An applied study on the assessment of concomitant pulmonary hypertension in patients with chronic obstructive pulmonary disease based on whole-lung CT imaging histologic features

Abstract

Background

Pulmonary hypertension (PH) is a common complication in the progression of chronic obstructive pulmonary disease (COPD) and plays a pivotal role in the advancement of chronic pulmonary heart disease. This condition is closely associated with acute exacerbations and poor prognosis in COPD patients. Currently, methods used in clinical practice to assess PH each have certain limitations. This study aimed to investigate the diagnostic value of whole-lung computed tomography (CT)-based Radiomics features in identifying PH in patients with COPD.

Methods

A total of 171 patients diagnosed with COPD by clinical and pulmonary function testing at the First Affiliated Hospital of Hebei North University between August 2022 and November 2024 were retrospectively enrolled and randomly divided into a training cohort (n=119) and a validation cohort (n=52) at a 7:3 ratio. Clinical data including demographic characteristics, hematological parameters, and coagulation profiles were collected. On axial CT images, the diameters of the main pulmonary artery (MPA) and ascending aortic (AA) were measured, and the MPA/AA ratio was calculated. SPSS.27.0 statistical software was used to perform univariate logistic regression to initially screen potentially significant variables, and multifactorial logistic analysis to screen independent risk factors for concomitant PH in COPD patients. Radiomics features were extracted from the entire lung using the three-dimensional (3D) Slicer Radiomics plug-in, and the feature data were normalized and downscaled, and then the most informative features were screened using the Random Forest (RF) algorithm. Clinical, MPA/AA, Radiomics, and joint models were constructed using R software, and their diagnostic performance was compared using receiver operating characteristic (ROC) curves, calibration plots, and decision curves analysis (DCA).

Results

Multivariate logistic regression identified D-dimer [P=0.008; odds ratio (OR) =2.404, 95% confidence interval (CI): 1.264–4.572], and MPA/AA ratio (P=0.044; OR =2.249, 95% CI: 1.021–4.955) as independent risk factors for PH in patients with COPD. Among the 1,168 extracted Radiomics features, five were selected via RF as key predictors. The ROC curves analysis demonstrated that the combined model exhibited significantly superior diagnostic performance compared to the clinical model, the MPA/AA model, and the Radiomics model in both the training cohort area under the ROC curve (AUC =0.884) and the validation cohort (AUC =0.874). Calibration plots, Hosmer-Lemeshow test, and DCA confirmed the superior clinical utility of the combined model. A nomogram was developed to provide intuitive visualization of individual predictor contributions.

Conclusions

This combined model integrates CT imaging features, D-dimer levels, and the MPA/AA ratio, enabling noninvasive identification of COPD patients with concomitant PH. It serves as an effective tool for clinical diagnostic support.

Highlight box.

Key findings

• Whole-lung radiomic features extracted from chronic obstructive pulmonary disease (COPD) patients represent an independent risk factor for the development of pulmonary hypertension (PH) in COPD patients.

• The combined model integrating computed tomography (CT) radiomics features with main pulmonary artery (MPA)/ascending aortic (AA) and D-dimer demonstrated superior diagnostic performance in assessing PH in COPD patients. This model significantly outperformed clinical models, MPA/AA models, and radiomics models in both the training cohort [area under the curve (AUC) =0.884] and validation cohort (AUC =0.874).

What is known and what is new?

• Right heart catheterization (RHC) and echocardiography are the standard clinical methods for evaluating PH in patients with COPD. Although RHC is considered the gold standard for diagnosing PH, its invasiveness limits widespread application. Echocardiography aids diagnosis by measuring tricuspid regurgitation velocity and assessing atrial and ventricular dimensions; however, operator dependency leads to measurement variability, thereby affecting diagnostic reliability.

• Whole-lung radiomic features in COPD patients can effectively assess the risk of PH. First-order features and texture features were identified as key indicators for PH risk. A combined model incorporating MPA/AA and D-dimer enables non-invasive identification of COPD patients with PH, serving as an effective tool for clinical diagnostic support.

