Combining radiomics based on high-resolution computed tomography with plasma metabolomics for diagnosing subtypes of rheumatoid arthritis-associated interstitial lung disease

Combining radiomics based on high-resolution computed tomography with plasma metabolomics for diagnosing subtypes of rheumatoid arthritis-associated interstitial lung disease | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Combining radiomics based on high-resolution computed tomography with plasma metabolomics for diagnosing subtypes of rheumatoid arthritis-associated interstitial lung disease Hongya Liu, Jianghua Chen, Jie Zhu, Dandan Wu, Kui Du, Fanxin Zeng, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7839620/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Apr, 2026 Read the published version in European Journal of Medical Research → Version 1 posted 13 You are reading this latest preprint version Abstract Objectives Rheumatoid arthritis-associated interstitial lung disease (RA-ILD) is primarily classified into usual interstitial pneumonia (UIP) and non-specific interstitial pneumonia (NSIP) patterns, which exhibit significant differences in prognosis and treatment response. This study aimed to differentiate UIP and NSIP patterns in RA-ILD by integrating HRCT-based radiomics with plasma metabolomics. Methods We included 350 RA-ILD patients, assigned to training, internal validation, and external validation sets. Radiomics features were extracted from HRCT images, and optimal features were selected using variance threshold, univariate selection, and LASSO regression. Three machine learning algorithms—random forest (RF), logistic regression (LR), and multi-layer perceptron (MLP)—were each used to construct three categories of models: clinical models, radiomics models, and combined clinical-radiomics models. This design resulted in a total of 9 models (3 algorithms × 3 model categories). A nomogram integrating the radiomics score (Rad-score) and clinical variables was developed. Model performance was assessed via AUC, calibration curves, and decision curve analysis (DCA). Plasma metabolomics analysis was performed using liquid chromatography-tandem mass spectrometry, and Spearman correlation was used to explore the relationship between radiomics features and metabolites. Results Twenty-five optimal radiomics features were selected, and 77 metabolites were significantly correlated with radiomics features (R > 0.4, P < 0.05). L-glutamate and alpha-ketoglutarate were co-enriched in key metabolic pathways (e.g., arginine biosynthesis, D-glutamine/D-glutamate metabolism). Radiomics and combined models outperformed clinical models: the MLP combined model showed more favourable discrimination in the training set (AUC = 0.915; 95% CI: 0.884–0.940) and internal validation set (AUC = 0.828; 95% CI: 0.728–0.916), while the RF combined model showed the best discrimination in the external validation set (AUC = 0.841; 95% CI: 0.729–0.916, specificity = 67%). The nomogram exhibited excellent calibration (Hosmer–Lemeshow: P = 0.051, 0.382, 0.073 for training, internal, external sets) and clinical utility (DCA: high net benefits across thresholds). Conclusions The combined clinical-radiomics model provides a promising auxiliary tool for distinguishing subtypes of RA-ILD, which may support early and accurate identification of RA-ILD subtypes to aid clinical decision-making. The Rad-score improves RA-ILD subtype discrimination, and L-glutamate/alpha-ketoglutarate merit further investigation as potential metabolic markers. radiomics nomogram metabolomics rheumatoid arthritis-associated interstitial lung disease usual interstitial pneumonia non-specific interstitial pneumonia Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction It is estimated that approximately 10% of individuals with rheumatoid arthritis (RA) will develop clinically significant interstitial lung disease (ILD) 1 – 3 . Patients with RA-associated ILD (RA-ILD) exhibit a higher mortality rate compared with RA patients without ILD 4 . RA-ILD can be classified as either a usual interstitial pneumonia (UIP) pattern or a non-UIP pattern (mainly non-specific interstitial pneumonia (NSIP)), with UIP as the pathologic manifestation in 54%–65% of patients with RA-ILD 5 – 7 . A UIP pattern on high-resolution computed tomography (HRCT) is associated with a worse prognosis and a higher risk of acute exacerbation compared to NSIP pattern in patients with RA-ILD 8 – 10 . Therefore, accurate differentiation between UIP and NSIP patterns is critical for guiding treatment decisions and improving patient outcomes. There is substantial evidence to support the use of HRCT instead of lung biopsy to distinguish ILD subtypes 11 – 12 . However, due to the large overlap of CT features between NSIP and UIP patterns, subjective differences in the identification and interpretation of abnormalities detected on imaging 13 , the overall diagnostic accuracy of CT in distinguishing between UIP and NSIP is about 70% 14 . As a result, increasing research is focusing on the development of computer-aided tools and the application of artificial intelligence for rapid assessment. Radiomics is a bridge between medical imaging and personalized medicine. By applying high-throughput quantitative image features extracted from medical images to clinical decision support systems, radiomics can improve the accuracy of diagnosis, prognosis and prediction 15 .For example, the machine learning model developed by Koo et al. can successfully differentiate pathologically confirmed UIP, NSIP, and chronic hypersensitivity pneumonitis 16 . Sun et al. 17 developed a lung graph-based machine learning model for identifying fibrous ILDs, which outperforms the diagnostic performance of radiologists and may help clinicians objectively assess ILDs. These results suggest that radiomics are a potential tool for the diagnosis of RA-ILD. However, radiomics faces a significant challenge: the lack of biological interpretability of radiomic features, which limits its clinical translation 18 . Metabolomics, which quantifies dynamic changes in metabolites, offers a complementary approach to uncover the biological mechanisms underlying disease processes 19 .Rindlisbacher et al. 20 identified the unique metabolic characteristics of patients with idiopathic pulmonary fibrosis through serum metabolic profiling, and lysophosphatidylcholine may be a potential marker for diagnosis and monitoring of idiopathic pulmonary fibrosis. Our previous multi-omics studies revealed the central role of glycerophospholipid metabolism in the development of rheumatoid arthritis 21 . Integrating radiomics with metabolomics provides a non-invasive and informative strategy to enhance diagnostic accuracy and reveal the biological significance of imaging features. Despite these advances, no studies have comprehensively explored the combined use of radiomics and metabolomics to differentiate UIP and NSIP patterns in RA-ILD. Therefore, this study aims to develop and validate combined clinical-radiomics models for RA-ILD diagnosis and to investigate the biological relevance of radiomic features through correlation analysis with plasma metabolites. Methods Our Institutional Review Board (The Army Medical University Medical Ethics Committee and Dazhou Central Hospital Medical Ethics Committee) approved this study and waived the requirement to obtain informed consent because this was a retrospective analysis. Study population The study populations comprised RA-ILD patients who were consecutively recruited from January 2019 to January 2024 at the First Affiliated Hospital of Army Medical University, and from January 2017 to December 2023 at Dazhou Central Hospital. All patients met the 2010 RA classification criteria of the American College of Rheumatology/European League Against Rheumatism (ACR/EULAR) 22 . The presence of ILD was confirmed by chest HRCT in all cases, evaluated by two senior radiologists using a single-blind method (who were unaware of the clinical information and imaging reports of the patients). The type of ILD was either UIP pattern or NSIP pattern, which was evaluated by a multidisciplinary team (MDT) (including pulmonologists, radiologists, pathologists and rheumatologists) based on the criteria of the American Thoracic Society/European Respiratory Society (ATS/ERS) International Multidisciplinary Consensus Classification of the Idiopathic Interstitial Pneumonias 23 . The flowchart of patient enrollment is shown in (Fig. 1 ). The research workflow of radiomics analysis is shown in (Fig. S1 ). HRCT image acquisition HRCT images of all enrolled patients were collected according to a standard chest imaging protocol. The images contained 116–560 layers, with layer thicknesses of 0.63–2.00 mm. Each layer was reconstructed to a matrix size of 512 × 512 with an in-plane pixel spatial resolution ranging from 0.56 mm × 0.56 mm to 0.92 mm × 0.92 mm. A lung window (L/W: -500 HU/1500 HU) was used to interpolate CT numbers (Hounsfield units) to 1 mm × 1 mm × 1 mm per scan, reducing inter-scan variations (e.g., slice thickness). Demographic, Laboratory Data and Clinical signature The following patient characteristics were included: clinical characteristics (age, sex), laboratory data such as rheumatoid factors (RF), erythrocyte sedimentation rate (ESR), c-reactive protein (CRP), white blood cell count (WBC), monocyte, lymphocyte, basophil, eosinophil, and neutrophil counts and percentages, glutamyl transpeptidase (GGT), alanine aminotransferase (ALT), creatinine (Cr), and other relevant markers. A univariate analysis of all clinical data was performed to screen for significant factors related to outcome events. Metabolomic Data In the external validation set, plasma samples from patients underwent liquid chromatography-tandem mass spectrometry analysis using an AB SCIEX UPLC-TripleTOF system. We acquired raw metabolic data and initially pre-processed it using Progenesis QI software (Waters Corporation, Milford, USA). This involved eliminating missing values exceeding 80% within each group, imputing missing values with the minimum observed, using sum normalization, discarding metabolites with relative standard deviations over 30% in quality control samples, and applying log10 transformations. Subsequently, we matched the mass spectral data of the metabolites with the Human Metabolome Database (HMDB, http://www.hmdb.ca/ ) to acquire the final metabolite expression profiles for further analysis. Spearman correlation analysis was used to identify plasma metabolites associated with radiomics features (R > 0.4, P < 0.05). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were carried out on the online platform of Majorbio ( https://cloud.majorbio.com ). A detailed description of the metabolomics methods can be found in Supplementary Material. Radiomics Feature extraction, selection and Rad-score calculation Feature extraction was performed using the AI Platform of the First Affiliated Hospital of Army Medical University. The platform’s U-net model automatically segmented lungs and extracted volumes of interest (VOI) (Fig. S2). We used three feature selection methods, variance threshold (> 0.8), univariate selection ( P < 0.05), and the least absolute shrinkage and selection operator (LASSO) to select the features. First, the variance threshold initially removed 263 features that contributed little to the model, then the univariate selection excluded 505 features that were not sufficiently significantly ( P > 0.05) associated with the RA-ILD subtype, and finally LASSO was conducted to choose 25 optimized subsets of features to construct the model (Fig. 2 ). Radiomics score (Rad-score) was calculated by summing the selected features weighted by their coefficients. Model construction and Nomogram Logistic regression (LR), random forest (RF), and multi-layer perceptron (MLP) were employed to build the radiomics model, clinical model, and combined radiomics-clinical model, respectively. Furthermore, a clinical-radiomics nomogram was constructed via multivariable logistic regression to visually present the probability of UIP pattern, incorporating Rad-score and independent clinical predictors. Statistical Analysis Statistical analyses were performed in R (version 4.3.0) and Python (version 3.7.0). Univariate analysis was used to compare the differences in clinical factors between UIP and NSIP patients. Statistical tests included corrected chi-square tests (categorical variables) and one-way analysis of variance (ANOVA) (normally distributed continuous variables); descriptive statistics were presented as frequencies (percentages) or mean ± standard deviation (SD), respectively. The Kruskal-Wallis H test was used if continuous variables were not normally distributed, and descriptive statistics were median (upper and lower quartiles).The receiver operating characteristic (ROC) curve was drawn and AUC was calculated to evaluate the diagnostic performance of the models. As well as analyzing the accuracy, sensitivity, and specificity of the models. Goodness-of-fit of the nomogram was evaluated using the Hosmer-Lemeshow test, while its predictive accuracy was assessed using calibration curves. Decision curve analysis (DCA) was performed to evaluate the clinical value of the models independently.. Calibration curves, DCA, and univariate analysis were implemented in R with the "rms" and "stats" packages. The Spearman correlation coefficient was calculated and visualized using the 'ComplexHeatmap' package in R. P < 0.05 was considered statistically significant. Results Baseline demographic and clinical features 301 patients (166 with UIP, 135 with NSIP) from the First Affiliated Hospital of Army Medical University were randomly divided into a training set (n = 240) and an internal validation set (n = 61) using computer-generated random numbers. 49 patients (27 in UIP and 22 in NSIP) from Dazhou Central Hospital were used as the external validation set. No significant clinical factor differences were found between the training set and the internal validation set ( P > 0.05) (Supplementary Table S1 ). Radiomics feature construction In this study, 1688 radiomics features were extracted from each VOI, including 324 first-order, 14 shape, 432 gray-level co-occurrence matrix, 252 gray-level dependence matrix, 288 gray-level run-length matrix, 90 neighboring gray-tone difference matrices, and 288 gray-level size zone matrix features. 