by radiology specialists.

Main Results:

• The WIA-LN processed LDCT images demonstrated superior visual and qualitative performance compared to original LDCT images.
• Statistically significant improvements were observed in noise and artifact suppression (P < 0.05).
• Enhancement of nodule structure visibility was achieved, aiding in clearer identification.
• The method proved effective even with substantial reductions in tube current and voltage.

Conclusions:

• The WIA-LN method offers a significant advancement in processing low-dose lung CT images.
• It effectively mitigates noise and artifacts while improving nodule conspicuity, crucial for accurate lung cancer screening.
• Algorithm optimization through parallelization enhances its clinical applicability, making it a viable tool for radiologists.