Deep-learning based quantitative evaluation of postoperative atelectasis following right upper lobectomy.

Area of Science:

• Pulmonology
• Medical Imaging
• Artificial Intelligence

Background:

• Current atelectasis grading is subjective and lacks scalability.
• Postoperative atelectasis assessment requires objective and automated methods.
• Deep learning offers potential for quantitative analysis in medical imaging.

Purpose of the Study:

• To develop and validate a deep learning framework for automated quantification and grading of postoperative atelectasis.
• To assess the correlation between volumetric lung metrics and atelectasis severity.
• To evaluate the association of volumetric changes with clinical outcomes.

Main Methods:

• Retrospective cohort of patients undergoing right upper lobectomy (2008-2023).
• Utilized nnU-Net v2 for segmentation of preoperative/postoperative lobes and airways.
• Performed volumetric quantification of lung lobes and assessed atelectasis using a 5-point radiological scale.

Main Results:

• Significant correlation found between atelectasis grade and right middle lobe (RML) volume loss (p < 0.001).
• Normalized RML/right lung and RML/total lung volume ratios differed significantly across atelectasis grades (p < 0.001).
• Increased normalized RLL volumes suggested compensatory hyperinflation; higher ΔRML/RL and ΔRML/TL associated with reduced need for bronchoscopy (p=0.01, p=0.03).

Conclusions:

• Deep learning-based volumetric assessment is feasible for quantifying atelectasis post-right upper lobectomy.
• Automated grading provides objective and scalable evaluation of postoperative atelectasis.
• Volumetric metrics demonstrate clinical relevance in predicting patient outcomes.