25 features were ultimately selected and contributed to the Rad-score construction. Nomogram variable screening In univariate analysis, the variables sex (odds ratio [OR] = 2.658, 95% confidence interval [CI]: 1.673–4.377, P < 0.001) and WBC count (OR = 0.876, 95% CI: 0.799–0.955, P = 0.004), monocyte count (OR = 0.326, 95% CI: [0.111–0.926], P = 0.038), and neutrophil count (OR = 0.876, 95% CI: [0.792–0.964], P = 0.008) were remained as independent predictors in the clinical factors model (Supplementary Table S2). Model development and evaluation Three categories of models were constructed using RF, LR, and MLP—with each algorithm applied to all model types, resulting in 9 total models. Specifically: Clinical models were built based on sex, WBC, monocyte, and neutrophil counts (Supplementary Fig. S3); Radiomics models were developed using the Rad-score (Supplementary Fig. S4); Combined clinical-radiomics models integrated clinical variables and the Rad-score. The combined models showed the best discriminative performance, with significantly higher AUC values than the clinical and radiomics models (Table 1). The MLP combined model showed more favourable discrimination in both the training set (AUC = 0.915; 95% CI: 0.844, 0.940) and internal validation set (AUC = 0.828; 95% CI: 0.728, 0.916).( Supplementary Fig. S5A). The RF combined model did the best discrimination in the external validation set (AUC = 0.841; 95% CI: 0.729, 0.916) (Supplementary Fig. S5C), but its specificity was low at 67%. Table 1: Performance of all 9 models (3 algorithms × 3 model categories) Data set MODEL AUC (95% CI) Accuracy Sensitivity Specificity Training set MLP Clinical model 0.683(0.625～0.738) 0.650 0.648 0.652 MLP Radiomics model 0.831(0.784～0.869) 0.750 0.704 0.788 MLP combined model 0.915(0.884～0.940) 0.821 0.917 0.742 LR Clinical model 0.645(0.589～0.706) 0.629 0.667 0.598 LR Radiomics model 0.820(0.774～0.862) 0.733 0.787 0.689 LR combined model 0.829(0.785～0.872) 0.750 0.750 0.750 RF Clinical model 0.772(0.729～0.825) 0.704 0.796 0.629 RF Radiomics model 0.839(0.797～0.876) 0.771 0.833 0.720 RF combined model 0.893(0.860～0.927) 0.833 0.833 0.833 Internal validation set MLP Clinical model 0.629(0.492～0.756) 0.672 0.519 0.794 MLP Radiomics model 0.801(0.710～0.889) 0.754 0.852 0.676 MLP combined model 0.828(0.728～0.916) 0.787 0.815 0.765 LR Clinical model 0.641(0.523～0.760) 0.641 0.672 0.704 LR Radiomics model 0.803(0.688～0.706) 0.803 0.787 0.667 LR combined model 0.820(0.714～0.902) 0.820 0.803 0.778 RF Clinical model 0.536(0.390～0.643) 0.536 0.607 0.593 RF Radiomics model 0.815(0.713～0.893) 0.815 0.754 0.852 RF combined model 0.815(0.712～0.892) 0.815 0.803 0.815 External Validation Set MLP Clinical model 0.589(0.448～0.724) 0.612 0.818 0.444 MLP Radiomics model 0.762(0.623～0.861) 0.776 0.773 0.778 MLP combined model 0.731(0.583～0.838) 0.714 0.864 0.593 LR Clinical model 0.579(0.435～0.715) 0.818 0.444 0.435 LR Radiomics model 0.778(0.643～0.878) 0.796 0.818 0.778 LR combined model 0.763(0.625～0.863) 0.755 0.818 0.704 RF Clinical model 0.554(0.409～0.713) 0.673 0.318 0.963 RF Radiomics model 0.826(0.692～0.907) 0.776 0.636 0.889 RF combined model 0.841(0.729～0.916) 0.755 0.864 0.667 Abbreviations: MLP, multi-layer perceptron; LR, logistic regression; RF, random forest. Nomogram construction and prediction performance assessment The sex, WBC, monocyte and neutrophil counts and Rad-score were incorporated into the clinical-radiomics nomogram (Fig. 3). To use the nomogram, first, we needed to determine the position of each variable (including sex, WBC, monocyte count, neutrophil count, and Rad-score) on the corresponding axis. Next, we drew a vertical line upwards from these positions to the point axis to obtain the corresponding score for each variable. We summed these scores to arrive at a total score. Lastly, we drew a vertical line downwards from the total point axis, and the value corresponding to this line represented the probability of the UIP pattern. Calibration curves (Fig. 4A-C) showed good consistency and the Hosmer–Lemeshow tes yielded non-significant results for both the training set ( P = .051), the internal validation set ( P = 0.382) and the external validation set ( P = 0.073). The DCA (Fig. 4D-F) showed that the nomogram exhibits high net benefits over most threshold probability ranges.This indicated that clinical intervention based on the nomogram could benefit patients in the training set, and also in the internal and external validation sets. Potential metabolic features of radiomics model Next, we analyzed the plasma metabolites of RA-ILD and identified a total of 330 different plasma metabolites in the HMDB database. The workflow of correlation analysis between radiomics features and metabolomics features is shown in Fig. 5A. Spearman correlation analysis of the 25 optimal radiomics features with 330 metabolites showed that 77 metabolites and radiomics features were significantly correlated (R > 0.4, P < 0.05). The 77 correlated metabolites were mainly enriched in pathways involving D-glutamine and D-glutamate metabolism, arginine biosynthesis, alanine, aspartate and glutamate metabolism, glycerophospholipid metabolism, and caffeine metabolism (corrected P < 0.05, Fig. 5B). There were 11 metabolites enriched in these pathways, with L-glutamate and alpha-ketoglutarate co-enriched in the arginine biosynthesis, D-glutamine and D-glutamate metabolism, alanine, aspartate, and glutamate metabolism pathways comprising the regulatory network (Fig. 5C). Discussion Despite significant advances in the diagnosis and treatment of RA, early diagnosis of RA-ILD remains challenging. A Danish population-based cohort study showed a 5-year mortality rate of 39.0% in patients with RA-ILD compared with 18.2% in patients without ILD 24 . In patients with RA-ILD, a UIP pattern demonstrated by HRCT scan or histology has a higher mortality rate compared to other patterns of pulmonary fibrosis 25-27 . Therefore, accurate differentiation between UIP and NSIP patterns is essential to guide treatment decisions and improve patient outcomes 28 . In this study, we developed and validated combined clinical-radiomics models to distinguish UIP and NSIP patterns, demonstrating superior diagnostic performance compared to traditional clinical models. The introduction of CT radiomics has provided a new perspective on the diagnosis of RA-ILD. While traditional radiological methods focus on macro-level analysis, radiomics are capable of uncovering hidden information in the image data, such as heterogeneity in lung texture and microscopic structura changes. In this study, the Rad-score was used to construct radiomics models, which exhibited good diagnostic performance on the training set, internal validation set and external validation set. The RF radiomics model demonstrated favorable discrimination, with AUC values of 0.839 in the training set, 0.815 in the internal validation set, and 0.826 in the external validation set. Compared with the clinical models, the diagnostic performance of the radiomics models was significantly improved, which reflected the increased value of Rad-Score in classification diagnosis. At the same time, comprehensive evaluation of the performance of all models showed that the performance of radiomics models in the three datasets was relatively stable, which confirmed the stability and consistency of radiomics features in the identification of RA-ILD subtypes. Interestingly, the study by Qiu et al. 29 also found that Rad-score has potential ability to differentiate between immune checkpoint inhibitor-related pneumonitis and radiation pneumonitis. Unlike radiomics studies that were based on cases diagnosed by pathology 16 30 , our study was based on clinical real-world cases, which somewhat avoids selection bias and improves the clinical applicability of radiomics studies. In addition, most of the combined models constructed by the same algorithm further improved their performance compared with their corresponding radiomics and clinical models, showing higher AUC, ACC, sensitivity and specificity. This reflects the multiple advantages of the combined models, which not only makes full use of existing medical resources, but also can realistically simulate the diagnostic thinking mode of clinicians. To transform the combined model into an intuitive visual computational tool, we drew a nomogram of the combined model in this study. The calibration curve showed that the UIP pattern prediction probability matched the actual observation. DCA showed that the combined model had better net benefits. This indicates that clinicians can use the nomogram to intuitively and easily distinguish RA-UIP from RA-NSIP. Recently, some studies have used deep learning methods 31-34 , for example, Chung et al. 34 developed a deep learning-based radiomics classifier for the diagnosis of UIP from chest CT scans in a retrospective study, which included a total of 2907 chest CT scans, and the classifier demonstrated a high level of sensitivity and specificity in different cohorts. Univariate analysis indicated that sex, WBC, monocyte and neutrophil counts were independent predictors for diagnosing RA-ILD patterns. Although risk factors for the development of RA-ILD include older age, male sex, smoking history, poor control of arthritis activity, and high levels of RF and anti-citrullinated protein antibodies 7 . However, no study has reported significant differences in RF between UIP and NSIP, and our study is also the case. Male sex was associated with the presence of UIP patterns, which is consistent with previous studies 35 , but no study has mentioned that WBC, monocyte and neutrophil count were significantly different between UIP and NSIP patterns. The previous study showed that monocytes, platelets, neutrophils, and lymphocytes play a crucial role in the immune response in RA and that platelet/lymphocyte ratio can be used as a biomarker to differentiate RA-ILD from RA 36 . This suggests that inflammation may play a role in the pathogenesis and progression of pulmonary fibrosis. Recent studies have reported that older age, seropositivity, and male sex were strongly associated with RA-UIP, whereas RA-related autoantibodies were associated with RA-NSIP 37 . The study by Wang et al. 38 showed that the AUC of serum UA for predicting RA-UIP was 0.845 (95%CI: 0.78-0.91; P < 0.01), indicating that the elevated serum UA concentration may be a diagnostic marker for RA-UIP. However, in this study, no significant difference was found in older age, seropositivity, and serum UA between UIP and NSIP. One possible reason is that the sample size of this study is small,and a large prospective cohort study is needed to confirm. Recently, there has been a surge of research focusing on the genomic, cytokine, antibody, proteomic and metabolomic profiles of RA-ILD, aimed to enhance the development of biomarkers for this condition 39-40 . Proteomic analyses have identified molecular signatures that are strongly associated with the presence and severity of RA-ILD, providing insights into unexplored disease pathways 40 . Notably, researchers found that serum levels of decanoic acid, glycerol, and morpholine were different in RA patients with or without ILD 41 . The partial least squares-discriminant analysis model generated from these three metabolites could successfully discriminate ILD in RA (AUC: 0.919, 95% CI: 0.867-0.968, sensitivity 0.880, specificity 0.780). In our study, 330 plasma metabolites of RA-ILD were correlated with 25 radiomics features, and 77 correlated metabolites were identified. In the pathway enrichment analysis, D-glutamine and D-glutamate metabolism, arginine biosynthesis, alanine, aspartate and glutamate metabolism, glycerophospholipid metabolism, and caffeine metabolism pathways were significantly enriched. These results suggest that these five metabolic pathways may be closely related to the pathophysiological process of RA-ILD. Among them, arginine biosynthesis, alanine, aspartate and glutamate metabolism are associated with disorders of amino acid metabolism in different stages of RA 42 . L-glutamate and alpha-ketoglutarate are enriched in a variety of metabolic pathways, and their elevation may be related to the increased activity of RA-ILD, suggesting that they may be have potential for further research, but this needs to be further verified by a larger sample size. According to recent investigations, glutamate contributes to pulmonary fibrosis 43-44 and citrulline is also higherexpression in RA-ILD lung tissues 45 . By deeply exploring the correlation between radiomics features and metabolomics, we are expected to reveal the potential biological significance behind these radiomics features. This study has several limitations that should be considered when interpreting the findings. First, the retrospective design inherently introduces potential selection bias, as patient enrollment relied on existing medical records at two single-center institutions. While we ensured no significant differences in baseline characteristics between the training and internal validation sets, the generalizability of our results to broader populations—especially those from diverse geographic regions or healthcare settings—may be limited. Prospective, multicenter studies with standardized data collection protocols are therefore needed to validate the performance of the combined clinical-radiomics model. Second, RA-ILD subtype classification was based on HRCT findings and MDT assessment (including pulmonologists, radiologists, pathologists, and rheumatologists) rather than surgical lung biopsy—the pathological gold standard for interstitial lung disease (ILD) diagnosis. Although MDT consensus aligns with clinical practice guidelines for ILD evaluation 23 and reduces interobserver variability, the absence of histopathological confirmation may introduce residual diagnostic uncertainty, particularly for cases with overlapping radiological features between UIP and NSIP patterns. Third, radiomics features were extracted from the entire lung parenchyma rather than focal ILD lesions. While this approach captures global lung involvement, it may dilute the signal from lesion-specific textural or density characteristics that are most relevant to subtype differentiation. Future studies could optimize feature extraction by focusing on manually or automatically segmented lesion regions to enhance the specificity of radiomic signatures. Fourth, incomplete outpatient medical records led to missing data on key variables with reported associations with RA-ILD pathogenesis, including smoking history and anti-citrullinated protein antibody (ACPA) status 46-47 . The exclusion of these variables from model construction may have underestimated their potential predictive value and limited the comprehensiveness of the clinical-radiomics integration. Fifth, our cohort only included patients with UIP or NSIP patterns, which are the most common RA-ILD subtypes but do not represent the full spectrum of RA-associated ILDs (e.g., cryptogenic organizing pneumonia, lymphoid interstitial pneumonia). The performance of our model in distinguishing these less frequent subtypes remains unknown and requires further investigation. Finally, while we identified correlations between radiomic features