What is the implication, and what should change now?

• This combined model provides clinicians with a rapid, cost-effective, and non-invasive tool for concurrently assessing the risk of concomitant PH during routine chest CT scans. By identifying high-risk patients early, this approach facilitates timely intervention and optimized treatment strategies to halt disease progression, thereby reducing the financial burden on patients.

Introduction

Chronic obstructive pulmonary disease (COPD) is a common chronic respiratory disease characterized by persistent airflow limitation, which can progressively lead to cor pulmonale and respiratory failure. Pulmonary hypertension (PH) is a frequent complication in the development of COPD and plays a critical role in the progression to chronic cor pulmonale. It is closely related to acute exacerbations and poor prognosis in COPD patients. If not promptly and properly managed, PH poses a serious threat to the health and survival of individuals with COPD (1,2).

Right heart catheterization (RHC) is considered the gold standard for diagnosing PH (3); however, its invasive nature limits widespread clinical application. Transthoracic echocardiography is one of the most commonly used noninvasive methods, aiding in PH diagnosis by measuring systolic tricuspid regurgitation velocity and assessing atrial and ventricular size and function. Nevertheless, echocardiographic assessments are subject to operator-dependent variability, which may lead to measurement bias and reduced reliability (4). Therefore, the development of non-invasive, accurate and standardized diagnostic approaches for PH holds significant clinical value.

Computed tomography (CT) is a standard method of clinical evaluation of patients with COPD, as it not only accurately identifies structural changes in the airways and parenchyma, but also allows assessment of the pulmonary vasculature (5). A previous study has confirmed a significant correlation between CT-measured main pulmonary artery (MPA) diameter and PH, and this metric has been widely incorporated into clinical guidelines for CT scanning (6). Imaging histology is used to identify a wide range of thoracic diseases and for prognostic assessment by non-invasively capturing phenotypic features that are not recognizable by the human eye and that reflect the diversity within a lesion (7-11). For COPD patients, automated extraction and quantification of whole-lung radiomic features provide valuable insight into the disease’s pathophysiology. Analyzing lung texture and structural features through Radiomics can provide more accurate diagnostic and staging results than conventional CT, providing an important reference for this study (12,13).

In this study, we extracted radiomic features from routine chest CT images and combined them with the conventional clinical indicators such as the MPA-to-ascending aorta diameter ratio and relevant clinical data. Our goal was to construct a diagnostic model for identifying COPD patients with concomitant PH at an early stage, enabling timely intervention and ultimately improving patient quality of life and prognosis. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1998/rc).

Methods

Study subjects and subgroups

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of the First Affiliated Hospital of Hebei North University (No. k2024257; Zhangjiakou, China). In view of the retrospective nature of the present study, the local institutional review board waived the requirement for informed consent. Finally, a total of 171 patients diagnosed with COPD based on clinical assessment and pulmonary function tests from August 2022 to November 2024 were enrolled (screening flowchart shown in Figure 1).

Figure 1.

Flowchart for screening research subjects. COPD, chronic obstructive pulmonary disease; CT, computed tomography.

Inclusion criteria: (I) adult patients diagnosed with COPD by clinical and pulmonary function testing; (II) underwent pulmonary function testing, chest CT, and echocardiography within 2 weeks (14).

Exclusion criteria: (I) history of congenital heart disease, valvular disease, pulmonary vascular malformations, or presence of cardiac pacemakers or defibrillators; (II) coexisting pulmonary conditions such as pulmonary tuberculosis, atelectasis, lung tumors, high-risk lung nodules larger than 6 mm (14), chronic pulmonary embolism, or pleural effusion; (III) PH attributable to causes other than COPD; (IV) history of lung surgery or severe thoracic trauma; (V) history of malignancy; (VI) incomplete clinical data or non-assessable imaging quality.

