Emphysema quantification using low-dose computed tomography with deep learning-based kernel conversion comparison.

Area of Science:

• Radiology
• Medical Imaging
• Pulmonary Medicine

Background:

• Low-dose computed tomography (LDCT) is crucial for emphysema quantification but is sensitive to kernel selection.
• Standard-dose computed tomography (SDCT) provides a benchmark, but involves higher radiation exposure.
• Kernel differences can significantly impact emphysema index (EI) and lung density measurements.

Purpose of the Study:

• To evaluate the impact of dose reduction and kernel selection on emphysema quantification using LDCT.
• To assess the efficacy of a deep learning-based kernel conversion technique for normalizing LDCT images for emphysema quantification.

Main Methods:

• 131 participants underwent both LDCT and SDCT.
• LDCT images were reconstructed with B31f (smooth) and B50f (sharp) kernels; SDCT images used B30f (smooth) kernels.
• A deep learning model converted B50f LDCT images to a B31f equivalent for comparison.

Main Results:

• Sharp kernel (B50f) LDCT significantly differed from smooth kernel LDCT (B31f) and SDCT (B30f) in emphysema quantification.
• Deep learning-based normalization of the sharp kernel (B50f to B31f) significantly reduced variation.
• Agreement between normalized LDCT (B31f and converted B50f) and SDCT (B30f) was within acceptable limits (-2.9 to 4.4%).

Conclusions:

• Deep learning-based kernel conversion effectively normalizes sharp kernels in LDCT for emphysema quantification.
• Smooth kernels in LDCT demonstrate adequate performance for emphysema quantification compared to SDCT.
• This deep learning approach reduces variation and potential overestimation associated with sharp kernels in emphysema assessment.