and plasma metabolites (e.g., L-glutamate, alpha-ketoglutarate), the biological mechanisms linking these imaging biomarkers to metabolic dysregulation in RA-ILD remain unclear. Radiomics features often lack direct biological interpretability 18 , and additional multi-omics studies (e.g., integrating genomics or proteomics) are needed to uncover the molecular pathways underlying the observed associations, which could improve the clinical translatability of radiomic-based tools. Conclusions In this study, the clinical-radiomics combined model (nomogram) can be used as a tool to distinguish the subtypes of RA-ILD, which is useful for the early and accurate identification of RA-ILD subtypes to support clinical decision-making. The Rad-score has considerable potential in distinguishing RA-ILD subtypes and deserves further exploration. Building on traditional radiomics, this study integrated metabolomics with radiomics to identify metabolites associated with RA-ILD-related radiomics features, highlighting the potential of multi-omics strategies to advance the precision diagnosis and treatment of RA-ILD. Future research should focus on validating these findings in larger cohorts, exploring the biological roles of key metabolites, and further optimizing the combined models for clinical application. Abbreviations RA rheumatoid arthritis ILD interstitial lung disease RA-ILD Rheumatoid arthritis-associated interstitial lung disease UIP usual interstitial pneumonia NSIP non-specific interstitial pneumonia HRCT high-resolution computed tomography RF random forest / rheumatoid factors LR logistic regression MLP multi-layer perceptron Rad-score radiomics score AUC area under the curve DCA decision curve analysis ACR/EULAR American College of Rheumatology/European League Against Rheumatism MDT multidisciplinary team ATS/ERS American Thoracic Society/European Respiratory Society ESR erythrocyte sedimentation rate CRP c-reactive protein WBC white blood cell count GGT glutamyl transpeptidase ALT alanine aminotransferase Cr creatinine KEGG Kyoto Encyclopedia of Genes and Genomes VOI volume of interest LASSO least absolute shrinkage and selection operator ROC receiver operating characteristic OR odds ratio CI confidence interval ACPA anti-citrullinated protein antibody Declarations Ethics approval and consent to participate This study involves human participants and human data/material, and was conducted in compliance with the World Medical Association Declaration of Helsinki (WMA Declaration of Helsinki – Ethical Principles for Medical Research Involving Human Participants). It was approved by the Ethics Committees of the First Affiliated Hospital of Army Medical University (Ethic Reference: (B)KY2023148) and Dazhou Central Hospital (Ethic Reference: 2021-022). Informed consent was waived due to the retrospective design and de-identification of patient data/material. Consent for publication Not applicable. Data availability statement The data supporting the findings of this study are available from the corresponding author upon reasonable request. The high-resolution computed tomography (HRCT) radiomics features, plasma metabolomics profiles, and clinical data used in this study have been de-identified to protect the privacy of participants (in compliance with the Declaration of Helsinki and ethical approval requirements of participating institutions). All data relevant to the study are included in the article and its supplementary materials. Competing interests The authors declare no competing interests. Funding This work was funded by the Individualized Training Grant for Key Talents of the Outstanding Talent Pool of Army Medical University (XA-2019-505-40) and the key projects of the Chongqing Science and Health Joint Traditional Chinese Medicine Research Project(2024ZYZD007). The funding sources did not have any role in the collection, analysis and interpretation of data. The Financial contributors did not influence the study design, collection, analysis and interpretation of data, the writing of the manuscript or the decision to submit the manuscript for publication. Author contributors HL, JC, JZ, DW and DK: patient selection and recruitment; HL, JC, JZ, FZ and QZ: experimental design; HL, JC, JZ, DW and DK: data collection; HL, JC, JZ and QZ: data analysis; HL, JC, JZ, FZ and QZ: manuscript writing and revision; and QZ: funding. Guarantor: QZ. Acknowledgements The authors would like to thank the doctors and nursing team of Department of Rheumatology and Immunology of the First Affiliated Hospital of Army Medical University and the Clinical Medical Research Center of Dazhou Central Hospital. Clinical trial registration http://www.chictr.org.cn (ChiCTR2400084726). Name of registry Chinese Clinical Trial Registry. References Olson AL, Swigris JJ, Sprunger DB, et al. Rheumatoid arthritis-interstitial lung disease-associated mortality. American journal of respiratory and critical care medicine 2011;183(3):372–8. Zou YQ, Li YS, Ding XN, et al. The clinical significance of HRCT in evaluation of patients with rheumatoid arthritis-associated interstitial lung disease: a report from China. Rheumatology international 2012;32(3):669–73. Oh JH, Kim GHJ, Cross G, et al. Automated quantification system predicts survival in rheumatoid arthritis-associated interstitial lung disease. Rheumatology (Oxford) 2022;61(12):4702–10. Manfredi A, Cassone G, Luppi F, et al. Rheumatoid arthritis related interstitial lung disease. Expert review of clinical immunology 2021;17(5):485–97. Zamora-Legoff JA, Krause ML, Crowson CS, et al. Patterns of interstitial lung disease and mortality in rheumatoid arthritis. Rheumatology (Oxford, England) 2017;56(3):344–50. Lee HK, Kim DS, Yoo B, et al. Histopathologic pattern and clinical features of rheumatoid arthritis-associated interstitial lung disease. Chest 2005;127(6):2019–27. Akiyama M, Alshehri W, Kaneko Y. Does a window of opportunity for rheumatoid arthritis-associated interstitial lung disease exist?. Autoimmunity Reviews, 2024, 23(2): 103501. Atzeni F, Gerardi MC, Barilaro G, et al. Interstitial lung disease in systemic autoimmune rheumatic diseases: a comprehensive review. Expert review of clinical immunology 2018;14(1):69–82. Izuka S, Yamashita H, Iba A, et al. Acute exacerbation of rheumatoid arthritis-associated interstitial lung disease: clinical features and prognosis. Rheumatology (Oxford, England) 2021;60(5):2348–2354. Akiyama, Mitsuhiro, and Yuko Kaneko. Pathogenesis, clinical features, and treatment strategy for rheumatoid arthritis-associated interstitial lung disease. Autoimmunity reviews 2022;21(5):103056. Kim EJ, Collard HR, King TE, Jr. Rheumatoid arthritis-associated interstitial lung disease: the relevance of histopathologic and radiographic pattern. Chest 2009;136(5):1397–405. Tanaka N, Kim JS, Newell JD, et al. Rheumatoid arthritis-related lung diseases: CT findings. Radiology 2004;232(1):81–91. Watadani T, Sakai F, Johkoh T, et al. Interobserver variability in the CT assessment of honeycombing in the lungs. Radiology 2013;266(3):936–44. Mink SN, Maycher B. Comparative manifestations and diagnostic accuracy of high-resolution computed tomography in usual interstitial pneumonia and nonspecific interstitial pneumonia. Current opinion in pulmonary medicine 2012;18(5):530–4. Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749–762. Koo CW, Williams JM, Liu G, et al. Quantitative CT and machine learning classification of fibrotic interstitial lung diseases. European radiology 2022;32(12):8152–61. Sun H, Liu M, Liu A, et al. Developing the Lung Graph-Based Machine Learning Model for Identification of Fibrotic Interstitial Lung Diseases. Journal of imaging informatics in medicine 2024;37(1):268–79. Tomaszewski M R, Gillies R J. The biological meaning of radiomic features. Radiology, 2021, 298(3): 505–516. Li R, Li L, Xu Y, et al. Machine learning meets omics: applications and perspectives. Briefings in Bioinformatics, 2022, 23(1): bbab460. Rindlisbacher B, Schmid C, Geiser T, et al. Serum metabolic profiling identified a distinct metabolic signature in patients with idiopathic pulmonary fibrosis - a potential biomarker role for LysoPC. Respiratory research 2018;19(1):7. Jian C, Wei L, Wu T, et al. Comprehensive multi-omics analysis reveals the core role of glycerophospholipid metabolism in rheumatoid arthritis development. Arthritis research & therapy 2023;25(1):246. Aletaha D, Neogi T, Silman AJ, et al. 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis and rheumatism 2010;62(9):2569–81. Travis WD, Costabel U, Hansell DM, et al. An official American Thoracic Society/European Respiratory Society statement: Update of the international multidisciplinary classification of the idiopathic interstitial pneumonias. American journal of respiratory and critical care medicine 2013;188(6):733–48. Hyldgaard C, Hilberg O, Pedersen AB, et al. A population-based cohort study of rheumatoid arthritis-associated interstitial lung disease: comorbidity and mortality. Annals of the rheumatic diseases 2017;76(10):1700–06. Nieto M A, Rodriguez-Nieto M J, Sanchez-Pernaute O, et al. Mortality rate in rheumatoid arthritis-related interstitial lung disease: the role of radiographic patterns. BMC Pulmonary Medicine, 2021, 21: 205. Singh N, Varghese J, England BR, et al. Impact of the pattern of interstitial lung disease on mortality in rheumatoid arthritis: A systematic literature review and meta-analysis. Seminars in arthritis and rheumatism 2019;49(3):358–65. Solomon JJ, Chung JH, Cosgrove GP, et al. Predictors of mortality in rheumatoid arthritis-associated interstitial lung disease. The European respiratory journal 2016;47(2):588–96. Raghu G, Remy-Jardin M, Richeldi L, et al. Idiopathic pulmonary fibrosis (an Update) and progressive pulmonary fibrosis in adults: An official ATS/ERS/JRS/ALAT clinical practice guideline. American Journal of Respiratory and Critical Care Medicine, 2022, 205(9): e18-e47. Qiu Q, Xing L, Wang Y, et al. Development and Validation of a Radiomics Nomogram Using Computed Tomography for Differentiating Immune Checkpoint Inhibitor-Related Pneumonitis From Radiation Pneumonitis for Patients With Non-Small Cell Lung Cancer. Frontiers in immunology 2022;13:870842. De Giacomi F, Raghunath S, Karwoski R, et al. Short-term Automated Quantification of Radiologic Changes in the Characterization of Idiopathic Pulmonary Fibrosis Versus Nonspecific Interstitial Pneumonia and Prediction of Long-term Survival. Journal of thoracic imaging 2018;33(2):124–31. Pawar SP, Talbar SN. Two-Stage Hybrid Approach of Deep Learning Networks for Interstitial Lung Disease Classification. BioMed research international 2022;2022:7340902. Huang S, Lee F, Miao R, et al. A deep convolutional neural network architecture for interstitial lung disease pattern classification. Medical & biological engineering & computing 2020;58(4):725–37. Walsh SLF, Calandriello L, Silva M, et al. Deep learning for classifying fibrotic lung disease on high-resolution computed tomography: a case-cohort study. The Lancet Respiratory medicine 2018;6(11):837–45. Chung JH, Chelala L, Pugashetti JV, et al. A Deep Learning-Based Radiomic Classifier for Usual Interstitial Pneumonia. Chest 2024;165(2):371–80. Bonilla Hernán MG, Gómez-Carrera L, Fernández-Velilla Peña M, et al.Prevalence and clinical characteristics of symptomatic diffuse interstitial lung disease in rheumatoid arthritis in a Spanish population. Revista clinica espanola 2022;222(5):281–287. Chen Q, Chen DY, Xu XZ, et al. Platelet/Lymphocyte, Lymphocyte/Monocyte, and Neutrophil/Lymphocyte Ratios as Biomarkers in Patients with Rheumatoid Arthritis and Rheumatoid Arthritis-Associated Interstitial Lung Disease. Medical science monitor: international medical journal of experimental and clinical research 2019;25:6474–81. McDermott GC, Hayashi K, Juge PA, et al. Impact of Sex, Serostatus, and Smoking on Risk for Rheumatoid Arthritis-Associated Interstitial Lung Disease Subtypes. Arthritis Care & Research, 2024. Wang Z, Wang W, Xiang T, et al. Serum uric acid as a diagnostic biomarker for rheumatoid arthritis–associated interstitial lung disease. Inflammation, 2022, 45(4): 1800–1814. Furukawa H, Oka S, Higuchi T, et al. Biomarkers for interstitial lung disease and acute-onset diffuse interstitial lung disease in rheumatoid arthritis. Therapeutic advances in musculoskeletal disease 2021;13:1759720x211022506. Wu X, Jeong Y, Poli de Frías S, et al. Serum proteomic profiling of rheumatoid arthritis-interstitial lung disease with a comparison to idiopathic pulmonary fibrosis. Thorax 2022;77(10):1041–44. Furukawa H, Oka S, Shimada K, et al. Serum Metabolomic Profiling in Rheumatoid Arthritis Patients With Interstitial Lung Disease: A Case-Control Study. Frontiers in medicine 2020;7:599794. Zhang X, Yin M, Zhang D, et al. Metabolomics reveals disturbed amino acid metabolism during different stages of RA in collagen-induced arthritis mice. Inflammation, 2024. Zhao YD, Yin L, Archer S, et al. Metabolic heterogeneity of idiopathic pulmonary fibrosis: a metabolomic study. BMJ open respiratory research 2017;4(1):e000183. Liu RM, Vayalil PK, Ballinger C, et al. Transforming growth factor β suppresses glutamate-cysteine ligase gene expression and induces oxidative stress in a lung fibrosis model. Free radical biology & medicine 2012;53(3):554–63. England BR, Duryee MJ, Roul P, et al. Malondialdehyde-Acetaldehyde Adducts and Antibody Responses in Rheumatoid Arthritis-Associated Interstitial Lung Disease. Arthritis & rheumatology 2019;71(9):1483–1493. Leavy O C, Kawano-Dourado L, Stewart I D, et al. Rheumatoid arthritis and idiopathic pulmonary fibrosis: a bidirectional Mendelian randomisation study[J]. Thorax, 2024, 79(6): 538–544. Wang D, Zhang J, Lau J, et al. Mechanisms of lung disease development in rheumatoid arthritis. Rheumatology 2019;15(10):581–596. Additional Declarations No competing interests reported. Supplementary Files Supplementarymateria.docx Cite Share Download PDF Status: Published Journal Publication published 18 Apr, 2026 Read the published version in European Journal of Medical Research → Version 1 posted Editorial decision: Revision requested 11 Jan, 2026 Reviews received at journal 27 Dec, 2025 Reviews received at journal 17 Dec, 2025 Reviews received at journal 15 Dec, 2025 Reviewers agreed at journal 09 Dec, 2025 Reviewers agreed at journal 07 Dec, 2025 Reviewers agreed at journal 07 Dec, 2025 Reviewers agreed at journal 07 Dec, 2025 Reviewers agreed at journal 26 Nov, 2025 Reviewers invited by journal 29 Oct, 2025 Editor assigned by journal 24 Oct, 2025 Submission checks completed at journal 23 Oct, 2025 First submitted to journal 21 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7839620","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":541912196,"identity":"6cd3f3f0-2d4c-4365-bc11-3ea4b6246ddc","order_by":0,"name":"Hongya