All patients were randomly assigned to either a training set (n=119) or validation set (n=52) at a 7:3 ratio. Based on the presence or absence of PH, patients were categorized into a COPD without PH group and a COPD with PH group.

Diagnostic criteria for COPD and PH

COPD diagnostic criteria

According to the 2023 Global Initiative for Chronic Obstructive Lung Disease (GOLD) strategy for the diagnosis, treatment, and prevention of COPD (15), a diagnosis is made in individuals with persistent respiratory symptoms such as chronic cough, sputum production, and wheezing, who have a history of exposure to risk factors, Spirometric confirmation is required, defined as a post-bronchodilator forced expiratory volume in 1 second/forced vital capacity (FEV₁/FVC) <70%, indicating persistent airflow limitation.

PH diagnostic criteria

Based on the 2022 ESC/ERS Guidelines for the Diagnosis and Treatment of Pulmonary Hypertension (16). PH is suspected when the echocardiographic tricuspid regurgitation velocity was exceeds 2.8 m/s. Using the simplified Bernoulli equation, pulmonary arterial systolic pressure (PASP) is estimated. A PASP ≥36 mmHg is considered diagnostic for PH.

Clinical and laboratory data collection

Detailed clinical data were collected, including patient age, sex, history of hypertension, diabetes mellitus, coronary artery disease, history of smoking and alcohol consumption. Within 24 hours of hospital admission, the first fasting venous blood sample was drawn to complete the blood routine, including white blood cell count, absolute neutrophil count, absolute lymphocyte count, red cell distribution width, and plateletcrit. The neutrophil-to-lymphocyte ratio (NLR) was calculated. Coagulation function tests were also performed, including measurements of D-dimer and fibrinogen levels.

CT scanning parameters and methods

A Toshiba Aquilion 320-row spiral CT scanner was used Patients were positioned supine with arms raised overhead, entering the scanner head-first. Scanning parameters were as follows: tube voltage 120 kV with automatic tube current modulation; pitch 0.175 mm, collimation 1.2 mm, and field of view (FOV) 380 mm × 380 mm. Image reconstruction was performed at a 1mm slice thickness. The scanning range extended from the lung apex to the lung base, covering the entire pulmonary region, including both pleural cavities. All scans were performed during a breath-hold at the end of maximal inspiration.

MPA/ascending aortic (AA) measurement methods

Measurements were performed on axial mediastinal window images from CT scans (Figure 2). The diameter of the MPA was measured at its widest point within 3 mm of the bifurcation level. The diameter of the AA was measured at the same axial level as the MPA, using its shortest transverse diameter. The MPA/AA was then calculated. All measurements were independently conducted by two radiologists. The average of the two measurements was used for analysis. If there was a significant discrepancy between the two results, a senior radiologist remeasured the parameters to ensure accuracy and consistency.

Figure 2.

Measurement of main pulmonary artery trunk and ascending aortic diameter.

Image segmentation, image histology feature extraction and screening

Non-contrast chest CT images in the lung window were exported in DICOM format. Whole-lung the regions of interest (ROIs) were automatically segmented using the 3D-Slicer (http://www.slicer.org) software. Two experienced thoracic radiologists (with 5 years of experience each) reviewed the segmentation results, and any inaccuracies were manually corrected (Figure 3). Throughout the entire process of defining ROIs, extracting texture features, and collecting clinical indicators, this study implemented a blinded design for patient names, PH comorbidity status, and other assessment factors. Specifically, personnel involved in these operations were unaware of whether patients had PH comorbidity and could not access information about other predictive factors when evaluating a specific predictor. This design effectively prevents cross-contamination of assessment information and interference from other factors, ensuring the objectivity of both the evaluation process and its outcomes.

Figure 3.

Whole-lung ROI outlining methodology. ROI, region of interest.

