Research on the construction and application of a multi-feature auxiliary diagnostic model for uremic pneumonia based on radiomics and deep learning: a retrospective bi-centre study

Research on the construction and application of a multi-feature auxiliary diagnostic model for uremic pneumonia based on radiomics and deep learning: a retrospective bi-centre study | 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 Research on the construction and application of a multi-feature auxiliary diagnostic model for uremic pneumonia based on radiomics and deep learning: a retrospective bi-centre study Yanan Wang, Jingfeng Zhang, Qi Dai, Minghui Yan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7079017/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objective This study aimed to develop and validate a diagnostic model that integrates radiomics and deep learning (DL) for the early differentiation of uremic pneumonia (UP) from pulmonary infection (PI) in non-dialysis patients. Methods This study retrospectively analyzed the clinical data and non-contrast CT images of 334 non-dialysis patients with stage 5 chronic kidney disease (CKD). In this study, univariate and multivariate logistic regression analysis methods were employed to screen for clinical risk factors. Radiomics and DL features of lesions were extracted based on non-contrast chest CT images. Clinical model, radiomics model, and DL model were constructed separately, and a combined deep learning-radiomics-clinical (DLRC) model was established using feature fusion methods. The performance of each model was evaluated using the receiver operating characteristic (ROC) curve, the area under the curve (AUC), the calibration curve, and decision curve analysis (DCA). In addition, the generalization ability of each model was validated using an external validation cohort (n = 93). Results The DLRC model demonstrated the highest diagnostic performance, with AUC values of 0.944 (95% CI: 0.911–0.978) in the training set and 0.930 (95% CI: 0.873–0.985) in the test set. The AUC values of each single-feature model in the training set and the test set were as follows: clinical model (0.828 and 0.894), radiomics model (0.862 and 0.813), and DL model (0.881 and 0.855). In the external validation set, the DLRC model continued to show stable diagnostic performance (AUC = 0.885, 95% CI: 0.817–0.952). Conclusion The DLRC model demonstrated outstanding performance in differentiating between UP and PI, which facilitated the early identification of UP and the formulation of individualized diagnosis and treatment plans. Uremic pneumonitis Pulmonary infection Computed tomography Radiomics Deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Uremic pneumonitis (UP) is a secondary non-infectious pulmonary injury resulting from the accumulation of metabolites and fluid retention in end-stage renal disease (ESRD), also known as uremic pulmonary edema. As a common pulmonary complication of uremia and chronic renal failure, UP has an extremely high incidence rate. Related literature reports that its incidence rate can reach 50% [ 1 ] . As one of the most urgent symptoms in patients with chronic kidney disease (CKD), UP has a mortality rate of 6.6% and a five-year survival rate of 21.2% [ 2 ] . In a study involving 36 patients with UP, the misdiagnosis rate reached 58.3% [ 3 ] . Due to the low specificity of UP chest imaging and mild initial symptoms, UP is prone to being misdiagnosed as pulmonary infection (PI). Relevant reports indicate that the incidence rate of PI in this population can reach 19% [ 4 ] . In fact, as the two most common pulmonary diseases among CKD patients, undifferentiated UP and PI share similarities in clinical symptoms and imaging manifestations, yet their treatment approaches differ significantly. Timely initiation of dialysis treatment is the most effective treatment for patients with UP. The "Chinese Expert Consensus on Vascular Access for Hemodialysis (2nd Edition)" states that the success rate of emergency dialysis rescue for UP can reach 95% [ 5 ] . For patients with PI, targeted anti-inflammatory therapy should be implemented based on specific pathogen types. Given that ESRD patients face a significantly increased cardiovascular risk when UP occurs before dialysis, preliminary diagnosis is of great importance for subsequent treatment. Therefore, early and accurate identification of non-dialysis UP patients is crucial for scientifically and efficiently managing this