Liu","email":"","orcid":"","institution":"First Affiliated Hospital of Army Medical University","correspondingAuthor":false,"prefix":"","firstName":"Hongya","middleName":"","lastName":"Liu","suffix":""},{"id":541912197,"identity":"2b54a547-fe96-47b3-b15d-176bf822d9ec","order_by":1,"name":"Jianghua Chen","email":"","orcid":"","institution":"Dazhou Vocational College of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jianghua","middleName":"","lastName":"Chen","suffix":""},{"id":541912198,"identity":"a8986b86-b3cf-4631-bb54-deecab3497fc","order_by":2,"name":"Jie Zhu","email":"","orcid":"","institution":"First Affiliated Hospital of Army Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jie","middleName":"","lastName":"Zhu","suffix":""},{"id":541912199,"identity":"c95ad62d-8bb6-4ec9-a7a2-53256ffe40f3","order_by":3,"name":"Dandan Wu","email":"","orcid":"","institution":"First Affiliated Hospital of Army Medical University","correspondingAuthor":false,"prefix":"","firstName":"Dandan","middleName":"","lastName":"Wu","suffix":""},{"id":541912200,"identity":"5559bd40-6d69-461d-a49a-5e2486e87879","order_by":4,"name":"Kui Du","email":"","orcid":"","institution":"First Affiliated Hospital of Army Medical University","correspondingAuthor":false,"prefix":"","firstName":"Kui","middleName":"","lastName":"Du","suffix":""},{"id":541912201,"identity":"a5d71d45-c9b5-4892-975a-6244490751cc","order_by":5,"name":"Fanxin Zeng","email":"","orcid":"","institution":"Dazhou Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"Fanxin","middleName":"","lastName":"Zeng","suffix":""},{"id":541912202,"identity":"14f4bb0a-de90-4f70-be3f-e18b4b0dfa99","order_by":6,"name":"Qinghua Zou","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBElEQVRIiWNgGAWjYLCCBDBiP3DgQwVUhIc4LTyJB2ecMSBSC0QXg/Fh3jYitMi39xjeeLijNs/g+IGEgzPn/YnWnZHA+OBtG4O8OQ4tjD1njC0SzxwvNjiTeODAx20GudtuJDAbzm1jMNzZgF0Ls0SOmURi27HEDQcSgLZAtLBJ87YxJBgcwK6FDa7l/AODw7xzwFrYf+PTwgPRUpO44UYCUEsDxBZmfFokeI4VWyS2HUiceeNNwsEZx4xzt5152Cw555yE4QYcWuTbmzfe/NlWl9h3Pv3whw81crnbjicf/PCmzEYely1gmxgYDiPzGRsggngAULYOr4JRMApGwSgY4QAApitp5I5wZPoAAAAASUVORK5CYII=","orcid":"","institution":"First Affiliated Hospital of Army Medical University","correspondingAuthor":true,"prefix":"","firstName":"Qinghua","middleName":"","lastName":"Zou","suffix":""}],"badges":[],"createdAt":"2025-10-12 09:23:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7839620/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7839620/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s40001-026-04452-3","type":"published","date":"2026-04-18T15:57:13+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":95661736,"identity":"4a3a54cd-0547-4309-85fc-9b82ee265c33","added_by":"auto","created_at":"2025-11-11 16:36:48","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1103658,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/003fbc0e839603fc7a33b113.docx"},{"id":95661743,"identity":"4fc00c2f-515c-4a52-a8ab-0427aed58174","added_by":"auto","created_at":"2025-11-11 16:36:49","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9626,"visible":true,"origin":"","legend":"","description":"","filename":"2cf61ee6a7c84d989647afce57be8ee0.json","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/3e381e616c1b1c030a66d730.json"},{"id":95661878,"identity":"c2b6a1c1-b312-4d17-a1f2-a4899cb77ecb","added_by":"auto","created_at":"2025-11-11 16:36:58","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1046392,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymateria.docx","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/f706b7499ae0d30dc4e6ba2b.docx"},{"id":95661860,"identity":"0494b7d0-13b8-4ef3-b326-b61d7f731715","added_by":"auto","created_at":"2025-11-11 16:36:56","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":130806,"visible":true,"origin":"","legend":"","description":"","filename":"2cf61ee6a7c84d989647afce57be8ee01enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/d7752c4d0898ba6547e516b8.xml"},{"id":95661881,"identity":"77081b1c-0ada-4f6a-9d29-b94662707b7e","added_by":"auto","created_at":"2025-11-11 16:36:58","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":44224,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/696408acbfa6abbad4f34f9f.png"},{"id":95662059,"identity":"b26e826e-a959-4385-b503-61f2a77c2536","added_by":"auto","created_at":"2025-11-11 16:37:08","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":53021,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/c58b6221504c89e1522bc5e9.png"},{"id":95662106,"identity":"6bf829ab-d365-4701-9878-aaadc0423824","added_by":"auto","created_at":"2025-11-11 16:37:11","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":27793,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/59958daddcd662d107d36fef.png"},{"id":95662087,"identity":"b220f78d-cff3-466a-9c6b-b92bf262fea2","added_by":"auto","created_at":"2025-11-11 16:37:10","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":38944,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/7e643a8cb6bd6440ab780c6f.png"},{"id":95662055,"identity":"ef3d0eb9-3cb1-4148-9220-51ddfc77285c","added_by":"auto","created_at":"2025-11-11 16:37:07","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":49984,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/75b951c3286b7e13cf006266.png"},{"id":95662061,"identity":"729f8b81-142b-4e84-9107-e1dd33914a95","added_by":"auto","created_at":"2025-11-11 16:37:08","extension":"xml","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":130759,"visible":true,"origin":"","legend":"","description":"","filename":"2cf61ee6a7c84d989647afce57be8ee01structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/765677b392d8e8cc845f138a.xml"},{"id":95661892,"identity":"67f262f8-b53d-466b-9de7-b7f109ae660f","added_by":"auto","created_at":"2025-11-11 16:37:00","extension":"html","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":141032,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/691f517112c16032739b5110.html"},{"id":95661760,"identity":"a79ddd08-af75-4cf1-a29e-b2ee706a8189","added_by":"auto","created_at":"2025-11-11 16:36:51","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":122787,"visible":true,"origin":"","legend":"\u003cp\u003ePatients enrollment flowchart. Note: HRCT: high-resolution computed tomography; UIP: usual interstitial pneumonia; NSIP: non-specific interstitial pneumonia. Institution 1: First Affiliated Hospital of Army Medical University. Institution 2: Dazhou Central Hospital.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/511bc8c7d78186cf36f6dc54.png"},{"id":95662236,"identity":"357a312e-15c4-44ee-8772-0c13b179dd0c","added_by":"auto","created_at":"2025-11-11 16:37:18","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":223256,"visible":true,"origin":"","legend":"\u003cp\u003eLASSO for VOI-based screening of radiomics feature sets and process maps.A: the optimal set of features retained by the LASSO ; B: a plot showing the mean square error of the optimal tuning parameter (λ) determined by the LASSO regression model and applying 10-fold cross-validation; C: a plot showing the coefficient variations of the 25 radiomics features with non-zero coefficients based on the selected value of λ, where each colored line represents the trajectory of a feature coefficient.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/1af64ab85c30bee1522caadf.png"},{"id":95661890,"identity":"b59a875f-b0a4-421c-b30e-5986bf3bc9fa","added_by":"auto","created_at":"2025-11-11 16:36:59","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":262747,"visible":true,"origin":"","legend":"\u003cp\u003eRadiomics-clinical combined model predicts UIP Patterns in Nomogram. WBC: white blood cell count; Rad-score: Radiomics score.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/b3fd2966b868bfac5902974b.png"},{"id":95662066,"identity":"45782736-be10-41be-b4f6-fdcf2e7967be","added_by":"auto","created_at":"2025-11-11 16:37:08","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":164077,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration curves of the nomogram for UIP prediction in the training set, internal validation set, and external validation set and DCA curves for three models. Calibration curve on the training set(A), internal validation set(B) and external validation set(C). DCA curves for each model on the training set(D), internal validation(E) and external validation set(F).\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/a1e6a2f2d4529d2cd729bcc1.png"},{"id":95661896,"identity":"d58dfd57-d587-4860-9c4e-51777ae08e0f","added_by":"auto","created_at":"2025-11-11 16:37:00","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":247523,"visible":true,"origin":"","legend":"\u003cp\u003eAssociation analysis of radiomics and metabolomics features. (A) The flow chart of the association analysis of omics features. (B) The KEGG enrichment analysis of 77 metabolites. (C) The metabolic regulatory network has a significant enrichment of metabolic features. Blue circles represent metabolic pathways; orange squares represent metabolites.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/69cd8b926c0c3de28cc4d6ea.png"},{"id":107351008,"identity":"069fd204-41b6-41b8-8b15-b74c794a11f6","added_by":"auto","created_at":"2026-04-20 16:07:41","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1449204,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/c1dd24ce-cea6-45f7-a6d4-9f107cfe8ef3.pdf"},{"id":95662090,"identity":"17867dd0-29fc-4aea-9ad8-c163159e90f2","added_by":"auto","created_at":"2025-11-11 16:37:10","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":1046392,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymateria.docx","url":"https://assets-eu.researchsquare.com/files/rs-7839620/v1/6bb3489da8455e25bb3433ca.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Combining radiomics based on high-resolution computed tomography with plasma metabolomics for diagnosing subtypes of rheumatoid arthritis-associated interstitial lung disease","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIt is estimated that approximately 10% of individuals with rheumatoid arthritis (RA) will develop clinically significant interstitial lung disease (ILD) \u003csup\u003e\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Patients with RA-associated ILD (RA-ILD) exhibit a higher mortality rate compared with RA patients without ILD \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. RA-ILD can be classified as either a usual interstitial pneumonia (UIP) pattern or a non-UIP pattern (mainly non-specific interstitial pneumonia (NSIP)), with UIP as the pathologic manifestation in 54%\u0026ndash;65% of patients with RA-ILD \u003csup\u003e\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. A UIP pattern on high-resolution computed tomography (HRCT) is associated with a worse prognosis and a higher risk of acute exacerbation compared to NSIP pattern in patients with RA-ILD \u003csup\u003e\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. Therefore, accurate differentiation between UIP and NSIP patterns is critical for guiding treatment decisions and improving patient outcomes.\u003c/p\u003e\u003cp\u003eThere is substantial evidence to support the use of HRCT instead of lung biopsy to distinguish ILD subtypes \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. However, due to the large overlap of CT features between NSIP and UIP patterns, subjective differences in the identification and interpretation of abnormalities detected on imaging \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, the overall diagnostic accuracy of CT in distinguishing between UIP and NSIP is about 70% \u003csup\u003e14\u003c/sup\u003e. As a result, increasing research is focusing on the development of computer-aided tools and the application of artificial intelligence for rapid assessment. Radiomics is a bridge between medical imaging and personalized medicine. By applying high-throughput quantitative image features extracted from medical images to clinical decision support systems, radiomics can improve the accuracy of diagnosis, prognosis and prediction \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e.For example, the machine learning model developed by Koo et al. can successfully differentiate pathologically confirmed UIP, NSIP, and chronic hypersensitivity pneumonitis \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Sun et al. \u003csup\u003e17\u003c/sup\u003e developed a lung graph-based machine learning model for identifying fibrous ILDs, which outperforms the diagnostic performance of radiologists and may help clinicians objectively assess ILDs. These results suggest that radiomics are a potential tool for the diagnosis of RA-ILD. However, radiomics faces a significant challenge: the lack of biological interpretability of radiomic features, which limits its clinical translation \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eMetabolomics, which quantifies dynamic changes in metabolites, offers a complementary approach to uncover the biological mechanisms underlying disease processes \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e.Rindlisbacher et al. \u003csup\u003e20\u003c/sup\u003e identified the unique metabolic characteristics of patients with idiopathic pulmonary fibrosis through serum metabolic profiling, and lysophosphatidylcholine may be a potential marker for diagnosis and monitoring of idiopathic pulmonary fibrosis. Our previous multi-omics studies revealed the central role of glycerophospholipid metabolism in the development of rheumatoid arthritis \u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Integrating radiomics with metabolomics provides a non-invasive and informative strategy to enhance diagnostic accuracy and reveal the biological significance of imaging features.\u003c/p\u003e\u003cp\u003eDespite these advances, no studies have comprehensively explored the combined use of radiomics and metabolomics to differentiate UIP and NSIP patterns in RA-ILD. Therefore, this study aims to develop and validate combined clinical-radiomics models for RA-ILD diagnosis and to investigate the biological relevance of radiomic features through correlation analysis with plasma metabolites.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eOur Institutional Review Board (The Army Medical University Medical Ethics Committee and Dazhou Central Hospital Medical Ethics Committee) approved this study and waived the requirement to obtain informed consent because this was a retrospective analysis.