Quantitative image histology features were extracted using the Slicer Radiomics plug-in. The extracted texture feature was standardized using Z-score normalization with R software (https://www.r-project.org/). First, dimensionality reduction was performed on the training dataset. Perform dimensionality reduction on features within the training set. Utilize correlation analysis to examine the relationships between feature parameters. For features with a correlation coefficient >0.9, retain only one of the two. A Random Forest (RF) model was then constructed using the dimensionally reduced training set. The model was built by setting a random seed and performing 100 rounds of random permutations to assess feature importance. Finally, radiomic features with P values <0.05 were retained for model construction. A complete case analysis strategy was adopted, utilizing available data from 171 patients without imputation of missing values.

The workflow for this predictive model consists of four steps: First, collect baseline information, initial clinical laboratory test results, and imaging data from patients with COPD at hospital admission. Second, extract radiomic features from the initial imaging studies performed prior to admission. Third, input the extracted radiomic features, clinical laboratory results, and baseline information into the model. Fourth, the model performs calculations to determine whether the patient has concomitant PH.

Model construction and validation

Multifactorial logistic regression analysis was applied to screen patient baseline characteristics, complete blood count, coagulation function, and MPA/AA parameters to identify independent risk factors and construct clinical models as well as MPA/AA models. A Radiomics model was developed using logistic regression based on the retained Radiomics features. To verify diagnostic performance, the clinical model, MPA/AA model and Radiomics model were further integrated to establish a joint model. Receiver operating characteristic (ROC) curves were used to evaluate the diagnostic efficacy of each model. The Hosmer-Lemeshow goodness-of-fit test and calibration curves were used to assess the fitting effect and predictive accuracy of each model in diagnosing COPD with PH. Clinical decision curves analysis (DCA) was used to assess the clinical utility of each model. In addition, a nomogram was developed to provide a visual representation of the prediction results.

Statistical analysis

This study employed SPSS 27.0 software for statistical analysis, with the following specific methods: The Shapiro-Wilk test was used to assess the normality of quantitative data. Normally distributed data were described as “mean ± standard deviation”, while non-normally distributed data were expressed as “median (interquartile range)”. Intergroup comparisons utilized either the independent samples t-test or nonparametric tests based on data distribution characteristics. Categorical data were presented as “proportions”, with intergroup comparisons performed using chi-square tests or Fisher’s exact tests. Logistic regression analysis explored the association between various indicators in COPD patients and the outcome of concomitant PH. The predictive performance of the model was evaluated using ROC curve analysis. Correlation analysis employed Pearson’s correlation test or Spearman’s correlation test based on variable type. Statistical significance was set at P<0.05.

Results

Basic clinical data, laboratory data, and imaging measurements of patients

A total of 119 patients with COPD were enrolled in the training set, including 56 patients (47.1%) with comorbid PH and 63 patients (53.0%) without PH. The validation set included 52 patients with COPD, of whom 31 (59.6%) had PH and 21 (40.4%) did not. In the training set, significant differences were observed between COPD patients with and without PH in terms of white blood cell count (P=0.049), plateletcrit (P=0.004), red blood cell distribution width-standard deviation (RDW-SD) (P<0.001), red blood cell distribution width-coefficient of variation (RDW-CV) (P<0.001), D-dimer (P<0.001), and MPA/AA ratio (P<0.001). Among these, plateletcrit and the MPA/AA ratio showed statistically significant differences in both the training set (P=0.004; P<0.001) and the validation set (P=0.02; P<0.001). However, the differences in other indicators were not statistically significant in the validation set (P>0.05) (Table 1).

Table 1. Clinical data, laboratory data and imaging measures of patients in the training and validation sets.

