A computed tomography-based lung radiomics nomogram to identify acute exacerbation of chronic obstructive pulmonary disease: a multi-institutional validation study

Introduction

Chronic obstructive pulmonary disease (COPD) is a chronic, progressive, and irreversible chronic respiratory disease characterized by abnormalities in airway (such as bronchitis and bronchiolitis) and/or alveoli (like emphysema), accompanied by persistent respiratory symptoms including dyspnea, cough, expectoration, and/or acute exacerbations (1). COPD constitutes a major global public-health challenge, marked by high morbidity and mortality (2,3).

In the course of COPD, patients experience worsening respiratory symptoms, a gradual decline in lung function, and even suffer from respiratory failure and systemic inflammatory response syndrome. An acute worsening of respiratory symptoms requiring additional therapy defines an acute exacerbation of chronic obstructive pulmonary disease (AECOPD) (4). AECOPD plays a critical role in the natural history of COPD, accelerating lung-function decline, aggravating comorbidities, eroding quality of life, and driving high morbidity and mortality. Early identification and proper assessment of AECOPD can facilitate timely treatment and rehabilitation, thereby improving quality of life and reducing the risk of hospitalization. Currently, clinical diagnosis and evaluation of AECOPD rely mainly on the level of serum inflammatory factors and pulmonary function indices, and there is a lack of clinically effective quantitative and objective data to assess the degree of acute exacerbation. Although spirometry remains the diagnostic cornerstone for COPD, its limited capacity to characterize structural lung damage hampers precise AECOPD risk stratification. High-resolution computed tomography (CT) uniquely quantifies three-dimensional patterns of disease progression inaccessible to pulmonary function tests, including airway remodeling, emphysema heterogeneity (5), and vascular pruning (6). The “Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstructive Lung Disease 2025” (GOLD 2025) has updated the importance of CT in evaluating COPD, as CT imaging provides critical structural insights into COPD phenotypes. Our previous studies (7-9) have likewise demonstrated the high value of chest CT qualitative, quantitative, and functional parameters in diagnosing COPD, grading its severity, and classifying its phenotypes. However, the pronounced heterogeneity of COPD still precludes effective, comprehensive assessment of lung status, particularly in AECOPD patients. Radiomics represents an emerging methodology that enables high-throughput extraction of quantitative features from medical imaging (e.g., CT), clarifying that while predominantly used in pulmonary oncology (10). This approach captures complex, visually imperceptible patterns within images, demonstrating significant potential for advancing clinical decision-making. Current applications in COPD research are rapidly expanding, with mounting evidence indicating distinct advantages of radiomic analysis for characterizing disease manifestations in this population (11,12). Radiomics utilizes computational machine learning (ML) to extract and quantify sub-visual imaging biomarkers that capture disease heterogeneity associated with the pathogenesis of AECOPD (13), while lung function parameters can objectively quantify ventilation injury. The integration of these two approaches enables a comprehensive view of both structural and functional alterations in AECOPD patients. It is hypothesized that this combined strategy will enhance diagnostic accuracy and clinical utility. Accordingly, this study aims to diagnose and assess AECOPD by combining CT radiomic features with pulmonary function parameters, thereby improving the understanding of disease severity and guiding personalized therapeutic strategies. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-972/rc).

Methods

Patient selection

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committee of the Second Affiliated Hospital of PLA Naval Medical University (approval No. 2022SL068) and individual consent for this retrospective analysis was waived. All participating hospitals were informed and agreed the study. A retrospective analysis was conducted on 343 COPD patients admitted to Zhejiang Provincial People’s Hospital from January 2013 to December 2022, including 158 patients with AECOPD and 185 patients who were not necessarily hospitalized patients. The 343 COPD patients were partitioned into a 240-member training cohort (107 AECOPD/133 non-AECOPD) and a 103-member internal validation cohort (51 AECOPD/52 non-AECOPD). Additionally, 132 patients were included from the Second Affiliated Hospital of PLA Naval Medical University and Harbin Chest Hospital, including 60 patients with AECOPD and 72 patients with non-AECOPD as an external validation set. The inclusion criteria were as follows: (I) the diagnostic criteria of COPD are the forced expiratory volume in one second (FEV₁)/forced vital capacity (FVC) being less than 70%; (II) underwent pulmonary function testing and chest CT scanning at an interval of less than 7 days; (III) had complete routine clinical data. The exclusion criteria were as follows: (I) CT images have artifacts that affect quantitative analysis; (II) history of other lung diseases, such as lung cancer, sarcoidosis, lymphoma, etc.; (III) history of lung surgery. AECOPD is independently diagnosed by two respiratory physicians, using the Rome proposal criteria (14), incorporating quantitative severity grading through symptom scores, inflammatory biomarkers, and airflow limitation progression, while the non-AECOPD group was defined as stable COPD patients without acute exacerbation, who are admitted for follow-up or routine medication, and undergo chest CT scans to assess lung conditions (such as emphysema). The routine clinical data of the patients included age, gender and body mass index (BMI). Based on the 2020 Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines (15), patients were divided into four stages (I–IV) based on their FEV₁/FVC ratio and ratio of FEV₁ to predicted value. Pulmonary function tests included vital capacity (VC), FVC, FEV₁, ratio of FEV₁ to predicted value (FEV₁%pred), FEV₁/FVC ratio, ratio of residual volume (RV) to total lung capacity (TLC), peak expiratory flow (PEF), and maximum ventilatory volume (MVV).

