# Habitat imaging radiomics of chest CT identifies noninfectious acute exacerbations in chronic obstructive pulmonary disease

## TL;DR

This study uses CT scans and radiomic analysis to identify noninfectious acute exacerbations in COPD patients with high accuracy.

## Contribution

A novel CT-based habitat imaging radiomic model is developed for diagnosing noninfectious AECOPD.

## Key findings

- The habitat-total model achieved an AUC of 0.800 (LR) and 0.807 (SVM) in the test cohort for identifying noninfectious AECOPD.
- Habitat-total score and GOLD stage were identified as independent predictors of noninfectious AECOPD.
- The model segments whole-lung CT scans into three habitat subregions for analysis.

## Abstract

Noninfectious acute exacerbations of chronic obstructive pulmonary disease (AECOPD) pose significant diagnostic challenges due to the lack of reliable biomarkers. This study aims to develop and validate a CT-based habitat imaging radiomic model for precise identification of noninfectious AECOPD.

This retrospective study included 352 eligible chronic obstructive pulmonary disease (COPD) patients who received treatment at Tianjin Chest Hospital from January 2019 to December 2023. Among these patients (181 with noninfectious AECOPD, 171 with stable COPD), stratified randomization allocated cohorts to training (n = 211) and testing (n = 141) cohorts. Whole-lung CT scans were subjected to habitat mapping by voxel-wise K-means clustering, with radiomic features derived from habitat subregions and optimized using least absolute shrinkage and selection operator regression. Logistic regression (LR) and support vector machine (SVM) models combined habitat-derived traits with clinical factors.

The CT-based whole lung was segmented into three habitat subregions: habitat subregion 1 (emphysema/bullae-associated), habitat subregion 2 (bronchovascular bundle), and habitat subregion 3 (lung parenchyma). The habitattotal model showed predictive power for identifying noninfectious AECOPD (training: AUC = 0.853 [LR], 0.897 [SVM]; test: AUC = 0.800 [LR], 0.807 [SVM]). Multivariate analysis identified habitattotal score and GOLD stage as independent predictors of noninfectious AECOPD (p < 0.001).

In conclusion, this study segmented whole-lung CT scans of noninfectious AECOPD patients into habitat subregions, developing a radiomics model that demonstrated strong diagnostic efficacy. This approach provides an objective imaging biomarker and a potential tool for quantifying COPD heterogeneity.