Factors	Training cohort (n=119)				Validation cohort (n=52)
	COPD without PH (n=63)	COPD with PH (n=56)	P value		COPD without PH (n=21)	COPD with PH (n=31)	P value
Gender			0.51				0.28
Male	71.4	76.8			71.4	83.9
Female	28.6	23.2			28.6	16.1
Age (years)	68.25±7.47	69.50±7.17	0.36		71.14±6.72	74.26±7.60	0.14
Hypertension			0.60				0.60
Yes	34.9	30.4			28.6	35.5
No	65.1	69.6			71.4	64.5
Diabetes			0.95				0.51
Yes	11.1	10.7			4.8	9.7
No	88.9	89.3			95.2	90.3
Asthma			0.60				0.24
Yes	6.3	8.9			0	6.5
No	93.7	91.1			100	93.5
Smoking history			0.55				0.83
Yes	68.3	73.2			71.4	74.2
No	31.7	26.8			28.6	25.8
Drinking history			0.17				0.52
Yes	55.6	42.9			47.6	38.7
No	44.4	57.1			52.4	61.3
White cell count (109/L)	7.20 (4.29)	6.52±1.79	0.049		7.26±2.22	6.90±1.73	0.52
Blood platelet ratio (%)	0.21 (0.09)	0.18 (0.08)	0.004		0.21±0.44	1.74±0.53	0.02
Red blood cell width SD (fL)	46.0 (5.30)	46.68±4.19	<0.001		46.32±4.08	48.20 (7.00)	0.28
Red blood cell width CV (%)	13.17±0.80	13.80 (1.97)	<0.001		13.41±1.35	13.5 (1.80)	0.44
D-dimer (mg/L)	0.29 (0.40)	0.90 (0.95)	<0.001		0.37 (0.25)	0.62 (0.41)	0.008
Fibrinogen content (g/L)	3.68 (1.98)	3.26 (1.92)	0.11		3.44±1.21	2.80 (1.67)	0.54
NLR	3.28 (4.88)	3.40 (1.96)	0.58		3.97 (3.25)	3.12 (2.67)	0.96
MPA/AA	0.86±0.11	1.03 (0.15)	<0.001		0.91±0.07	1.02 (0.14)	<0.001

Normally distributed data are expressed as mean ± standard deviation; non-normally distributed data are expressed as median (interquartile range); categorical data are expressed as proportion. AA, ascending aortic; COPD, chronic obstructive pulmonary disease; CV, coefficient of variation; MPA, main pulmonary artery; NLR, neutrophil-to-lymphocyte ratio; PH, pulmonary hypertension; SD, standard deviation.

Screening for independent risk factors

Univariate logistic regression analysis showed that white blood cell count (P=0.01), plateletcrit (P=0.04), RDW-SD (P<0.001), RDW-CV (P<0.001), D-dimer (P=0.01), and MPA/AA ratio (P<0.001) were the significantly associated with the presence of PH in COPD patients. Multivariate logistic regression analysis further identified D-dimer (P=0.008) and the MPA/AA ratio (P=0.044) as independent risk factors for PH in COPD patients (Table 2).

Table 2. Comparison of laboratory indices between the two groups in the training set.

Variant	One-way logistic analysis			Multi-factor logistic analysis
	OR (95% CI)	P value		OR (95% CI)	P value
White cell count	0.601 (0.401–0.901)	0.01		–	–
Blood platelet ratio	0.308 (0.102–0.928)	0.04		–	–
Red blood cell width SD	2.587 (1.621–4.131)	<0.001		–	–
Red blood cell width CV	3.540 (1.773–7.066)	<0.001		–	–
D-dimer	2.225 (1.178–4.203)	0.01		2.404 (1.264–4.572)	0.008
Fibrinogen content	0.739 (0.507–1.078)	0.12		–	–
NLR	0.822 (0.543–1.243)	0.35		–	–
MPA/AA	4.327 (2.508–7.467)	<0.001		2.249 (1.021–4.955)	0.044

AA, ascending aortic; CI, confidence interval; CV, coefficient of variation; MPA, main pulmonary artery; NLR, neutrophil-to-lymphocyte ratio; OR, odds ratio; SD, standard deviation.