CT examination method and CT image preprocessing

All patients underwent chest CT scans following respiratory training. They were positioned in a supine position with their arms raised above their heads and underwent CT scanning after taking a deep inhalation. The scanning range covered the area from the level of the thoracic inlet to the diaphragm. The scanning parameters were as follows: tube voltage of 120 kV, tube current of 200 mA, collimation of 0.564, matrix of 512×512, layer thickness and spacing of 1 or 2 mm, window width and level of 1,350/−600 HU, standard reconstruction algorithm.

Before extracting the radiomic features, the images were preprocessed, which consisted of three steps. First, spatial standardization via trilinear interpolation to isotropic 1 mm³ × 1 mm³ × 1 mm³ resolution harmonized voxel dimensions across different scanners. Second, intensity discretization with fixed-bin-width quantization ensured gray-level consistency. Third, hybrid multiscale filtering combined Laplacian of Gaussian (LoG) denoising and 3-level Symlet wavelet decomposition separated frequency-specific signatures.

CT image segmentation and radiomics feature extraction

The ITK-Snap software (Version 3.8.0; http://www.itksnap.org/) was employed for semi-automatic whole-lung volume of interest (VOI) segmentation: its 3D Segment tool first auto-detected lung parenchyma, after which manual refinements finalized the VOI. Then, in Python software (Version 3.11.1; https://www.python.org/), the open-source “Pyradiomics” software package was used to extract radiomics features for the included patients. A total of 100 original image features and 1,118 derived features were extracted, with the latter generated using Wavelet and LoG filtering. All features comply with the Image Biomarker Standardization Initiative and were Z-score normalized to remove scale differences.

Features selection and radiomics score construction

Feature selection proceeded in two steps. First, Pearson’s rank correlation coefficient was also used to assess the correlation between features, and one of the features with a correlation coefficient greater than 0.9 between any two features was retained. Second, the least absolute shrinkage and selection operator (LASSO) regression algorithm, with penalty parameter tuning, performed ten-fold cross-validation. The optimal feature dataset with the smallest cross-validation binomial deviation was selected, and the non-zero coefficients were defined as the weight of the selected feature, representing the correlation between the feature and AECOPD. Each patient’s Rad-score was computed as the linear combination of selected features and their coefficients, forming the radiomic model.

Model construction and evaluation

Univariate and multivariate analyses were performed to compare the clinical characteristics between the non-AECOPD and the AECOPD groups. Clinical parameters with P<0.05 were selected to construct a clinical model.

The final retained radiomics features were input into 11 ML classifiers, including logistic regression (LR), NaiveBayes, support vector machine (SVM), K-nearest neighbor (KNN), random forest (RF), extra tree (ET), extreme gradient enhancer (XGBoost), light gradient boosting machine (LightGBM), GradientBoosting, AdaBoost and multi-layer perception (MLP), to develop the Rad-signature for identifying AECOPD patients. The receiver operating characteristic (ROC) curve and Hosmer-Lemeshow (HL) test were employed to select the best-performing model. LR was then used to combine clinical and radiomics features, resulting in an optimal clinical radiomic model. A clinical-radiomics nomogram was subsequently constructed for bedside use. To evaluate the predictive capabilities of the three models, ROC curves were drawn for both the training and validation cohorts, and the average area under the curve (AUC), accuracy, sensitivity, and specificity were calculated. The clinical practicability of the models was evaluated using a calibration curve and decision curve analysis (DCA).