patient group and reducing the risk of adverse cardiovascular events. In recent years, artificial intelligence technologies such as radiomics and deep learning (DL) have been widely applied in various aspects of clinical practice, including auxiliary diagnosis, surgical path planning, lesion segmentation, and measurement. During the COVID-19 pandemic, radiomics and DL models provided technical support for accurate identification and disease monitoring of COVID-19 [ 6 , 7 ] . To address the limitations of manually defined features in traditional radiomics, some scholars have integrated DL with radiomics features, demonstrating excellent differential diagnostic capabilities in previous studies [ 8 , 9 ] . Therefore, integrating these two quantitative techniques into the differential diagnosis of UP is feasible and has clinical value. This study aimed to develop a scalable auxiliary diagnostic model that integrates radiomics and DL. The model incorporated clinical manifestations, laboratory examination results, and imaging data of UP to provide a scientific basis for the early identification and intervention of UP in clinical practice. Materials and methods Study participants This study strictly adhered to the ethical principles outlined in the Declaration of Helsinki and obtained formal approval from the Ethics Committee of the Ningbo Yinzhou NO.2 Hospital (registration number: 2025-007). In accordance with relevant regulations, the requirement for research participants to sign informed consent forms was waived. A total of 1,373 patients diagnosed with CKD5 who received medical treatment at two hospitals from January 2019 to December 2023 were initially recruited. After applying the inclusion and exclusion criteria, 334 patients were ultimately included in this study. The development cohort (n = 241) was randomly divided into a training set (n = 169) and a test set (n = 72) at a ratio of 7:3. All 93 patients in the external cohort were incorporated into the external validation set. The detailed process of case screening and grouping is illustrated in Fig. 1 . It is worth noting that in the PI group (n = 150) of the development cohort, the distribution of infection types was as follows: 41 cases of bacterial infection, 85 cases of viral infection, 15 cases of fungal infection, 3 cases of mycoplasma infection, and 1 case of chlamydia infection. In the PI group of the external cohort (n = 56), there were 16 cases of bacterial infection, 32 cases of viral infection, 7 cases of fungal infection, and 1 case of mycoplasma infection. The inclusion criteria were as follows: (1) Patients' renal function had progressed to CKD5, which was defined as an estimated glomerular filtration rate (eGFR) of less than 15 mL/(min·1.73 m²). (2) Patients who had been diagnosed with newly-developed exudative pulmonary lesions confirmed by standard chest CT, and UP had been confirmed through pathological examination and clinical assessment. (3) Patients had been diagnosed with PI based on respiratory symptoms and pathogen detection results (e.g., positive culture results from sputum, blood, pleural effusion, or bronchoalveolar lavage fluid). The exclusion criteria were as follows: (1) Patients who were undergoing hemodialysis or peritoneal dialysis. (2) Patients with a history of tuberculosis, Goodpasture's syndrome, pulmonary embolism, pulmonary tumors, pulmonary trauma, pulmonary surgery, cardiogenic diseases, etc. (3) Patients with both UP and PI. (4) Patients with poor-quality imaging and incomplete clinical data. The baseline clinical data included variables such as gender, age, chest tightness, edema, cough, expectoration, body temperature (T), systolic blood pressure (SBP), diastolic blood pressure (DBP), C-reactive protein level (CRP), white blood cell count (WBC), hemoglobin (Hb), serum creatinine level (SCr), procalcitonin (PCT), N-terminal pro-B-type natriuretic peptide (NT-proBNP), arterial blood pH value (pH), and blood oxygen saturation (SpO₂). CT imaging and image segmentation The CT scans in this study were performed using Kangda CT (China), GE optima 620 (USA), and Siemens Somatom go.Top (Germany) scanners. The CT scanning parameters were set as follows: a tube voltage of 120 kV, a tube current of 150–300 mA, a layer thickness of 1-1.25 mm, and a matrix size of 512 × 512. This study utilized the open-source software 3D Slicer (version 5.2.2, https://www.slicer.org/ ) to carry out semi-automatic image segmentation. The image analysis