Deep Learning-Based Kernel Adaptation Enhances Quantification of Emphysema on Low-Dose Chest CT for Predicting...

Area of Science:

• Radiology
• Pulmonary Medicine
• Artificial Intelligence in Medicine

Background:

• Low-dose computed tomography (LDCT) is used for lung cancer screening.
• Quantifying emphysema on LDCT can be challenging due to image quality.
• Deep learning offers potential for image enhancement and quantitative analysis.

Purpose of the Study:

• To evaluate the predictive value of emphysema quantification using LDCT after deep learning-based kernel adaptation.
• To assess the association between emphysema severity and long-term nonaccidental mortality.

Main Methods:

• Retrospective analysis of LDCT scans from asymptomatic individuals aged 60+.
• Application of a deep learning algorithm for kernel adaptation to LDCT images.
• Quantification of emphysema using LAA-950 (lung volume percentage with attenuation ≤ -950 HU) before and after adaptation.
• Survival analysis using Cox proportional hazards models, linked to national registry data.

Main Results:

• Kernel adaptation significantly reduced median LAA-950 and the proportion of emphysema-positive LDCTs.
• Emphysema quantification before adaptation showed no association with mortality.
• After adaptation, higher LAA-950 and emphysema-positivity independently predicted nonaccidental death (adjusted HRs: 1.01 and 1.36, respectively).

Conclusions:

• Deep learning-based kernel adaptation enhances emphysema quantification on LDCT.
• This improved quantification serves as a potential predictor of long-term mortality in asymptomatic populations.