Statistical analysis

Data were analyzed using Python (Version 3.11.1) and R (Version 4.2.3 software). For continuous variables, such as clinical and pulmonary function indicators were expressed by the mean ± standard deviation. The Shapiro-Wilk method was used to test for normality. For normally distributed data, an independent sample t-test was used, whereas for non-normally distributed data, a Mann-Whitney U test was used. Categorical variables were presented as frequencies (percentages) and were analyzed using a chi-square test. P<0.05 was considered statistically significant.

Results

Clinical data and pulmonary function results of patients

This study retrospectively enrolled 532 COPD patients; 475 met the inclusion and exclusion criteria. Among them, 257 non-AECOPD patients (41 female) averaged 70.89±9.92 years, and 218 AECOPD patients (52 female) averaged 69.56±9.85 years. The patient selection flowchart is shown in Figure 1. In the training cohort, there was a significant difference in age (P=0.03) between the non-AECOPD and AECOPD groups. Gender and GOLD grade showed no significant differences between groups in the training or internal-validation cohorts (all P>0.05). In the external validation cohort, there are statistical differences in gender between the non-AECOPD and AECOPD groups (P<0.001). BMI did not differ significantly between non-AECOPD and AECOPD groups across any cohort (all P>0.05). Additionally, significant differences were found in pulmonary function parameters, including FVC, FEV₁, FEV₁/FVC%, PEF and MVV between the non-AECOPD and AECOPD groups in the training cohort (P<0.05) (Table 1).

Figure 1 Flow chart of patient collection. AECOPD, acute exacerbation of chronic obstructive pulmonary disease; COPD, chronic obstructive pulmonary disease; CT, computed tomography; PFT, pulmonary function test.

Construction of Radscore and radiomics model

A total of 1,218 lung radiomic features were extracted from each VOI. After LASSO regression, 61 features with non-zero coefficients were retained (Figure 2). These features and their corresponding coefficients were linearly combined to compute the Radscore, whose formula is provided in the Appendix 1. The 61 selected radiomics features were then utilized to construct a radiomics model for each classifier, and subsequently, the performance of each model was evaluated in the training, internal, and external validation cohorts. Table 2 shows the comparison of AUC values of 11 ML models for identifying AECOPD, and presents statistics on the diagnostic efficacy of the radiomics models constructed by the various classifiers. Notably, the LR radiomics model achieved excellent performance, with AUCs of 0.974 (95% CI: 0.957–0.990), 0.836 (95% CI: 0.759–0.913) and 0.944 (95% CI: 0.902–0.986) in the training, internal and external validation cohorts, respectively. ROC analyses for the different radiomics models across the three cohorts are illustrated in Figure 3.

Figure 2 Radiomics feature selection by using the LASSO LR. (A) Selection of the tuning parameter (λ) in the LASSO model via 10-fold cross-validation based on minimum criteria. Binomial deviances from the LASSO regression cross-validation model are plotted as a function of log(λ). The y-axis shows binomial deviances and the lower x-axis the log(λ). Numbers along the upper x-axis indicate the average number of predictors. Red dots indicate average deviance values for each model with a given λ, and vertical bars through the red dots indicate the upper and lower values of the deviances. The vertical black lines define the optimal values of λ, where the model provides its best fit to the data. (B) The coefficients have been plotted vs. log(λ). (C) The 61 features with nonzero coefficients are shown in the plot. LASSO, least absolute shrinkage and selection operator; LR, logistic regression; MSE, mean square error.

Figure 3 The ROC curves for the 11 ML models in the training (A), internal (B) and external (C) cohorts. Eleven ML models include LR, NaiveBayes, SVM, KNN, RF, ExtraTrees, XGBoost, LightGBM, GradientBoosting, AdaBoost, MLP. The y-axis represents the true positive rate (sensitivity), and the x-axis represents the false positive rate (1-specificity). The diagonal line represents the performance of a random classifier. AUC, area under the curve; CI, confidence interval; KNN, K-nearest neighbor; LightGBM, light gradient boosting machine; LR, logistic regression; ML, machine learning; MLP, multi-layer perceptron; non, not a number; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machine; XGBoost, extreme gradient enhancer.

Establishment and performance of the clinical model and the Radiomics-Clinical model

Given the small sample size, univariate analysis of clinical parameters was performed across all patients. The results revealed that there were significant differences in FEV₁, FVC, VC, FEV₁%pred, FEV₁/FVC%, PEF and MVV between the non-AECOPD and AECOPD cohorts (all P<0.05). Subsequently, multivariate analysis of these six clinical parameters in the training cohort showed that FEV₁ was an independent predictor of AECOPD (P=0.04). Table 3 shows the results of univariate and multivariate analysis.

