Superior performance of three-dimensional to two-dimensional convolutional neural network for predicting airflow limitation in patients with chronic obstructive pulmonary disease

This study compared two-dimensional (2D) and three-dimensional (3D) convolutional neural networks (CNNs) for predicting airflow limitation in patients with chronic obstructive pulmonary disease (COPD). Models were trained using CT data from the Hokkaido COPD cohort and externally validated with the Kyoto-Himeji COPD cohort. The 3D CNN demonstrated superior performance, achieving a higher area under the curve (AUC) compared to the 2D CNN. It effectively captured CT-derived features such as emphysema and airway wall thickness, which are critical for assessing airflow limitation. Quantitative CT analysis was performed using AVIEW (Coreline Soft), a 3D analysis software that leverages the full CT volume to compute emphysema index and airway measurements. These results highlight the advantage of 3D deep learning approaches in COPD assessment.