Screening of imaging histologic features

In this study, a total of 1,168 Radiomics features were initially extracted from the training set using 3D Slicer Radiomics module. After dimensionality reduction and correlation analysis, a total of 67 radiomics features were retained. The RF model finally screened out five important imaging histological features: first-order statistics: log-sigma-2-0-mm-3D-firstorder-10Percentile; log-sigma-2-0-mm-3D-firstorder-Maximum; texture features: log-sigma-2-0-mm-3D-glszm-ZoneEntropy; and first-order statistics. First-order-Maximum; texture features: log-sigma-2-0-mm-3D-glszm-ZoneEntropy; wavelet-LLL-glrlm-GrayLevelVariance; wavelet-HLH-glcm-Imc2 (Figure 4).

Figure 4.

Random forest screening for representative imaging histology features. *, P<0.05; **, P<0.01. MCC, maximal correlation coefficient; MSE, mean squared error.

Comparison of diagnostic efficacy of models

Based on the results of multivariate logistic regression analysis, the clinical model (D-dimer) and an MPA/AA model were established, and the imaging histology model was constructed by finally screening 5 imaging histology features using the RF method; meanwhile, this study further integrates the clinical model, the MPA/AA model, and the imaging histology model to establish a joint model.

The results of the ROC curves showed that the AUC values of the joint model was 0.884 in the training set and 0.874 in the validation set (Figure 5, Table 3). Within the training set, the DeLong test further validated statistically significant differences in AUC values between the clinical model and the combined model (P=0.001), the MPA/AA model and the combined model (P=0.03), and the radiomics model and the combined model (P=0.004). In the validation set, significant differences in AUC were observed between the clinical model and the combined model (P=0.03) and between the radiomics model and the combined model (P=0.049). These results indicate that the combined model demonstrates superior diagnostic performance compared to individual models.

Figure 5.

ROC curves for each model in training set (A) and validation set (B). AA, ascending aortic; AUC, area under the curve; CI, confidence interval; MPA, main pulmonary artery; ROC, receiver operating characteristic.

Table 3. Comparison of ROC curves for diagnostic performance of each model in training and validation sets.

Model	AUC (95% CI)	Sensitivity	Specificity	Truncation value
Clinical model
Training set	0.724 (0.629–0.818)	0.732	0.683	0.545
Validation set	0.719 (0.571–0.866)	0.710	0.667	0.500
MPA/AA model
Training set	0.819 (0.740–0.899)	0.714	0.921	0.997
Validation set	0.811 (0.691–0.931)	0.710	0.905	0.991
Radiomics model
Training set	0.774 (0.690–0.857)	0.821	0.619	0.410
Validation set	0.770 (0.634–0.905)	0.742	0.762	0.562
Combined model
Training set	0.884 (0.825–0.944)	0.875	0.762	0.381
Validation set	0.874 (0.775–0.973)	0.935	0.714	0.454

AA, ascending aortic; AUC, area under the curve; CI, confidence interval; MPA, main pulmonary artery; ROC, receiver operating characteristic.

DCA showed that within a threshold probability ranged from 7% to 94%, and the joint model provided the highest net clinical benefit and the best diagnostic performance for COPD patients with PH, confirming its highest clinical value (Figure 6). The Hosmer-Lemeshow test confirmed good calibration of the joint model in both the training set risk probability threshold (P=0.92) and the validation set risk probability threshold (P=0.65). The calibration curves show that a high degree of agreement between predicted probabilities and actual observations in both datasets (Figure 7), indicating excellent consistency and accuracy in diagnosing PH in COPD patients. A nomogram was also constructed to visually represent the joint model, integrating all key indicators into a practical diagnostic tool for predicting PH in COPD patients (Figure 8).

Figure 6.

Decision curves for training set (A) and validation set (B). AA, ascending aortic; MPA, main pulmonary artery.

Figure 7.

Training set (A) and validation set (B) calibration curves.

Figure 8.

Nomogram of training set. AA, ascending aortic; MPA, main pulmonary artery.

