Machine learning slice-wise whole-lung CT emphysema score correlates with airway obstruction

Area of Science:

• Pulmonary Medicine
• Radiology
• Artificial Intelligence in Medical Imaging

Background:

• Quantitative CT imaging is a key biomarker for emphysema, particularly in smokers.
• Current quantitative methods often lack strong correlation with radiologists' visual assessments.
• There is a need for improved quantitative CT analysis for emphysema.

Purpose of the Study:

• To develop and validate a novel neural network-based slice-wise whole-lung emphysema score (SWES) for chest CT.
• To compare the performance of SWES against a conventional quantitative CT method (LAV950) and visual scoring.
• To assess SWES's correlation with spirometry measures of airway obstruction.

Main Methods:

• Developed SWES using a neural network on chest CT data.
• Validated SWES on independent cohorts, including the Swedish CArdioPulmonary bioImage Study (SCAPIS).
• Compared SWES with LAV950 and radiologists' visual emphysema scores (sum-visual) using correlation and ROC analysis, referencing spirometry (FEV1/FVC).

Main Results:

• SWES demonstrated a significantly stronger correlation with visual emphysema scores than LAV950 (r=0.78 vs. r=0.41, p<0.001).
• SWES showed a superior area under the ROC curve for predicting airway obstruction compared to LAV950 (0.76 vs. 0.61, p=0.007).
• SWES exhibited stronger correlations with FEV1/FVC across all cohorts than LAV950 and visual scores.

Conclusions:

• The developed SWES provides a more accurate quantitative assessment of emphysema on CT images.
• SWES outperforms traditional threshold-based scoring (LAV950) and visual assessment in correlating with clinical measures.
• SWES offers a promising tool for emphysema quantification in large-scale research and clinical settings.