Single Inspiratory Chest CT-based Generative Deep Learning Models to Evaluate Functional Small Airways Disease.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonary Medicine

Background:

• Functional small airways disease (fSAD) is a key indicator of early lung disease, often challenging to diagnose with standard imaging.
• Parametric Response Mapping (PRM) is a validated technique for assessing lung disease but typically requires paired inspiratory and expiratory CT scans.
• Developing methods to predict fSAD and perform PRM using single-view CT scans can improve diagnostic efficiency and accessibility.

Purpose of the Study:

• To develop and validate a deep learning (DL) model capable of performing PRM and predicting fSAD from a single inspiratory chest CT scan.
• To compare the performance of generative and predictive DL models in this task.
• To assess the model's generalizability across different test datasets, including an external validation set.

Main Methods:

• Retrospective development of predictive and generative DL models for PRM using inspiratory CT scans.
• Fivefold cross-validation on a model development dataset, with PRM from paired respiratory CT serving as the reference standard.
• Evaluation using voxelwise metrics (sensitivity, AUC, SSIM) and testing on internal and external datasets.

Main Results:

• The generative DL model significantly outperformed the predictive model in detecting fSAD (sensitivity 86.3% vs 38.9%; AUC 0.86 vs 0.70).
• The generative model demonstrated robust performance across multiple internal and external test sets, with AUCs for fSAD detection ranging from 0.83 to 0.85.
• Exceptional performance was noted in the preserved ratio impaired spirometry group (AUC 0.88 for fSAD).

Conclusions:

• A generative DL model utilizing a single inspiratory CT scan can effectively perform PRM and predict fSAD.
• This approach achieves performance comparable to traditional methods requiring paired CT scans.
• The developed model shows promise for improving the diagnosis and management of lung diseases like COPD.