was conducted by two radiologists with extensive experience in chest imaging diagnosis (5 years and 7 years, respectively). They conducted the image analysis without knowledge of the experimental results. Firstly, the CT images reconstructed by maximum intensity projection (MIP) were resampled to achieve voxel sizes of 1 mm × 1 mm × 1 mm and then imported into the 3D Slicer software (Fig. 2 a). Subsequently, a junior physician utilized the "Interactive Lobe Segmentation" plugin to perform semi-automatic image segmentation based on thresholds to determine the approximate extent of the lungs. After that, the doctor manually adjusted to accurately depict the region of interest (ROI) encompassing the entire lung tissue (Fig. 2 b-c). Ultimately, the volume of the 3D region of interest (VOI) was calculated (Fig. 2 d). To ensure the accuracy and reliability of the ROI, a senior physician reviewed and modified the delineation results. If there were significant differences in the delineated lesion areas, the physician redefined the boundaries and made corresponding adjustments. Extraction of radiomics and deep learning features Using the 3D Slicer software, a total of 851 radiomics features were extracted from each ROI. These features included first-order features, 2D shape features, texture features, and transformed features. By cropping and normalizing the ROI into a 224 × 224 pixel 3D image block, DL features were calculated and then input into the model. Transfer learning techniques were applied to compensate for insufficient data, and a moderately sized ResNet34 architecture was used to extract DL features, yielding a total of 512 features. The ResNet34 model comprised 34 learnable layers, and its specific architecture was as follows: (1) An initial feature extraction layer with 64 channels and a 7 × 7 convolutional kernel. (2) A 3 × 3 max-pooling layer (with a stride of 2) that reduced the dimensionality of the feature map to 56 × 56. (3) Residual blocks with skip connections were utilized to gradually extract deep-level features and compress the feature map to 7 × 7; (4) A 512-channel feature map was processed by average pooling to output a 512-dimensional feature vector [ 10 ] (Fig. 3 ). To ensure the reproducibility of feature extraction, the two radiologists previously mentioned conducted reliability assessments on a randomly selected sample of 30 cases. For the intra-rater reliability assessment, the first radiologist obtained radiomics and DL features following specific procedures. The same methodology and steps were repeated two weeks later. Then, intra-rater correlation analyses were performed on the features extracted from these two sets. As for the inter-rater reliability assessment, the second radiologist employed the same methodologies and procedures as the first one. Subsequently, the features obtained by the second radiologist were compared and analyzed with those extracted by the first radiologist. The results demonstrated that both the intra-rater and inter-rater correlation coefficients exceeded 0.75 ( P < 0.05), indicating a high level of consistency in the extracted features. Selection of radiomics and deep learning features Firstly, the minimum redundancy maximum relevance (mRMR) algorithm was utilized to eliminate redundant and irrelevant features. After that, the least absolute shrinkage and selection operator (LASSO) algorithm was employed to pick out the most predictive features. In the process of LASSO regression, the regularization parameter (λ) was optimized through ten-fold cross-validation to achieve precise feature selection. Finally, the selected features were combined linearly with their respective coefficients obtained from the LASSO regression. This linear combination led to the calculation of the radiomics score (Rad-score) and the DL score (Deepscore). Construction and evaluation of models In this study, logistic regression analysis was employed to construct the following four models. These models were constructed based on clinical risk factors, the Rad-score, and the Deepscore: a clinical model, a radiomics model, a DL model, and a combined deep learning-radiomics-clinical (DLRC) model. A multi-dimensional evaluation framework was adopted to assess the performance of these models: (1) The ability of the four models to distinguish between UP and PI was quantified using the receiver operating characteristic (ROC) curve and its area under the curve (AUC) value. Inter-group comparisons were