Discussion

CT imaging has been widely used in the diagnosis and treatment of COPD patients. In this study, we fully leveraged chest CT imaging data from COPD patients. Beyond the conventional measurement of the MPA/AA ratio, we extracted the Radiomics features from the entire lung CT images and constructed a diagnostic assessment model for COPD combined with PH. This model integrates radiomics features, D-dimer levels, and MPA/AA parameters to noninvasively identify patients with COPD complicated by PH, serving as an effective tool for clinical diagnostic support.

Currently, non-invasive evaluation of PH remains a research focus and challenges both domestically and internationally. Previous studies have confirmed that the MPA/AA diameter ratio based on CT images is an important imaging index for assessing PH. Gašparović et al. (17) have confirmed that a CT-measured PA/AA ratio ≥0.95 serves as an independent predictor of PH in COPD patients, with a specificity reaching 100%. However, this measurement is influenced by factors such as body size, age, CT slice thickness, and measurement method, which may lead to inaccurate results (18). A meta-analysis by Shen et al. (19) showed that MPA/AA demonstrated moderate discriminatory power in detecting PH diagnosis (AUC =0.84), but relatively low sensitivity at 0.74 [95% confidence interval (CI): 0.66–0.80], suggesting a lower false positive rate but a considerable risk of false negatives. The study also pointed out that CT-derived pulmonary artery diameter alone has limited reliability in diagnosing PH, and its diagnostic accuracy may improve when combined with other modalities such as echocardiography. In this study, MPA/AA was identified as an independent risk factor for COPD combined with PH. The constructed MPA/AA model achieved AUCs of 0.819 and 0.811 in the training and validation sets, respectively, and the values of the Hosmer-Lemeshow test were P=0.14 and P=0.49, indicating good diagnostic performance and calibration. These results suggest that MPA/AA can be used as an assessment index for COPD combined with PH.

In recent years, Radiomics has been widely used in the diagnosis, treatment, and prognosis of malignant tumors (7,20). Meanwhile, studies have shown that Radiomics also offers distinct advantages in non-neoplastic diseases (10,11). Radiomics is capable of sensitively detecting subtle structural changes in the lungs, which, though seemingly minor, are closely related to lung structure and functional impairments. Makimoto et al. (21) compared different feature selection methods and machine learning classifiers in predicting COPD status based on CT texture features, and concluded that a model using a Linear-Support Vector Machines (Linear-SVM) classifier obtained the highest AUC value of 0.78. These radiomic features not only reflect microscopic changes in lung structure, but may also indicate pulmonary vascular abnormalities. A multicenter study (22) combining nomograms built from Radiomics scores with independent clinical risk factors to evaluate cardiovascular disease risk in patients with COPD. The results showed that the Radiomics-based nomograms outperformed clinical models in identifying cardiovascular risk among in patients with COPD, demonstrating strong performance in the training, internal validation and external validation sets (AUC =0.731, 0.727, and 0.725, respectively). These results lay a solid foundation for the present study, which aims to assess the risk of comorbid PH in patients with COPD using CT-based Radiomics, and further validate the significant diagnostic and evaluative value of Radiomics in both respiratory and cardiovascular diseases, as well as its potential clinical feasibility. In this study, five important radiomic features were screened by the RF method, and the AUCs of the constructed training set and validation set imaging histology models were 0.774 (95% CI: 0.690–0.857) and 0.770 (95% CI: 0.634–0.905), respectively. Meanwhile, a nomogram incorporating multiple independent risk factors for COPD patients with comorbid PH was developed. This tool provides a more intuitive and convenient means of evaluating PH risk in patients with COPD, aiding clinicians in conducting individualized risk assessments and enabling more precise treatment planning, thereby improving patient outcomes.