Clinical predictor FEV₁ was finally combined with the LR-based radscore with high stability and reliability to construct the integrated nomogram in Figure 4A. The HL test indicated good agreement (P>0.05) for the calibration curves of the nomograms identifying AECOPD across the training, internal, and external validation cohorts (Figure 4B-4D). The clinical model performed poorly in the external validation cohort, likely owing to the limited number of independent predictive factors. In all three cohorts, the calibration curves of the clinical radiomics nomogram and the radiomics model fit the diagonal better than the clinical model, reflecting more accurate AECOPD predictions. The DCA (Figure 5A-5C) of the training, internal and external validation models were above the two reference lines, confirming their clinical net benefit.

Figure 4 Radiomics nomogram, calibration curves and DCA curves. (A) The radiomics nomogram, combining FEV₁, LR, was developed in the training cohort. (B-D) The nomogram calibration curves in training (B), internal validation (C) cohorts and external validation (D) cohorts. DCA, decision curve analysis; FEV₁, forced expiratory volume in one second; LR, logistic regression.

Figure 5 The DCA for different models. Decision curve analysis of the training (A), internal (B) and external (C) validation models. DCA, decision curve analysis.

The performance of the three models was compared from the aspects of accuracy, AUC, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), precision, recall and F1. All statistics are listed in Table 4. The clinical model showed an AUC of 0.630 (95% CI: 0.559–0.701) with balanced sensitivity and specificity of 0.393 and 0.850, respectively, in the training cohort. In the internal and external validation cohorts, the AUC was 0.636 (95% CI: 0.526–0.746) and 0.734 (95% CI: 0.645–0.823), respectively. The AUC of radiomics was 0.974 (95% CI: 0.957–0.991) in the training cohort and 0.836 (95% CI: 0.758–0.913) in the internal validation cohort and 0.944 (95% CI: 0.902–0.986) in the external validation cohort, respectively. The AUC of the clinical radiomics nomogram was 0.974 (95% CI: 0.957–0.991) in the training cohort, 0.849 (95% CI: 0.775–0.922) in the internal validation cohort, and 0.957 (95% CI: 0.921–0.993) in the external validation cohort, respectively (Figure 6). The clinical radiomics nomogram delivered the highest AUC in both training and validation cohort and is a reliable tool for clinical identification of AECOPD. An application example of a nomogram is shown in Figure 7. Similar to the points scoring system, we assigned points for each predictor of AECOPD and then equated these predictors with the risk of AECOPD. We can read the top score scale upward from the predictors to determine the points score associated with FEV₁ and the Radscore (LR). Once a score has been assigned to each predictor, an overall score is calculated. Then, the total score is converted to the probability of AECOPD by reading the associated prob ability of AECOPD from the total point scale.

Figure 6 The ROC curves of the clinical, radiomics, and nomogram in the training (A) , internal validation (B) and external validation (C) cohorts are displayed. It is plotted with the true positive rate (sensitivity) on the y-axis against the false positive rate (1-specificity) on the x-axis. The diagonal line represents the performance of a random classifier. Rad radiomics model (LR). AUC, area under the curve; CI, confidence interval; FPR, false positive rate; LR, logistic regression; ROC, receiver operating characteristic; TPR, true positive rate.

Figure 7 The risk scores of COPD in two patients were calculated by using the nomogram. (A,C,E) An 85-year-old woman with a BMI of 20.61 kg/m², Radscore 0.81 of non-AECOPD. Taking the 1 cm layer above the right diaphragm as an example; (A) thin-slice chest CT image; (C) the proportion of emphysema in the middle lobe of the right lung is 7% (blue and purple areas); (E) the nomogram shows that the total score was 51.1 points, corresponding to the probability of developing AECOPD is approximately 1.49%. Lung function examination showed that FEV₁/FVC =0.57, FEV₁ =1.61 L. (B,D,F) A 91-year-old male with BMI 20.45 kg/m², Radscore 0.02 of AECOPD. Taking the 1 cm layer above the right diaphragm as an example; (B) thin-slice chest CT image; (D) the proportion of emphysema in the middle lobe of the right lung is 18% (green, blue and purple areas); (F) the total score of the nomogram was 131, corresponding to the probability of developing AECOPD of approximately 96.2%. Pulmonary function examination showed that FEV₁/FVC =0.51, FEV₁ =1.02 L. *, P<0.05; ***, P<0.001. AECOPD, acute exacerbation of chronic obstructive pulmonary disease; BMI, body mass index; COPD, chronic obstructive pulmonary disease; CT, computed tomography; FEV₁, forced expiratory volume in one second; FEV₁/FVC, ratio of the first second forced expiratory volume to forced vital capacity; LR, logistic regression.