conducted using the DeLong test. (2) The generalization performance was evaluated using an external validation dataset. (3) A comprehensive evaluation was performed by integrating multiple metrics, including sensitivity, specificity, and accuracy. (4) The calibration of the models was validated by plotting calibration curves. (5) Decision curve analysis (DCA) was used to quantify the clinical utility of the four models. (6) The final results were presented visually through nomogram. Statistical analysis In this study, statistical analysis was conducted using the R language (version 4.3.2, https://www.r-project.org/ ). For categorical variables like gender and clinical symptoms, either the chi-square test or Fisher's exact test was employed. For continuous variables like age and laboratory indicators, once normality testing had been conducted, the t-test was applied to variables that were found to conform to a normal distribution. Meanwhile, non-parametric tests were used for those variables that did not meet the normality criterion. The results were presented as mean ± standard deviation or median (interquartile range). In-depth analyses were conducted using specialized packages within the R software. Feature selection was performed through the mRMR algorithm using the "survcomp" package. LASSO regression was conducted by turning to the "glmnet" package. Both nomograms and calibration curves, which were crucial for visualizing and validating the models, were constructed by applying the "rms" package. DCA was carried out by utilizing the "rmda" package. The ROC curves were plotted using the "pROC" package. The significance level was set at P < 0.05. Results Clinical characteristics The characteristics of patients in the training set, the test set, and the external validation set are presented in Table 1 . Univariate and multivariate logistic regression analyses were carried out on the basic information, clinical symptoms, and laboratory examination results of the patients in the training set. The analysis revealed significant differences in SBP, NT-proBNP, Hb, and CRP between the two groups ( P < 0.05, Table 2 ). Table 1 Clinical laboratory results and statistical outcomes of the patients Characteristic Training set (n = 169) P Testing set (n = 72) P External validation set (n = 93) P UP (n = 63) PI (n = 106) UP (n = 27) PI (n = 45) UP (n = 37) PI (n = 56) Gender, n (%) 0.870 0.205 0.668 Male 43.0 (68.3) 68.0 (64.2) 15.0 (55.6) 16.0 (35.6) 20.0 (54.1) 34.0 (60.7) Female 23.0 (31.7) 38.0 (35.8) 12.0 (44.4) 29.0 (64.4) 17.0 (45.9) 22.0 (39.3) Age, M(Q₃-Q₁) 60.5 (24) 64.0 (20.0) 0.070 66.0 (28.8) 65.0 (18.0) 0.492 61.5 (17.8) 72.0 (20.3) 0.001 Edema, n (%) 0.426 0.795 0.119 No 33.0 (52.3) 63.0(59.4) 18.0 (66.7) 32.0 (71.1) 21.0 (56.8) 41.0 (73.2) Yes 30.0 (47.7) 43.0(40.6) 9.0 (33.3) 13.0 (28.9) 16.0 (43.2) 15.0 (26.8) Chest tightness, n (%) 0.207 0.805 0.396 No 25.0 (39.7) 55.0 (51.9) 17.0 (60.7) 26.0 (57.8) 19.0 (51.4) 23.0 (41.1) Yes 38.0 (60.3) 51.0 (48.1) 10.0 (39.3) 19.0 (42.2) 18.0 (48.6) 33.0 (58.9) Cough, n (%) 0.265 0.316 0.026 No 21.0 (33.3) 47.0 (44.3) 17.0 (60.7) 19.0 (42.2) 17.0 (45.9) 13.0 (23.2) Yes 42.0 (66.7) 59.0 (55.7) 11.0 (39.3) 26.0 (57.8) 20.0 (54.1) 43.0 (76.8) Expectoration, n (%) 0.431 0.803 0.003 No 37.0 (58.7) 54.0 (50.9) 16.0 (59.3) 28.0 (62.2) 23.0 (62.2) 17.0 (30.3) Yes 27.0 (41.3) 52.0 (49.1) 11.0 (40.7) 17.0 (37.8) 14.0 (37.8) 39.0 (69.7) T, M (Q₃-Q₁) a 36.6 (0.4) 36.8 (0.7) 0.009 36.7 (0.5) 36.8 (0.7) 0.545 36.5 (0.5) 36.8 (0.7) 0.005 SBP, Mean ± SD 158.2 ± 19.7 140.6 ± 23.2 0.001 157.7 ± 21.3 140.4 ± 23.8 0.004 156.3 ± 27.5 143.6 ± 24.0 0.022 DBP, M(Q₃-Q₁) 85.0 (17.0) 80.0 (13.0) 0.017 86.5 (25.8) 80.0 (18.8) 0.040 82.0 (16.25) 72.0 (20.5) 0.005 CRP, M(Q₃-Q₁) 3.6 (8.9) 10.9 (39.7) 0.001 2.2 (3.6) 11.4 (33.1) 0.001 9.7 (18.5) 71.9 (87.5) 0.001 WBC, M(Q₃-Q₁) 6.2 (3.9) 6.8 (4.2) 0.280 5.9 (2.1) 7.0 (4.0) 0.017 6.5 (3.2) 7.4 (5.2) 0.289 Hb, M(Q₃-Q₁) 88.3 ± 22.2 96.9 ± 18.8 0.008 80.5 ± 28.5 106.0 ± 31.5 0.001 83.5 ± 26.8 85.5 ± 26.0 0.287 SCr, M(Q₃-Q₁) 631.0 (378.0) 630.0 (403.0) 0.805 591.5 (536.0) 649.0 (369.8) 0.455 639.0 (331.5) 499.5 (303.8) 0.068 NT-proBNP, M(Q₃-Q₁) 18750.0 (20252.5) 7090.0 (14670.0) 0.001 20720.0 (27157.5) 7205.0 (18989.5) 0.004 35000 (18610.0) 9638.5 (19384.8) 0.001 PCT, M(Q₃-Q₁) 0.1 (0.1) 0.2 (0.5) 0.335 0.0 (0.0) 0.1 (0.5) 0.002 0.5 (0.2) 0.5 (1.1) 0.133 PH, M(Q₃-Q₁) 7.4 (0.1) 7.4 (0.1) 0.108 7.4 (0.1) 7.4 (0.1) 0.117 7.4 (0.1) 7.4 (0.1) 0.295 SpO2, M(Q₃-Q₁) 95.9 (4.8) 