In Radiomics studies, first-order and texture features capture distinct biological properties of a lesion from different dimensions. First-order features mainly reflect the gray intensity distribution within the lesion, while texture features characterize spatial relationship between pixels. In this study, the significant first-order features included log-sigma-2-0-mm-3D-firstorder-10Percentile and log-sigma-2-0-mm-3D-firstorder-Maximum, both of which capture density differences in lesion areas by quantifying the intensity distribution of CT voxels. The selected texture features comprised log-sigma-2-0-mm-3D-glszm-ZoneEntropy [a gray-level size zone matrix (GLSZM) feature], wavelet-LLL-glrlm-GrayLevelVariance [a gray-level run-length matrix (GLRLM) feature], and wavelet-HLH-glcm-Imc2 [a gray-level co-occurrence matrix (GLCM) feature] (Information Measure Correlation 2). Among them, the GLCM-derived feature showed the highest clinical value in distinguishing COPD phenotypes. Recent studies have revealed that whole-lung radiomic signatures are significantly correlated with airflow limitation severity and patient-reported health status in COPD, with some features independently associated with the risk of disease progression. Furthermore, a Radiomics model based on native T1 mapping has been shown to predict short-term therapeutic response in PH patients. These findings underscore the potential of both first-order and texture features as multidimensional imaging markers for the precision diagnosis and treatment of respiratory and cardiovascular comorbidities (11,23-25).

COPD is a chronic inflammatory disease characterized by a prothrombotic state driven by persistent systemic inflammation, which can induce coagulation abnormalities through mechanisms such as endothelial damage and platelet activation (26). D-dimer, a fibrin degradation product, serves as a widely used clinical biomarker for coagulation system activation (27). Smits et al. (28) demonstrated significantly elevated D-dimer levels in the blood of PH patients (P=0.001), and fibrinogen showed an upward trend without statistical significance (P=0.09), supporting the hypothesis that hypercoagulability contributes to the pathophysiology of PH. Consistent with these findings, the present study identified D-dimer as an independent risk factor for COPD-combined PH (P=0.008), which is consistent with the findings of Smits et al. (28). However, despite the clinical value of D-dimer in the early identification and prognostic assessment of COPD combined with PH, its limited specificity—due to elevated levels in various conditions such as infections, tumors, and trauma (29)—restricts its standalone diagnostic value. Therefore, dynamic monitoring of D-dimer, in combination with radiologic features and other coagulation indicators, is essential to support individualized clinical decision-making in COPD management.

There are three limitations in this study. First, the relatively small size may have limited the statistical validity and generalizability of the findings. Future studies with larger cohorts are warranted to enhance the robustness of the conclusions. Second, as a single-center retrospective study, the data were collected from a single institution, lacking external validation sets. This may introduce selection bias and limit the external validity of the findings. Third, although echocardiography has achieved near-gold-standard accuracy in diagnosing concomitant PH compared to RHC, relying solely on ultrasound for PH classification may still introduce classification bias. Future research plans to integrate multiple echocardiographic parameters for comprehensive assessment, aiming to minimize such bias. Although the combined model demonstrated good discriminatory ability in validation (AUC ≈0.87), indicating its potential value in early detection and risk stratification, its robustness and clinical feasibility require further validation through multicenter studies. This is necessary due to limitations in single-center data sources and the absence of independent external validation, aiming to provide more reliable support in clinical practice.

Conclusions

In conclusion, the combined model based on CT imaging, D-dimer, and MRA/AA demonstrated strong diagnostic performance for identifying COPD combined with PH. This approach offers a reliable and clinically applicable tool for early detection and risk stratification of PH in COPD patients. This method provides a reliable and clinically valuable tool for the early detection and risk stratification of PH in COPD patients. It enables noninvasive identification of COPD patients with concomitant PH, thereby facilitating early intervention to improve patient outcomes.

Supplementary

The article’s supplementary files as

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 Institutional Review Board of the First Affiliated Hospital of Hebei North University (No. k2024257; Zhangjiakou, China). In view of the retrospective nature of the present study, the local institutional review board waived the requirement for informed consent.

Data Sharing Statement

Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1998/dss

jtd-18-01-18-dss.pdf ^{(82.3KB, pdf)}

DOI: 10.21037/jtd-2025-1998