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy population\u003c/h2\u003e\u003cp\u003eThe study populations comprised RA-ILD patients who were consecutively recruited from January 2019 to January 2024 at the First Affiliated Hospital of Army Medical University, and from January 2017 to December 2023 at Dazhou Central Hospital. All patients met the 2010 RA classification criteria of the American College of Rheumatology/European League Against Rheumatism (ACR/EULAR)\u003csup\u003e22\u003c/sup\u003e. The presence of ILD was confirmed by chest HRCT in all cases, evaluated by two senior radiologists using a single-blind method (who were unaware of the clinical information and imaging reports of the patients). The type of ILD was either UIP pattern or NSIP pattern, which was evaluated by a multidisciplinary team (MDT) (including pulmonologists, radiologists, pathologists and rheumatologists) based on the criteria of the American Thoracic Society/European Respiratory Society (ATS/ERS) International Multidisciplinary Consensus Classification of the Idiopathic Interstitial Pneumonias \u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. The flowchart of patient enrollment is shown in (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The research workflow of radiomics analysis is shown in (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eHRCT image acquisition\u003c/h3\u003e\n\u003cp\u003eHRCT images of all enrolled patients were collected according to a standard chest imaging protocol. The images contained 116\u0026ndash;560 layers, with layer thicknesses of 0.63\u0026ndash;2.00 mm. Each layer was reconstructed to a matrix size of 512 \u0026times; 512 with an in-plane pixel spatial resolution ranging from 0.56 mm \u0026times; 0.56 mm to 0.92 mm \u0026times; 0.92 mm. A lung window (L/W: -500 HU/1500 HU) was used to interpolate CT numbers (Hounsfield units) to 1 mm \u0026times; 1 mm \u0026times; 1 mm per scan, reducing inter-scan variations (e.g., slice thickness).\u003c/p\u003e\n\u003ch3\u003eDemographic, Laboratory Data and Clinical signature\u003c/h3\u003e\n\u003cp\u003eThe following patient characteristics were included: clinical characteristics (age, sex), laboratory data such as rheumatoid factors (RF), erythrocyte sedimentation rate (ESR), c-reactive protein (CRP), white blood cell count (WBC), monocyte, lymphocyte, basophil, eosinophil, and neutrophil counts and percentages, glutamyl transpeptidase (GGT), alanine aminotransferase (ALT), creatinine (Cr), and other relevant markers. A univariate analysis of all clinical data was performed to screen for significant factors related to outcome events.\u003c/p\u003e\n\u003ch3\u003eMetabolomic Data\u003c/h3\u003e\n\u003cp\u003eIn the external validation set, plasma samples from patients underwent liquid chromatography-tandem mass spectrometry analysis using an AB SCIEX UPLC-TripleTOF system. We acquired raw metabolic data and initially pre-processed it using Progenesis QI software (Waters Corporation, Milford, USA). This involved eliminating missing values exceeding 80% within each group, imputing missing values with the minimum observed, using sum normalization, discarding metabolites with relative standard deviations over 30% in quality control samples, and applying log10 transformations. Subsequently, we matched the mass spectral data of the metabolites with the Human Metabolome Database (HMDB, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.hmdb.ca/\u003c/span\u003e\u003cspan address=\"http://www.hmdb.ca/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) to acquire the final metabolite expression profiles for further analysis. Spearman correlation analysis was used to identify plasma metabolites associated with radiomics features (R\u0026thinsp;\u0026gt;\u0026thinsp;0.4, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were carried out on the online platform of Majorbio (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cloud.majorbio.com\u003c/span\u003e\u003cspan address=\"https://cloud.majorbio.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). A detailed description of the metabolomics methods can be found in Supplementary Material.\u003c/p\u003e\n\u003ch3\u003eRadiomics Feature extraction, selection and Rad-score calculation\u003c/h3\u003e\n\u003cp\u003eFeature extraction was performed using the AI Platform of the First Affiliated Hospital of Army Medical University. The platform\u0026rsquo;s U-net model automatically segmented lungs and extracted volumes of interest (VOI) (Fig. S2). We used three feature selection methods, variance threshold (\u0026gt;\u0026thinsp;0.8), univariate selection (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05), and the least absolute shrinkage and selection operator (LASSO) to select the features. First, the variance threshold initially removed 263 features that contributed little to the model, then the univariate selection excluded 505 features that were not sufficiently significantly (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05) associated with the RA-ILD subtype, and finally LASSO was conducted to choose 25 optimized subsets of features to construct the model (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Radiomics score (Rad-score) was calculated by summing the selected features weighted by their coefficients.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eModel construction and Nomogram\u003c/h2\u003e\u003cp\u003eLogistic regression (LR), random forest (RF), and multi-layer perceptron (MLP) were employed to build the radiomics model, clinical model, and combined radiomics-clinical model, respectively. Furthermore, a clinical-radiomics nomogram was constructed via multivariable logistic regression to visually present the probability of UIP pattern, incorporating Rad-score and independent clinical predictors.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eStatistical analyses were performed in R (version 4.3.0) and Python (version 3.7.0). Univariate analysis was used to compare the differences in clinical factors between UIP and NSIP patients. Statistical tests included corrected chi-square tests (categorical variables) and one-way analysis of variance (ANOVA) (normally distributed continuous variables); descriptive statistics were presented as frequencies (percentages) or mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD), respectively. The Kruskal-Wallis H test was used if continuous variables were not normally distributed, and descriptive statistics were median (upper and lower quartiles).The receiver operating characteristic (ROC) curve was drawn and AUC was calculated to evaluate the diagnostic performance of the models. As well as analyzing the accuracy, sensitivity, and specificity of the models. Goodness-of-fit of the nomogram was evaluated using the Hosmer-Lemeshow test, while its predictive accuracy was assessed using calibration curves. Decision curve analysis (DCA) was performed to evaluate the clinical value of the models independently.. Calibration curves, DCA, and univariate analysis were implemented in R with the \"rms\" and \"stats\" packages. The Spearman correlation coefficient was calculated and visualized using the 'ComplexHeatmap' package in R. \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eBaseline demographic and clinical features\u003c/h2\u003e\u003cp\u003e301 patients (166 with UIP, 135 with NSIP) from the First Affiliated Hospital of Army Medical University were randomly divided into a training set (n\u0026thinsp;=\u0026thinsp;240) and an internal validation set (n\u0026thinsp;=\u0026thinsp;61) using computer-generated random numbers. 49 patients (27 in UIP and 22 in NSIP) from Dazhou Central Hospital were used as the external validation set. No significant clinical factor differences were found between the training set and the internal validation set (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05) (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eRadiomics feature construction\u003c/h2\u003e\u003cp\u003eIn this study, 1688 radiomics features were extracted from each VOI, including 324 first-order, 14 shape, 432 gray-level co-occurrence matrix, 252 gray-level dependence matrix, 288 gray-level run-length matrix, 90 neighboring gray-tone difference matrices, and 288 gray-level size zone matrix features. 25 features were ultimately selected and contributed to the Rad-score construction.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eNomogram variable screening\u003c/h2\u003e\u003cp\u003eIn univariate analysis, the variables sex (odds ratio [OR]\u0026thinsp;=\u0026thinsp;2.658, 95% confidence interval [CI]: 1.673\u0026ndash;4.377, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and WBC count (OR\u0026thinsp;=\u0026thinsp;0.876, 95% CI: 0.799\u0026ndash;0.955, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.004), monocyte count (OR\u0026thinsp;=\u0026thinsp;0.326, 95% CI: [0.111\u0026ndash;0.926], \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.038), and neutrophil count (OR\u0026thinsp;=\u0026thinsp;0.876, 95% CI: [0.792\u0026ndash;0.964], \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.008) were remained as independent predictors in the clinical factors model (Supplementary Table S2).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eModel development and evaluation\u003c/h2\u003e\u003cp\u003eThree categories of models were constructed using RF, LR, and MLP\u0026mdash;with each algorithm applied to all model types, resulting in 9 total models. Specifically: Clinical models were built based on sex, WBC, monocyte, and neutrophil counts (Supplementary Fig. S3); Radiomics models were developed using the Rad-score (Supplementary Fig. S4); Combined clinical-radiomics models integrated clinical variables and the Rad-score.\u003c/p\u003e\u003cp\u003eThe combined models showed the best discriminative performance, with significantly higher AUC values than the clinical and radiomics models (Table\u0026nbsp;1). The MLP combined model showed more favourable discrimination in both the training set (AUC\u0026thinsp;=\u0026thinsp;0.915; 95% CI: 0.844, 0.940) and internal validation set (AUC\u0026thinsp;=\u0026thinsp;0.828; 95% CI: 0.728, 0.916).( Supplementary Fig. S5A). The RF combined model did the best discrimination in the external validation set (AUC\u0026thinsp;=\u0026thinsp;0.841; 95% CI: 0.729, 0.916) (Supplementary Fig. S5C), but its specificity was low at 67%.\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;1: Performance of all 9 models (3 algorithms \u0026times; 3 model categories)\u003c/p\u003e\u003c/div\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"134%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003eData set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMODEL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003eAUC (95% CI)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003eSensitivity\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003eTraining set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.683(0.625～0.738)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.650\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.648\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.652\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP Radiomics model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.831(0.784～0.869)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.750\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.704\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.788\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.915(0.884～0.940)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.821\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.917\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.742\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR Clinical model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.645(0.589～0.706)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.629\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.667\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.598\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR Radiomics model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.820(0.774～0.862)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.733\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.787\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.689\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.829(0.785～0.872)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.750\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.750\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.750\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.772(0.729～0.825)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.704\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.796\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.629\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF Radiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.839(0.797～0.876)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.771\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.833\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.720\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.893(0.860～0.927)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.833\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.833\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.833\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003eInternal validation set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.629(0.492～0.756)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.672\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.519\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.794\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP Radiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.801(0.710～0.889)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.754\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.852\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.676\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.828(0.728～0.916)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.787\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.815\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.765\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.641(0.523～0.760)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.641\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.672\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.704\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR Radiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.803(0.688～0.706)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.803\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.787\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.667\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.820(0.714～0.902)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.820\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.803\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.778\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.536(0.390～0.643)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.536\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.607\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.593\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF Radiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.815(0.713～0.893)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.815\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.754\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.852\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.815(0.712～0.892)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.815\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.803\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.815\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003eExternal Validation Set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.589(0.448～0.724)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.612\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.818\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.444\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP Radiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.762(0.623～0.861)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.776\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.773\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.778\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eMLP combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.731(0.583～0.838)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.714\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.864\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.593\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.579(0.435～0.715)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.818\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.444\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.435\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR Radiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.778(0.643～0.878)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.796\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.818\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.778\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eLR combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.763(0.625～0.863)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.755\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.818\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.704\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF Clinical model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.554(0.409～0.713)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.673\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.318\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.963\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF Radiomics model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.826(0.692～0.907)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.776\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.636\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.889\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 15.7738%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5675%;\"\u003e\n \u003cp\u003eRF combined model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.6508%;\"\u003e\n \u003cp\u003e0.841(0.729～0.916)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7817%;\"\u003e\n \u003cp\u003e0.755\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.4603%;\"\u003e\n \u003cp\u003e0.864\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.7659%;\"\u003e\n \u003cp\u003e0.667\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAbbreviations: MLP, multi-layer perceptron; LR, logistic regression; RF, random forest.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNomogram construction and prediction performance assessment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe\u0026nbsp;sex,\u0026nbsp;WBC, monocyte and\u0026nbsp;neutrophil\u0026nbsp;counts and Rad-score were incorporated into the clinical-radiomics nomogram (Fig. 3). To use the nomogram, first, we needed to determine the position of each variable (including sex, WBC, monocyte count, neutrophil count, and Rad-score) on the corresponding axis. Next, we drew a vertical line upwards from these positions to the point axis to obtain the corresponding score for each variable. We summed these scores to arrive at a total score. Lastly, we drew a vertical line downwards from the total point axis, and the value corresponding to this line represented the probability of the UIP pattern. Calibration curves (Fig. 4A-C) showed good consistency and the Hosmer\u0026ndash;Lemeshow tes\u0026nbsp;yielded non-significant results for both the training set (\u003cem\u003eP\u003c/em\u003e = .051), the internal validation set (\u003cem\u003eP\u003c/em\u003e = 0.382) and the external validation set (\u003cem\u003eP\u003c/em\u003e = 0.073). The DCA (Fig. 4D-F) showed that the nomogram exhibits high net benefits over most threshold probability ranges.This indicated that clinical intervention based on the nomogram could benefit patients in the training set, and also in the internal and external validation sets.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePotential metabolic features of radiomics model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNext, we analyzed the plasma metabolites of RA-ILD and identified a total of 330 different plasma metabolites in the HMDB database. The workflow of correlation analysis between radiomics features and metabolomics features is shown in Fig. 5A. Spearman correlation analysis of the 25 optimal radiomics features with 330 metabolites showed that 77 metabolites and radiomics features were significantly correlated (R \u0026gt; 0.4, \u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u0026lt; 0.05). The 77 correlated metabolites were mainly enriched in pathways involving D-glutamine and D-glutamate metabolism, arginine biosynthesis, alanine, aspartate and glutamate metabolism, glycerophospholipid metabolism, and caffeine metabolism (corrected \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05, Fig. 5B). There were 11 metabolites enriched in these pathways, with L-glutamate and alpha-ketoglutarate co-enriched in the arginine biosynthesis, D-glutamine and D-glutamate metabolism, alanine, aspartate, and glutamate metabolism pathways comprising the regulatory network (Fig. 5C).\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eDespite significant advances in the diagnosis and treatment of RA, early diagnosis of RA-ILD remains challenging. A Danish population-based cohort study showed a 5-year mortality rate of 39.0% in patients with RA-ILD compared with 18.2% in patients without ILD \u003csup\u003e24\u003c/sup\u003e. In patients with RA-ILD, a UIP pattern demonstrated by HRCT scan or histology has a higher mortality rate compared to other patterns of pulmonary fibrosis \u003csup\u003e25-27\u003c/sup\u003e. Therefore, accurate differentiation between UIP and NSIP patterns is essential to guide treatment decisions and improve patient outcomes \u003csup\u003e28\u003c/sup\u003e. In this study, we developed and validated combined clinical-radiomics models to distinguish UIP and NSIP patterns, demonstrating superior diagnostic performance compared to traditional clinical models.\u003c/p\u003e\n\u003cp\u003eThe introduction of CT radiomics has provided a new perspective on the diagnosis of RA-ILD. While traditional radiological methods focus on macro-level analysis, radiomics are capable of uncovering hidden information in the image data, such as heterogeneity in lung texture and microscopic structura changes. In this study, the Rad-score was used to construct radiomics models, which exhibited good diagnostic performance on the training set, internal validation set and external validation set. The RF radiomics model demonstrated favorable discrimination, with AUC values of 0.839 in the training set, 0.815 in the internal validation set, and 0.826 in the external validation set. Compared with the clinical models, the diagnostic performance of the radiomics models was significantly improved, which reflected the increased value of Rad-Score in classification diagnosis. At the same time, comprehensive evaluation of the performance of all models showed that the performance of radiomics models in the three datasets was relatively stable, which confirmed the stability and consistency of radiomics features in the identification of RA-ILD subtypes. Interestingly, the study by Qiu et al. \u003csup\u003e29\u003c/sup\u003ealso found that Rad-score has potential ability to differentiate between immune checkpoint inhibitor-related pneumonitis and radiation pneumonitis. Unlike radiomics studies that were based on cases diagnosed by pathology\u003csup\u003e16 30\u003c/sup\u003e, our study was based on clinical real-world cases, which somewhat avoids selection bias and improves the clinical applicability of radiomics studies. In addition, most of the combined models constructed by the same algorithm further improved their performance compared with their corresponding radiomics and clinical models, showing higher AUC, ACC, sensitivity and specificity. This reflects the multiple advantages of the combined models, which not only makes full use of existing medical resources, but also can realistically simulate the diagnostic thinking mode of clinicians. To transform the combined model into an intuitive visual computational tool, we drew a nomogram of the combined model in this study. The calibration curve showed that the UIP pattern prediction probability matched the actual observation. DCA showed that the combined model had better net benefits. This indicates that clinicians can use the nomogram to intuitively and easily distinguish RA-UIP from RA-NSIP. Recently, some studies have used deep learning methods\u003csup\u003e31-34\u003c/sup\u003e, for example, Chung et al. \u003csup\u003e34\u003c/sup\u003edeveloped a deep learning-based radiomics classifier for the diagnosis of UIP from chest CT scans in a retrospective study, which included a total of 2907 chest CT scans, and the classifier demonstrated a high level of sensitivity and specificity in different cohorts.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUnivariate analysis indicated that\u0026nbsp;sex, WBC, monocyte and\u0026nbsp;neutrophil\u0026nbsp;counts were independent predictors for diagnosing RA-ILD patterns. Although risk factors for the development of RA-ILD include older age, male sex, smoking history, poor control of arthritis activity, and high levels of RF and anti-citrullinated protein antibodies \u003csup\u003e7\u003c/sup\u003e. However, no study has reported significant differences in RF between UIP and NSIP, and our study is also the case. Male sex was associated with the presence of UIP patterns, which is consistent with previous studies\u003csup\u003e35\u003c/sup\u003e, but no study has mentioned that\u0026nbsp;WBC, monocyte and\u0026nbsp;neutrophil\u0026nbsp;count\u0026nbsp;were significantly different between UIP and NSIP patterns.\u0026nbsp;The previous study showed that monocytes, platelets, neutrophils, and lymphocytes play a crucial role in the immune response in RA and that platelet/lymphocyte ratio can be used as a biomarker to differentiate RA-ILD from RA\u003csup\u003e36\u003c/sup\u003e. This suggests that inflammation may play a role in the pathogenesis and progression of pulmonary fibrosis.\u0026nbsp;Recent studies have reported that older age, seropositivity, and male sex were strongly associated with RA-UIP, whereas RA-related autoantibodies were associated with RA-NSIP \u003csup\u003e37\u003c/sup\u003e. The study by Wang et al.\u003csup\u003e38\u0026nbsp;\u003c/sup\u003eshowed that the AUC of serum UA for predicting RA-UIP was 0.845 (95%CI: 0.78-0.91; \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.01), indicating that the elevated serum UA concentration may be a diagnostic marker for RA-UIP. However, in this study, no significant difference was found in older age, seropositivity, and serum UA between UIP and NSIP. One possible reason is that the sample size of this study is small,and a large prospective cohort study is needed to confirm.\u003c/p\u003e\n\u003cp\u003eRecently, there has been a surge of research focusing on the genomic, cytokine, antibody, proteomic and metabolomic profiles of RA-ILD, aimed to enhance the development of biomarkers for this condition \u003csup\u003e39-40\u003c/sup\u003e. Proteomic analyses have identified molecular signatures that are strongly associated with the presence and severity of RA-ILD, providing insights into unexplored disease pathways \u003csup\u003e40\u003c/sup\u003e. Notably, researchers found that serum levels of decanoic acid, glycerol, and morpholine were different in RA patients with or without ILD\u003csup\u003e41\u003c/sup\u003e. The partial least squares-discriminant analysis model generated from these three metabolites could successfully discriminate ILD in RA (AUC: 0.919, 95% CI: 0.867-0.968, sensitivity 0.880, specificity 0.780). In our study, 330 plasma metabolites of RA-ILD were correlated with 25 radiomics features, and 77 correlated metabolites were identified. In the pathway enrichment analysis, D-glutamine and D-glutamate metabolism, arginine biosynthesis, alanine, aspartate and glutamate metabolism, glycerophospholipid metabolism, and caffeine metabolism pathways were significantly enriched. These results suggest that these five metabolic pathways may be closely related to the pathophysiological process of RA-ILD.\u0026nbsp;Among them, arginine biosynthesis, alanine, aspartate and glutamate metabolism are associated with disorders of amino acid metabolism in different stages of RA \u003csup\u003e42\u003c/sup\u003e. L-glutamate and\u0026nbsp;alpha-ketoglutarate are enriched in a variety of metabolic pathways, and their elevation may be related to the increased activity of RA-ILD, suggesting that they may be\u0026nbsp;have potential for further research, but this needs to be further verified by a larger sample size. According to recent investigations, glutamate contributes to pulmonary fibrosis \u003csup\u003e43-44\u003c/sup\u003e and citrulline is also higherexpression in RA-ILD lung tissues \u003csup\u003e45\u003c/sup\u003e.\u0026nbsp;By deeply exploring the correlation between radiomics features and metabolomics, we are expected to reveal the potential biological significance behind these radiomics features.\u003c/p\u003e\n\u003cp\u003eThis study has several limitations that should be considered when interpreting the findings. First, the retrospective design inherently introduces potential selection bias, as patient enrollment relied on existing medical records at two single-center institutions. While we ensured no significant differences in baseline characteristics between the training and internal validation sets, the generalizability of our results to broader populations—especially those from diverse geographic regions or healthcare settings—may be limited. Prospective, multicenter studies with standardized data collection protocols are therefore needed to validate the performance of the combined clinical-radiomics model.