97.5 (2.8) 0.789 96.9 (3.8) 97.5 (3.6) 0.962 94.9 (5.7) 94.2 (9.1) 0.689 a M: Median, Q₁: 1st Quartile, Q₃: 3st Quartile Table 2 Results of multivariate logistic regression analysis Variable OR 95% CI P SBP 1.032 1.011–1.053 0.003 CRP 0.973 0.955–0.991 0.003 Hb 0.978 0.958–0.998 0.033 NT-proBNP 1.000 1.000–1.000 0.001 Development of radiomics and deep learning models A total of 851 radiomics features and 512 DL features were obtained in this study. By applying the mRMR algorithm and LASSO regression, 7 radiomics features and 9 DL features were ultimately selected (Fig. 4 ). The Rad-score was constructed using the 7 selected features and their corresponding regression coefficients. The specific formula is as follows: Rad-score = 1.009 × 10Percentile + (-0.35) × High Gray Level Zone Emphasis + (-0.383) × Flatness + (-0.39) × Large Area Low Gray Level Emphasis + 0.46 × Major Axis Length + (-0.066) × Large Dependence High Gray Level Emphasis + (-0.252) × Maximum2D Diameter Row + (-0.76) The Deepscore was constructed using the 9 selected features and their corresponding regression coefficients. The specific formula is as follows: Deepscore = (-0.5018) × v_394 + (-0.736) × v_273 + (-0.585) × v_255 + 0.521 × v_357 + (-0.195) × v_17 + 0.121 × v_201 + (-0.2) × v_187 + (-0.263) × v_232 + (-0.209) × v_10 + (-0.986) In both the training set (Fig. 5 a, c) and the test set (Fig. 5 b, d), the Rad-score and Deepscore of the UP group were significantly higher than those of the PI group ( P < 0.001). The AUC values for these scores are as follows: 0.862 (95% CI: 0.808–0.917) and 0.881 (95% CI: 0.827–0.935) for the training set; 0.813 (95% CI: 0.705–0.921) and 0.855 (95% CI: 0.760–0.950) for the test set; and 0.828 (95% CI: 0.739–0.917) and 0.78 (95% CI: 0.675–0.886) for the external validation set, respectively. Evaluation and validation of models The ability of the four models to distinguish between UP and PI was quantitatively evaluated using the ROC curves and the AUC values from the training set (Fig. 6 a), the test set (Fig. 6 b), and the external validation set (Fig. 6 c), respectively. The accuracy (ACC), sensitivity (SEN), specificity (SPE), positive predictive value (PPV), and negative predictive value (NPV) of each model were calculated according to the confusion matrix (Table 3 ). The DeLong test results revealed statistically significant differences ( P < 0.05) in the AUC values between the combined DLRC model and each single-feature model. However, in the external validation set, there were no statistically significant differences in the ROC curves between the clinical model and the radiomics model ( P = 0.763). Evidently, the combined DLRC model demonstrated robust discriminative ability, with AUC values of 0.944 (95% CI: 0.911–0.978), 0.930 (95% CI: 0.873–0.986), and 0.885 (95% CI: 0.817–0.952) in the training set, the test set, and the external validation set, respectively. Their sensitivities were 0.894, 0.833, and 0.892, respectively; specificities‌ were 0.883, 0.896, and 0.751, respectively; and accuracies were 0.888, 0.875, and 0.806, respectively. DCA (Fig. 6 d) indicated that the DLRC model showed better clinical utility when the risk threshold was below 90%. Additionally, the calibration curve of the combined DLRC model (Fig. 6 e) indicated good consistency between the predicted probabilities and the actual observed probabilities ( P > 0.05). By constructing a deep learning radiomics clinical nomogram (DLRCN), complex regression equations were converted into visual graphs to more intuitively present the diagnostic efficacy of the DLRC model for UP (Fig. 7 ). Table 3 Performance of the four models in the training set, the test set, and the external validation set Model Cohort AUC 95%CI ACC SEN SPE PPV NPV Clinical Training 0.828 0.766–0.891 0.787 0.667 0.864 0.759 0.802 Test 0.894 0.821–0.967 0.833 0.833 0.833 0.714 0.909 External validation 0.809 0.716–0.902 0.796 0.811 0.786 0.714 0.863 Radiomics Training 0.862 0.808–0.917 0.798 0.788 0.806 0.722 0.856 Test 0.813 0.705–0.921 0.722 0.875 0.646 0.553 0.912 External validation 0.828 0.739–0.917 0.774 0.784 0.768 0.690 0.843 Deep learning Training 0.881 0.827–0.935 0.828 0.803 0.845 0.768 0.870 Test 0.855 0.760–0.950 0.806 0.792 0.813 0.679 0.886 External validation 0.780 0.675–0.886 0.774 0.703 0.821 0.722 0.807 DLRC Training 0.944 0.911–0.978 0.888 0.894 0.883 0.831 0.929 Test 0.930 0.873–0.988 0.875 0.833 0.896 0.800 0.915 External validation 0.885 0.817–0.952 0.806 0.892 