The Delong test was used to compare the AUC of the models. Significant differences were observed between the clinical model and the nomogram, and between the clinical model and the radiomics model, across all three cohorts (all P<0.05). However, no significant difference was detected between the radiomic model and the nomogram model in the training [AUC =0.974 (95% CI: 0.967–0.991) vs. AUC =0.974 (95% CI: 0.956–0.991); P=0.74], internal [AUC =0.836 (95% CI: 0.758–0.913) vs. AUC =0.849 (95% CI: 0.775–0.922); P=0.08] and external validation cohort [AUC =0.944 (95% CI: 0.902–0.986) vs. AUC =0.957 (95% CI 0.921–0.993); P=0.06].

Discussion

This study demonstrated that radiomics features extracted from chest CT images can effectively identify AECOPD, and that the model integrating clinical data offers superior performance, facilitating timely and effective COPD management in clinical practice.

At present, AECOPD diagnosis mainly relies on clinical manifestations, which are highly subjective and lack objective biomarkers (16). While pulmonary function test (PFT) remains the gold standard for diagnosing COPD, it is often challenging to perform satisfactorily in AECOPD patients due to compromised lung function, particularly a reduction in FEV₁. In this study, univariate and multivariate analyses of the training cohort showed that FEV₁ was an independent predictor of AECOPD (P=0.04), a finding consistent with findings from the Korean COPD Subgroup Study (KOCOSS) queue in South Korea (17).

AECOPD is a highly heterogeneous disorder marked by small airway remodeling, small vessel remodeling, and emphysema formation (18). CT scanning is considered to be the most effective method to characterize and quantify the features of COPD (19). Chest CT can visually depict structural changes, delineate their anatomical distribution, and enable individualised phenotyping of airways, vessels, and lung density, which is crucial for early COPD diagnosis and longitudinal evaluation (20). A large number of studies have demonstrated the diagnostic potential of imaging features for AECOPD (21-23). Through lung CT imaging, it has been found that CT imaging measurements can help predict the utilization of subsequent medical services by COPD patients (24). Furthermore, artificial intelligence (AI) has been widely used in the research of COPD (25). CT texture-based radiomics has proven effective in predicting COPD occurrence and severity. The performance of CT images was found to be effective in evaluating the severity classification of COPD. ML is a subfield of AI that utilizes algorithms and statistical models to identify patterns in data, capable of handling complex nonlinear relationships between predictor variables and generating novel results (26). Compared with traditional models, ML handles large, heterogeneous datasets more robustly, delivering accurate risk predictions even in the presence of noise or missing data, and obviates stringent preprocessing (27). While traditional early screening and risk assessment for AECOPD rely on direct clinical observation of respiratory symptoms and pulmonary function tests, along with evaluations of known risk factors, these methods are limited in handling multi-source data and capturing complex disease patterns, necessitating the use of ML to construct early screening models for AECOPD. Li et al. applied CT-based radiomics to the identification of COPD, SVM and LR ML algorithms were used, the AUCs of SVM and LR models in the training set were 0.992 and 0.993, respectively (28). Numerous machine-learning algorithms exist, yet which yields optimal performance remains unclear (29). In this study, 61 radiomics features were analyzed to construct a diagnostic ML model for AECOPD. Across eleven algorithms, the LR model achieved the highest diagnostic efficiency, with the training cohort AUC of 0.974 (95% CI: 0.957–0.990), internal validation cohort AUC of 0.836 (95% CI: 0.759–0.913), and an external validation cohort AUC of 0.944 (95% CI: 0.902–0.986). Consequently, the LR ML algorithm was selected and the corresponding Rad-score calculated from CT images for further analysis.