\u003c/p\u003e\n\u003cp\u003eSecond, RA-ILD subtype classification was based on HRCT findings and MDT assessment (including pulmonologists, radiologists, pathologists, and rheumatologists) rather than surgical lung biopsy—the pathological gold standard for interstitial lung disease (ILD) diagnosis. Although MDT consensus aligns with clinical practice guidelines for ILD evaluation \u003csup\u003e23\u003c/sup\u003e and reduces interobserver variability, the absence of histopathological confirmation may introduce residual diagnostic uncertainty, particularly for cases with overlapping radiological features between UIP and NSIP patterns.\u003c/p\u003e\n\u003cp\u003eThird, radiomics features were extracted from the entire lung parenchyma rather than focal ILD lesions. While this approach captures global lung involvement, it may dilute the signal from lesion-specific textural or density characteristics that are most relevant to subtype differentiation. Future studies could optimize feature extraction by focusing on manually or automatically segmented lesion regions to enhance the specificity of radiomic signatures.\u003c/p\u003e\n\u003cp\u003eFourth, incomplete outpatient medical records led to missing data on key variables with reported associations with RA-ILD pathogenesis, including smoking history and anti-citrullinated protein antibody (ACPA) status \u003csup\u003e46-47\u003c/sup\u003e. The exclusion of these variables from model construction may have underestimated their potential predictive value and limited the comprehensiveness of the clinical-radiomics integration.\u003c/p\u003e\n\u003cp\u003eFifth, our cohort only included patients with UIP or NSIP patterns, which are the most common RA-ILD subtypes but do not represent the full spectrum of RA-associated ILDs (e.g., cryptogenic organizing pneumonia, lymphoid interstitial pneumonia). The performance of our model in distinguishing these less frequent subtypes remains unknown and requires further investigation.\u003c/p\u003e\n\u003cp\u003eFinally, while we identified correlations between radiomic features and plasma metabolites (e.g., L-glutamate, alpha-ketoglutarate), the biological mechanisms linking these imaging biomarkers to metabolic dysregulation in RA-ILD remain unclear. Radiomics features often lack direct biological interpretability \u003csup\u003e18\u003c/sup\u003e, and additional multi-omics studies (e.g., integrating genomics or proteomics) are needed to uncover the molecular pathways underlying the observed associations, which could improve the clinical translatability of radiomic-based tools.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eIn this study, the clinical-radiomics combined model (nomogram) can be used as a tool to distinguish the subtypes of RA-ILD, which is useful for the early and accurate identification of RA-ILD subtypes to support clinical decision-making. The Rad-score has considerable potential in distinguishing RA-ILD subtypes and deserves further exploration. Building on traditional radiomics, this study integrated metabolomics with radiomics to identify metabolites associated with RA-ILD-related radiomics features, highlighting the potential of multi-omics strategies to advance the precision diagnosis and treatment of RA-ILD. Future research should focus on validating these findings in larger cohorts, exploring the biological roles of key metabolites, and further optimizing the combined models for clinical application.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eRA\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003erheumatoid arthritis\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eILD\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003einterstitial lung disease\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eRA-ILD\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eRheumatoid arthritis-associated interstitial lung disease\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eUIP\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eusual interstitial pneumonia\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eNSIP\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003enon-specific interstitial pneumonia\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eHRCT\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ehigh-resolution computed tomography\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eRF\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003erandom forest / rheumatoid factors\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eLR\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003elogistic regression\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eMLP\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003emulti-layer perceptron\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eRad-score\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eradiomics score\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003earea under the curve\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eDCA\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003edecision curve analysis\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eACR/EULAR\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eAmerican College of Rheumatology/European League Against Rheumatism\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eMDT\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003emultidisciplinary team\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eATS/ERS\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eAmerican Thoracic Society/European Respiratory Society\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eESR\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eerythrocyte sedimentation rate\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eCRP\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ec-reactive protein\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eWBC\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ewhite blood cell count\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eGGT\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eglutamyl transpeptidase\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eALT\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ealanine aminotransferase\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eCr\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ecreatinine\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eKEGG\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eKyoto Encyclopedia of Genes and Genomes\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eVOI\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003evolume of interest\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eLASSO\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eleast absolute shrinkage and selection operator\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eROC\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ereceiver operating characteristic\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eOR\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eodds ratio\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eCI\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003econfidence interval\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cb\u003eACPA\u003c/b\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eanti-citrullinated protein antibody\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study involves human participants and human data/material, and was conducted in compliance with the World Medical Association Declaration of Helsinki (WMA Declaration of Helsinki – Ethical Principles for Medical Research Involving Human Participants). It was approved by the Ethics Committees of the First Affiliated Hospital of Army Medical University (Ethic Reference: (B)KY2023148) and Dazhou Central Hospital (Ethic Reference: 2021-022). Informed consent was waived due to the retrospective design and de-identification of patient data/material.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability statement\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe data supporting the findings of this study are available from the corresponding author upon reasonable request. The high-resolution computed tomography (HRCT) radiomics features, plasma metabolomics profiles, and clinical data used in this study have been de-identified to protect the privacy of participants (in compliance with the Declaration of Helsinki and ethical approval requirements of participating institutions). All data relevant to the study are included in the article and its supplementary materials.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis work was funded by the Individualized Training Grant for Key Talents of the Outstanding Talent Pool of Army Medical University (XA-2019-505-40) and the key projects of the Chongqing Science and Health Joint Traditional Chinese Medicine Research Project(2024ZYZD007). The funding sources did not have any role in the collection, analysis and interpretation of data. The Financial contributors did not influence the study design, collection, analysis and interpretation of data, the writing of the manuscript or the decision to submit the manuscript for publication.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributors\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHL, JC, JZ, DW and DK: patient selection and recruitment; HL, JC, JZ, FZ and QZ: experimental design; HL, JC, JZ, DW and DK: data collection; HL, JC, JZ and QZ: data analysis; HL, JC, JZ, FZ and QZ: manuscript writing and revision; and QZ: funding. Guarantor: QZ.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to thank the doctors and nursing team of\u0026nbsp;Department of Rheumatology and Immunology of the First Affiliated Hospital of Army Medical University and the Clinical Medical Research Center of Dazhou Central Hospital.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial registration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ehttp://www.chictr.org.cn (ChiCTR2400084726).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eName of registry\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eChinese Clinical Trial Registry.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eOlson AL, Swigris JJ, Sprunger DB, et al. Rheumatoid arthritis-interstitial lung disease-associated mortality. \u003cem\u003eAmerican journal of respiratory and critical care medicine\u003c/em\u003e 2011;183(3):372\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZou YQ, Li YS, Ding XN, et al. The clinical significance of HRCT in evaluation of patients with rheumatoid arthritis-associated interstitial lung disease: a report from China. \u003cem\u003eRheumatology international\u003c/em\u003e 2012;32(3):669\u0026ndash;73.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOh JH, Kim GHJ, Cross G, et al. Automated quantification system predicts survival in rheumatoid arthritis-associated interstitial lung disease. \u003cem\u003eRheumatology (Oxford)\u003c/em\u003e 2022;61(12):4702\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eManfredi A, Cassone G, Luppi F, et al. Rheumatoid arthritis related interstitial lung disease. \u003cem\u003eExpert review of clinical immunology\u003c/em\u003e 2021;17(5):485\u0026ndash;97.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZamora-Legoff JA, Krause ML, Crowson CS, et al. Patterns of interstitial lung disease and mortality in rheumatoid arthritis. \u003cem\u003eRheumatology (Oxford, England)\u003c/em\u003e 2017;56(3):344\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLee HK, Kim DS, Yoo B, et al. Histopathologic pattern and clinical features of rheumatoid arthritis-associated interstitial lung disease. \u003cem\u003eChest\u003c/em\u003e 2005;127(6):2019\u0026ndash;27.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAkiyama M, Alshehri W, Kaneko Y. Does a window of opportunity for rheumatoid arthritis-associated interstitial lung disease exist?. Autoimmunity Reviews, 2024, 23(2): 103501.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAtzeni F, Gerardi MC, Barilaro G, et al. Interstitial lung disease in systemic autoimmune rheumatic diseases: a comprehensive review. \u003cem\u003eExpert review of clinical immunology\u003c/em\u003e 2018;14(1):69\u0026ndash;82.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIzuka S, Yamashita H, Iba A, et al. Acute exacerbation of rheumatoid arthritis-associated interstitial lung disease: clinical features and prognosis. \u003cem\u003eRheumatology (Oxford, England)\u003c/em\u003e 2021;60(5):2348\u0026ndash;2354.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAkiyama, Mitsuhiro, and Yuko Kaneko. Pathogenesis, clinical features, and treatment strategy for rheumatoid arthritis-associated interstitial lung disease. \u003cem\u003eAutoimmunity reviews\u003c/em\u003e 2022;21(5):103056.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim EJ, Collard HR, King TE, Jr. Rheumatoid arthritis-associated interstitial lung disease: the relevance of histopathologic and radiographic pattern. \u003cem\u003eChest\u003c/em\u003e 2009;136(5):1397\u0026ndash;405.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTanaka N, Kim JS, Newell JD, et al. Rheumatoid arthritis-related lung diseases: CT findings. \u003cem\u003eRadiology\u003c/em\u003e 2004;232(1):81\u0026ndash;91.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWatadani T, Sakai F, Johkoh T, et al. Interobserver variability in the CT assessment of honeycombing in the lungs. \u003cem\u003eRadiology\u003c/em\u003e 2013;266(3):936\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMink SN, Maycher B. Comparative manifestations and diagnostic accuracy of high-resolution computed tomography in usual interstitial pneumonia and nonspecific interstitial pneumonia. \u003cem\u003eCurrent opinion in pulmonary medicine\u003c/em\u003e 2012;18(5):530\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749\u0026ndash;762.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKoo CW, Williams JM, Liu G, et al. Quantitative CT and machine learning classification of fibrotic interstitial lung diseases. \u003cem\u003eEuropean radiology\u003c/em\u003e 2022;32(12):8152\u0026ndash;61.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSun H, Liu M, Liu A, et al. Developing the Lung Graph-Based Machine Learning Model for Identification of Fibrotic Interstitial Lung Diseases. \u003cem\u003eJournal of imaging informatics in medicine\u003c/em\u003e 2024;37(1):268\u0026ndash;79.