CT-Based Pulmonary Segment Delineation: An Imaging Informatics Review of Methods, Benchmarks, Robustness, and Clinical Deployment

Accurate bronchopulmonary segment delineation on computed tomography (CT) is essential for segmentectomy planning, computer-assisted thoracic interventions, and fine-grained quantitative lung analysis. However, segment-level parenchymal partitioning remains technically challenging because intersegmental boundaries are often subtle or indistinct on routine CT, requiring bronchovascular priors-particularly airway and vascular trees-to guide anatomically plausible delineation. This review presents an imaging informatics-based synthesis of CT-based pulmonary segment delineation, covering conventional multi-stage pipelines and recent deep learning frameworks, including convolutional neural networks (CNNs), graph neural networks (GNNs), implicit neural representations, and continuous spatial modeling parameterized by Gaussian mixture models (GMMs). We organize evaluation practices into three complementary categories: direct ground-truth-based metrics, indirect anatomy-referenced criteria, and reference-free plausibility or quality-control measures. We further review publicly available datasets, annotation protocols, and benchmark limitations that currently hinder reproducible and fair segment-level comparison. From a deployment perspective, we discuss clinical workflow requirements, robustness challenges, and the need for anatomy-aware and data-efficient methods. Illustrative analyses of a single cystic fibrosis (CF) case using a distance-based method show how removal of a segmental airway branch or omission of lobar constraints can substantially alter the resulting segment territories. These case-specific observations motivate broader robustness evaluation, uncertainty-aware quality control, boundary-level plausibility checks, and deployment-oriented validation beyond conventional overlap-based metrics.