Based on clinical data and pulmonary function parameters, regression analysis can be used to evaluate the risk of AECOPD with moderate power. A study indicated that texture features based on chest CT images are more effective in predicting AECOPD patients than clinical features such as BMI, airflow Obstruction, Dyspnea and Exercise capacity (BODE), biomechanics, or acute exacerbation history (30). To compare effectiveness, a clinical model was constructed from meaningful parameters. Its AUCs for identifying AECOPD were 0.630 (95% CI: 0.559–0.701) in the training cohort, 0.636 (95% CI: 0.526–0.746) in the internal validation cohort, and 0.734 (95% CI: 0.645–0.823) in the external validation cohort, all lower than the corresponding radiomics model. Preliminary research showed that the COPD identification model combining CT with meaningful clinical parameters outperformed the model built solely on CT imaging radiomics. A study found that a multi-modal data combination strategy based on chest high-resolution CT (HRCT) images and pulmonary function parameters can intelligently identify respiratory distress complications in COPD patients (31). Therefore, we also explored the combination of radiomics and integrating statistically significant pulmonary function parameters and Rad-score values to construct a comprehensive discrimination model, and found that the AUC values of the training cohort (AUC is 0.974), internal validation cohort (AUC is 0.849), and external validation (AUC is 0.957) cohort of this model were relatively higher than those of the rad-score and radiomics model of 0.974 (95% CI: 0.957–0.991), 0.836 (95% CI: 0.758–0.913), and 0.944 (95% CI: 0.902–0.986), respectively. However, the diagnostic performance of the Radiomics-Clinical model in this study still has room for improvement, possibly because routine blood examination indicators and cardiac function parameters were not included. Accumulating clinical evidence shows that specific indicators derived from routine complete blood count, such as the neutrophil-to-lymphocyte ratio and platelet-to-lymphocyte ratio, can significantly predict adverse outcomes in AECOPD patients (32). A neutrophil-to-lymphocyte ratio greater than 6.90 is considered an important biomarker for predicting in-hospital mortality in AECOPD patients (33).

There are some limitations in this study. Firstly, this study enrolled 475 patients, and after calculation, the sample size was sufficient. The gender imbalance in our external validation cohort (76.5% male), while reflective of real-world COPD epidemiology, may affect generalizability to female populations. However, in order to further enhance the generalizability of the model among different genders and ethnic groups, we will further validate it in public databases. Secondly, at present, this study focuses on independent validation of radiomics and does not include blood inflammatory indicators, aiming to lay the foundation for its mechanism specificity and avoid scientific ambiguity caused by premature integration. In future research, we will systematically evaluate the synergistic potential of combining radiomics with biomarkers to optimize the identification of high-risk populations for AECOPD. Thirdly, this study analyzed CT images of COPD patients during acute exacerbation to assist in diagnosis, but further exploration of stable baseline CT or inter-stage comparisons is needed. Fourthly, quantitative CT analysis is important for COPD, yet this study did not compare CT quantitative data with image-based radiomics models. Therefore, additional laboratory tests, stable baseline CT examinations, and CT quantitative parameters should be incorporated in future work to predict AECOPD risk. Lastly, it is crucial to seamlessly integrate the column chart into existing electronic health record systems and validate the generality and robustness of the model in clinical settings. We will also conduct further research and validation in the future.

Conclusions

In summary, the combination strategy of clinical data, pulmonary function parameters, and CT radiomics facilitates AECOPD diagnosis and furnishes objective, quantitative evidence for formulating clinical treatment strategies.

Acknowledgments

We thank the patients who participated in this study.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-972/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-972/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-972/prf