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTomaszewski M R, Gillies R J. The biological meaning of radiomic features. Radiology, 2021, 298(3): 505\u0026ndash;516.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi R, Li L, Xu Y, et al. Machine learning meets omics: applications and perspectives. Briefings in Bioinformatics, 2022, 23(1): bbab460.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRindlisbacher B, Schmid C, Geiser T, et al. Serum metabolic profiling identified a distinct metabolic signature in patients with idiopathic pulmonary fibrosis - a potential biomarker role for LysoPC. \u003cem\u003eRespiratory research\u003c/em\u003e 2018;19(1):7.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJian C, Wei L, Wu T, et al. Comprehensive multi-omics analysis reveals the core role of glycerophospholipid metabolism in rheumatoid arthritis development. Arthritis research \u0026amp; therapy 2023;25(1):246.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAletaha D, Neogi T, Silman AJ, et al. 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. \u003cem\u003eArthritis and rheumatism\u003c/em\u003e 2010;62(9):2569\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTravis WD, Costabel U, Hansell DM, et al. An official American Thoracic Society/European Respiratory Society statement: Update of the international multidisciplinary classification of the idiopathic interstitial pneumonias. \u003cem\u003eAmerican journal of respiratory and critical care medicine\u003c/em\u003e 2013;188(6):733\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHyldgaard C, Hilberg O, Pedersen AB, et al. A population-based cohort study of rheumatoid arthritis-associated interstitial lung disease: comorbidity and mortality. \u003cem\u003eAnnals of the rheumatic diseases\u003c/em\u003e 2017;76(10):1700\u0026ndash;06.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNieto M A, Rodriguez-Nieto M J, Sanchez-Pernaute O, et al. Mortality rate in rheumatoid arthritis-related interstitial lung disease: the role of radiographic patterns. BMC Pulmonary Medicine, 2021, 21: 205.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSingh N, Varghese J, England BR, et al. Impact of the pattern of interstitial lung disease on mortality in rheumatoid arthritis: A systematic literature review and meta-analysis. \u003cem\u003eSeminars in arthritis and rheumatism\u003c/em\u003e 2019;49(3):358\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSolomon JJ, Chung JH, Cosgrove GP, et al. Predictors of mortality in rheumatoid arthritis-associated interstitial lung disease. \u003cem\u003eThe European respiratory journal\u003c/em\u003e 2016;47(2):588\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRaghu G, Remy-Jardin M, Richeldi L, et al. Idiopathic pulmonary fibrosis (an Update) and progressive pulmonary fibrosis in adults: An official ATS/ERS/JRS/ALAT clinical practice guideline. American Journal of Respiratory and Critical Care Medicine, 2022, 205(9): e18-e47.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQiu Q, Xing L, Wang Y, et al. Development and Validation of a Radiomics Nomogram Using Computed Tomography for Differentiating Immune Checkpoint Inhibitor-Related Pneumonitis From Radiation Pneumonitis for Patients With Non-Small Cell Lung Cancer. \u003cem\u003eFrontiers in immunology\u003c/em\u003e 2022;13:870842.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDe Giacomi F, Raghunath S, Karwoski R, et al. Short-term Automated Quantification of Radiologic Changes in the Characterization of Idiopathic Pulmonary Fibrosis Versus Nonspecific Interstitial Pneumonia and Prediction of Long-term Survival. \u003cem\u003eJournal of thoracic imaging\u003c/em\u003e 2018;33(2):124\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePawar SP, Talbar SN. Two-Stage Hybrid Approach of Deep Learning Networks for Interstitial Lung Disease Classification. \u003cem\u003eBioMed research international\u003c/em\u003e 2022;2022:7340902.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang S, Lee F, Miao R, et al. A deep convolutional neural network architecture for interstitial lung disease pattern classification. \u003cem\u003eMedical \u0026amp; biological engineering \u0026amp; computing\u003c/em\u003e 2020;58(4):725\u0026ndash;37.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWalsh SLF, Calandriello L, Silva M, et al. Deep learning for classifying fibrotic lung disease on high-resolution computed tomography: a case-cohort study. \u003cem\u003eThe Lancet Respiratory medicine\u003c/em\u003e 2018;6(11):837\u0026ndash;45.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChung JH, Chelala L, Pugashetti JV, et al. A Deep Learning-Based Radiomic Classifier for Usual Interstitial Pneumonia. \u003cem\u003eChest\u003c/em\u003e 2024;165(2):371\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBonilla Hern\u0026aacute;n MG, G\u0026oacute;mez-Carrera L, Fern\u0026aacute;ndez-Velilla Pe\u0026ntilde;a M, et al.Prevalence and clinical characteristics of symptomatic diffuse interstitial lung disease in rheumatoid arthritis in a Spanish population. Revista clinica espanola 2022;222(5):281\u0026ndash;287.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen Q, Chen DY, Xu XZ, et al. Platelet/Lymphocyte, Lymphocyte/Monocyte, and Neutrophil/Lymphocyte Ratios as Biomarkers in Patients with Rheumatoid Arthritis and Rheumatoid Arthritis-Associated Interstitial Lung Disease. \u003cem\u003eMedical science monitor: international medical journal of experimental and clinical research\u003c/em\u003e 2019;25:6474\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMcDermott GC, Hayashi K, Juge PA, et al. Impact of Sex, Serostatus, and Smoking on Risk for Rheumatoid Arthritis-Associated Interstitial Lung Disease Subtypes. Arthritis Care \u0026amp; Research, 2024.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Z, Wang W, Xiang T, et al. Serum uric acid as a diagnostic biomarker for rheumatoid arthritis\u0026ndash;associated interstitial lung disease. Inflammation, 2022, 45(4): 1800\u0026ndash;1814.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFurukawa H, Oka S, Higuchi T, et al. Biomarkers for interstitial lung disease and acute-onset diffuse interstitial lung disease in rheumatoid arthritis. \u003cem\u003eTherapeutic advances in musculoskeletal disease\u003c/em\u003e 2021;13:1759720x211022506.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu X, Jeong Y, Poli de Fr\u0026iacute;as S, et al. Serum proteomic profiling of rheumatoid arthritis-interstitial lung disease with a comparison to idiopathic pulmonary fibrosis. \u003cem\u003eThorax\u003c/em\u003e 2022;77(10):1041\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFurukawa H, Oka S, Shimada K, et al. Serum Metabolomic Profiling in Rheumatoid Arthritis Patients With Interstitial Lung Disease: A Case-Control Study. \u003cem\u003eFrontiers in medicine\u003c/em\u003e 2020;7:599794.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang X, Yin M, Zhang D, et al. Metabolomics reveals disturbed amino acid metabolism during different stages of RA in collagen-induced arthritis mice. Inflammation, 2024.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhao YD, Yin L, Archer S, et al. Metabolic heterogeneity of idiopathic pulmonary fibrosis: a metabolomic study. \u003cem\u003eBMJ open respiratory research\u003c/em\u003e 2017;4(1):e000183.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu RM, Vayalil PK, Ballinger C, et al. Transforming growth factor β suppresses glutamate-cysteine ligase gene expression and induces oxidative stress in a lung fibrosis model. \u003cem\u003eFree radical biology \u0026amp; medicine\u003c/em\u003e 2012;53(3):554\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEngland BR, Duryee MJ, Roul P, et al. Malondialdehyde-Acetaldehyde Adducts and Antibody Responses in Rheumatoid Arthritis-Associated Interstitial Lung Disease. Arthritis \u0026amp; rheumatology 2019;71(9):1483\u0026ndash;1493.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLeavy O C, Kawano-Dourado L, Stewart I D, et al. Rheumatoid arthritis and idiopathic pulmonary fibrosis: a bidirectional Mendelian randomisation study[J]. Thorax, 2024, 79(6): 538\u0026ndash;544.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang D, Zhang J, Lau J, et al. Mechanisms of lung disease development in rheumatoid arthritis. Rheumatology 2019;15(10):581\u0026ndash;596.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"european-journal-of-medical-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ejmr","sideBox":"Learn more about [European Journal of Medical Research](http://eurjmedres.biomedcentral.com)","snPcode":"40001","submissionUrl":"https://submission.nature.com/new-submission/40001/3","title":"European Journal of Medical Research","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"radiomics, nomogram, metabolomics, rheumatoid arthritis-associated interstitial lung disease, usual interstitial pneumonia, non-specific interstitial pneumonia","lastPublishedDoi":"10.21203/rs.3.rs-7839620/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7839620/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjectives\u003c/h2\u003e\u003cp\u003eRheumatoid arthritis-associated interstitial lung disease (RA-ILD) is primarily classified into usual interstitial pneumonia (UIP) and non-specific interstitial pneumonia (NSIP) patterns, which exhibit significant differences in prognosis and treatment response. This study aimed to differentiate UIP and NSIP patterns in RA-ILD by integrating HRCT-based radiomics with plasma metabolomics.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eWe included 350 RA-ILD patients, assigned to training, internal validation, and external validation sets. Radiomics features were extracted from HRCT images, and optimal features were selected using variance threshold, univariate selection, and LASSO regression. Three machine learning algorithms\u0026mdash;random forest (RF), logistic regression (LR), and multi-layer perceptron (MLP)\u0026mdash;were each used to construct three categories of models: clinical models, radiomics models, and combined clinical-radiomics models. This design resulted in a total of 9 models (3 algorithms \u0026times; 3 model categories). A nomogram integrating the radiomics score (Rad-score) and clinical variables was developed. Model performance was assessed via AUC, calibration curves, and decision curve analysis (DCA). Plasma metabolomics analysis was performed using liquid chromatography-tandem mass spectrometry, and Spearman correlation was used to explore the relationship between radiomics features and metabolites.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eTwenty-five optimal radiomics features were selected, and 77 metabolites were significantly correlated with radiomics features (R\u0026thinsp;\u0026gt;\u0026thinsp;0.4, P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). L-glutamate and alpha-ketoglutarate were co-enriched in key metabolic pathways (e.g., arginine biosynthesis, D-glutamine/D-glutamate metabolism). Radiomics and combined models outperformed clinical models: the MLP combined model showed more favourable discrimination in the training set (AUC\u0026thinsp;=\u0026thinsp;0.915; 95% CI: 0.884\u0026ndash;0.940) and internal validation set (AUC\u0026thinsp;=\u0026thinsp;0.828; 95% CI: 0.728\u0026ndash;0.916), while the RF combined model showed the best discrimination in the external validation set (AUC\u0026thinsp;=\u0026thinsp;0.841; 95% CI: 0.729\u0026ndash;0.916, specificity\u0026thinsp;=\u0026thinsp;67%). The nomogram exhibited excellent calibration (Hosmer\u0026ndash;Lemeshow: P\u0026thinsp;=\u0026thinsp;0.051, 0.382, 0.073 for training, internal, external sets) and clinical utility (DCA: high net benefits across thresholds).\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003eThe combined clinical-radiomics model provides a promising auxiliary tool for distinguishing subtypes of RA-ILD, which may support early and accurate identification of RA-ILD subtypes to aid clinical decision-making. The Rad-score improves RA-ILD subtype discrimination, and L-glutamate/alpha-ketoglutarate merit further investigation as potential metabolic markers.\u003c/p\u003e","manuscriptTitle":"Combining radiomics based on high-resolution computed tomography with plasma metabolomics for diagnosing subtypes of rheumatoid arthritis-associated interstitial lung disease","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-11 16:18:40","doi":"10.21203/rs.3.rs-7839620/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-01-12T00:34:23+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-27T22:22:23+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-18T03:54:49+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-15T08:56:52+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"16295153463978005478274822777192153051","date":"2025-12-09T10:46:26+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"236037341856679452829387442952778254737","date":"2025-12-07T07:33:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"179176289633780499011079647299493908614","date":"2025-12-07T07:19:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"28882790465867272793871686248204075510","date":"2025-12-07T05:02:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"143263088822041059613504604871930505868","date":"2025-11-26T17:43:46+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-29T10:54:40+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-24T10:11:49+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-23T10:42:51+00:00","index":"","fulltext":""},{"type":"submitted","content":"European Journal of Medical Research","date":"2025-10-21T16:14:14+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"european-journal-of-medical-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ejmr","sideBox":"Learn more about [European Journal of Medical Research](http://eurjmedres.biomedcentral.com)","snPcode":"40001","submissionUrl":"https://submission.nature.com/new-submission/40001/3","title":"European Journal of Medical Research","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5659863b-e5c4-442b-a789-ab3473bab0c3","owner":[],"postedDate":"November 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-04-20T16:05:16+00:00","versionOfRecord":{"articleIdentity":"rs-7839620","link":"https://doi.org/10.1186/s40001-026-04452-3","journal":{"identity":"european-journal-of-medical-research","isVorOnly":false,"title":"European Journal of Medical Research"},"publishedOn":"2026-04-18 15:57:13","publishedOnDateReadable":"April 18th, 2026"},"versionCreatedAt":"2025-11-11 16:18:40","video":"","vorDoi":"10.1186/s40001-026-04452-3","vorDoiUrl":"https://doi.org/10.1186/s40001-026-04452-3","workflowStages":[]},"version":"v1","identity":"rs-7839620","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7839620","identity":"rs-7839620","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}