Funding: This work was supported by Naval Medical University Campus Level Project (No. 2023QN074, to X.Z.), National Natural Science Foundation of China (No. 82171926, to L.F., No. 81930049, to S.L.), National Key R&D Program of China (No. 2022YFC2010000, to S.L., No. 2022YFC2010002, to L.F., No. 2022YFC2010005, to X.Z.), The Program of Science and Technology Commission of Shanghai Municipality (No. 21DZ2202600, to L.F.), The Construction of CT Standardized Database for Chronic Obstructive Pulmonary Disease (No. YXFSC2022JJSJ002, to L.F.), The Clinical Innovation Project of Shanghai Changzheng Hospital (No. 2020YLCYJ-Y24, to L.F.), Zhejiang Province Medical and Health Technology Plan Project (No. 2022KY040, 2023KY472, to Y.M.), and Zhejiang Provincial Natural Science Foundation of China (No. LTGY24H180017, to Y.M.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-972/coif). X.Z. reports receiving funding from the Naval Medical University Campus Level Project (No. 2023QN074) and National Key R&D Program of China (No. 2022YFC2010005). Y.M. reports receiving funding from the Zhejiang Province Medical and Health Technology Plan Project (Nos. 2022KY040, 2023KY472) and Zhejiang Provincial Natural Science Foundation of China (No. LTGY24H180017). S.L. reports receiving funding from the National Natural Science Foundation of China (No. 81930049) and National Key R&D Program of China (No. 2022YFC2010000). L.F. reports receiving funding from the National Natural Science Foundation of China (No. 82171926), National Key R&D Program of China (No. 2022YFC2010002), the Program of Science and Technology Commission of Shanghai Municipality (No. 21DZ2202600), Construction of CT Standardized Database for Chronic Obstructive Pulmonary Disease (No. YXFSC2022JJSJ002), and Clinical Innovation Project of Shanghai Changzheng Hospital (No. 2020YLCYJ-Y24). The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committee of The Second Affiliated Hospital of PLA Naval Medical University (approval No. 2022SL068) and individual consent for this retrospective analysis was waived. All participating hospitals were informed and agreed the study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Agustí A, Celli BR, Criner GJ, et al. Global Initiative for Chronic Obstructive Lung Disease 2023 Report: GOLD Executive Summary. Am J Respir Crit Care Med 2023;207:819-37. [Crossref] [PubMed]
2. Deolmi M, Decarolis NM, Motta M, et al. Early Origins of Chronic Obstructive Pulmonary Disease: Prenatal and Early Life Risk Factors. Int J Environ Res Public Health 2023;20:2294. [Crossref] [PubMed]
3. Kwok WC, Tam TCC, Ho JCM, et al. Hospitalized acute exacerbation in chronic obstructive pulmonary disease - impact on long-term renal outcomes. Respir Res 2024;25:36. [Crossref] [PubMed]
4. Sharma M, Joshi S, Banjade P, et al. Global Initiative for Chronic Obstructive Lung Disease (GOLD) 2023 Guidelines Reviewed. Open Respir Med J 2024;18:e18743064279064. [Crossref] [PubMed]
5. Yuan NF, Hasenstab K, Retson T, et al. Unsupervised Learning Identifies Computed Tomographic Measurements as Primary Drivers of Progression, Exacerbation, and Mortality in Chronic Obstructive Pulmonary Disease. Ann Am Thorac Soc 2022;19:1993-2002. [Crossref] [PubMed]
6. Yang T, Chen C, Chen Z. The CT pulmonary vascular parameters and disease severity in COPD patients on acute exacerbation: a correlation analysis. BMC Pulm Med 2021;21:34. [Crossref] [PubMed]
7. Zhang D, Zhao M, Zhou X, et al. Single Inspiratory Chest CT-based Generative Deep Learning Models to Evaluate Functional Small Airways Disease. Radiol Artif Intell 2025;7:e240680. [Crossref] [PubMed]
8. Pu Y, Zhou X, Zhang D, et al. Quantitative Assessment Characteristics of Small Pulmonary Vessel Remodelling in Populations at High Risk for COPD and Smokers Using Low-Dose CT. Int J Chron Obstruct Pulmon Dis 2024;19:51-62. [Crossref] [PubMed]
9. Zhou X, Pu Y, Zhang D, et al. Development of machine learning model to predict pulmonary function with low-dose CT-derived parameter response mapping in a community-based chest screening cohort. J Appl Clin Med Phys 2023;24:e14171. [Crossref] [PubMed]
10. Russo L, Charles-Davies D, Bottazzi S, et al. Radiomics for clinical decision support in radiation oncology. Clin Oncol (R Coll Radiol) 2024;36:e269-81. [Crossref] [PubMed]
11. Zhou T, Guan Y, Lin X, et al. CT-based whole lung radiomics nomogram for identification of PRISm from non-COPD subjects. Respir Res 2024;25:329. [Crossref] [PubMed]
12. Koo MC, Au R, Hague CJ, et al. Expiration CT Gas Trapping Measures with Texture-Based Radiomics Improves Association with Lung Function and Lung Function Decline in COPD. Acad Radiol 2025;32:3711-21. [Crossref] [PubMed]
13. Zhou TH, Zhou XX, Ni J, et al. CT whole lung radiomic nomogram: a potential biomarker for lung function evaluation and identification of COPD. Mil Med Res 2024;11:14. [Crossref] [PubMed]
14. Celli BR, Fabbri LM, Aaron SD, et al. An Updated Definition and Severity Classification of Chronic Obstructive Pulmonary Disease Exacerbations: The Rome Proposal. Am J Respir Crit Care Med 2021;204:1251-8. [Crossref] [PubMed]
15. Global Initiative for Chronic Obstructive Lung Disease. Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstructive Pulmonary Disease: 2020 Report. (2019-11-05) [2019-11-30]. Available online: https://goldcopd.org/gold-reports/
16. Pratt AJ, Purssell A, Zhang T, et al. Complexity in clinical diagnoses of acute exacerbation of chronic obstructive pulmonary disease. BMC Pulm Med 2023;23:298. [Crossref] [PubMed]
17. Choi JY, Lee CH, Joo H, et al. Implications of Global Lung Function Initiative Spirometry Reference Equations in Northeast Asian Patients With COPD. Chest 2025;167:414-24. [Crossref] [PubMed]
18. Ritchie AI, Wedzicha JA. Definition, Causes, Pathogenesis, and Consequences of Chronic Obstructive Pulmonary Disease Exacerbations. Clin Chest Med 2020;41:421-38. [Crossref] [PubMed]
19. Baraghoshi D, Strand M, Humphries SM, et al. Quantitative CT Evaluation of Emphysema Progression over 10 Years in the COPDGene Study. Radiology 2023;307:e222786. [Crossref] [PubMed]
20. Iwasawa T, Matsushita S, Hirayama M, et al. Quantitative Analysis for Lung Disease on Thin-Section CT. Diagnostics (Basel) 2023;13:2988. [Crossref] [PubMed]
21. Chaudhary MFA, Hoffman EA, Guo J, et al. Predicting severe chronic obstructive pulmonary disease exacerbations using quantitative CT: a retrospective model development and external validation study. Lancet Digit Health 2023;5:e83-92. [Crossref] [PubMed]
22. Rangelov BA, Young AL, Jacob J, et al. Thoracic Imaging at Exacerbation of Chronic Obstructive Pulmonary Disease: A Systematic Review. Int J Chron Obstruct Pulmon Dis 2020;15:1751-87. [Crossref] [PubMed]
23. Hua Q, Chen G, Yang Y, et al. Quantitative Evaluation of Chronic Obstructive Pulmonary Disease and Risk Prediction of Acute Exacerbation by High-Resolution Computed Tomography. Evid Based Complement Alternat Med 2022;2022:6015766. [Crossref] [PubMed]
24. Moslemi A, Makimoto K, Tan WC, et al. Quantitative CT Lung Imaging and Machine Learning Improves Prediction of Emergency Room Visits and Hospitalizations in COPD. Acad Radiol 2023;30:707-16. [Crossref] [PubMed]
25. Kaplan A, Cao H, FitzGerald JM, et al. Artificial Intelligence/Machine Learning in Respiratory Medicine and Potential Role in Asthma and COPD Diagnosis. J Allergy Clin Immunol Pract 2021;9:2255-61. [Crossref] [PubMed]
26. Sarker IH. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN Comput Sci 2021;2:160. [Crossref] [PubMed]
27. Coates JT, de Koning C. Machine learning-driven critical care decision making. J R Soc Med 2022;115:236-8. [Crossref] [PubMed]
28. Li Z, Liu L, Zhang Z, et al. A Novel CT-Based Radiomics Features Analysis for Identification and Severity Staging of COPD. Acad Radiol 2022;29:663-73. [Crossref] [PubMed]
29. Makimoto K, Au R, Moslemi A, et al. Comparison of Feature Selection Methods and Machine Learning Classifiers for Predicting Chronic Obstructive Pulmonary Disease Using Texture-Based CT Lung Radiomic Features. Acad Radiol 2023;30:900-10. [Crossref] [PubMed]
30. Chaudhary MFA, Hoffman EA, Comellas AP, et al. CT Texture Features Predict Severe COPD Exacerbations in SPIROMICS. C5 C005 Diagn. Progn. COPD, American Thoracic Society; 2021:A1122. Available online: https://www.atsjournals.org/doi/pdf/10.1164/ajrccm-conference.2021.203.1_MeetingAbstracts.A1122
31. Yang Y, Chen Z, Li W, et al. Multi-modal data combination strategy based on chest HRCT images and PFT parameters for intelligent dyspnea identification in COPD. Front Med (Lausanne) 2022;9:980950. [Crossref] [PubMed]
32. Zinellu A, Zinellu E, Mangoni AA, et al. Clinical significance of the neutrophil-to-lymphocyte ratio and platelet-to-lymphocyte ratio in acute exacerbations of COPD: present and future. Eur Respir Rev 2022;31:220095. [Crossref] [PubMed]
33. Emami Ardestani M, Alavi-Naeini N. Evaluation of the relationship of neutrophil-to lymphocyte ratio and platelet-to-lymphocyte ratio with in-hospital mortality in patients with acute exacerbation of chronic obstructive pulmonary disease. Clin Respir J 2021;15:382-8. [Crossref] [PubMed]