0.751 0.702 0.913 Discussion Through a comprehensive systematic review of relevant domestic and international literature, this study is the first to integrate radiomics with DL for the diagnosis of UP, which showcases notable methodological innovation. Previous studies have utilized DL to differentiate generalized pulmonary edema from various types of pneumonia. Velichko E et al. [ 11 ] developed an auxiliary model named EDECOVID-net. This was the first method designed to distinguish between pulmonary edema and COVID-19, and it achieved an outstanding accuracy rate of 0.98. Similarly, Rachel et al. [ 12 ] integrated the squirrel search algorithm (SSA) with the backpropagation neural network (BPNN) for the differential diagnosis of pulmonary edema and COVID-19, and the study also exhibited excellent performance, attaining an accuracy rate of 0.88. However, these studies only used imaging data, did not incorporate other clinical factors, and failed to integrate radiomics and DL quantitative techniques. In contrast, this study established a combined DLRC model by integrating clinical factors, Rad-score, and Deepscore, exhibiting excellent diagnostic performance. In the training set, the AUC of the model for predicting UP reached 0.944, which was significantly higher than that of any single-feature model ( P < 0.05). In the external validation set, when compared with other models in the same group, the DLRC model (AUC = 0.885) exhibited superior discriminative ability, indicating that the combined model possessed good generalization ability and stability. In terms of accuracy, sensitivity, and specificity, the DLRC model demonstrated outstanding performance, guaranteeing high diagnostic accuracy and sensitivity for identifying UP cases, as well as robust discriminatory ability for non-target cases. In prior research on solid tumors, the DLRC model has been validated to yield comparable outcomes [ 13 ] . The current experimental findings offer crucial evidence-based support for the potential future application of this model in the field of diffuse lesions. To date, research on applying radiomics technology to address the differential diagnosis between UP and PI remains scarce. In recent years, the application of radiomics technology in the realm of pulmonary infectious diseases has become increasingly prevalent. Particularly after the onset of the COVID-19, radiomics has offered robust support for clinical decision-making and demonstrated its feasibility in identifying diffuse pulmonary lesions. This study extracted a total of 7 significant radiomics features, including 1 first-order feature, 3 shape features, and 3 texture features. The first-order feature, serving as the foundational level in radiomics analysis, is a feature value directly calculated based on the pixel intensity distribution of the original image [ 14 ] . The first-order feature "10Percentile" represents the gray-level value of the low-intensity 10th percentile within the ROI. The gray-level value in medical imaging is closely correlated with the composition of pulmonary exudate. In cases of UP, due to elevated pressure, pulmonary tissue fluid leaks out and accumulates in the alveoli and interstitium. When PI occurs, the exudate comprises epithelial cells, purulent exudate, inflammatory cells, and its composition is relatively complex. The difference in the composition of pulmonary exudate between UP and PI results in alterations in gray-level values, which is consistent with the pathological manifestations. Shape features are employed to quantitatively describe the geometric properties of the ROI, primarily representing morphological details such as the shape, size, and volume of lesions. These features assist in evaluating the growth rate and spatial distribution of lesions [ 15 ] . In this study, the values of "Flatness" and "Maximum 2D Diameter Row" in the UP group were both negative, indicating that the morphology of UP lesions was relatively flat. This finding was associated with the relatively limited extent of pulmonary involvement. It was hypothesized that this phenomenon might be related to the relative reduction in lung volume resulting from decreased lung function and alveolar collapse. As the third category of radiomics features in the results of this study, texture features reflect the microstructure and functional status of tissues by analyzing the 3D spatial arrangement of pixels, such as contrast, coarseness, and homogeneity [ 16 ] . In this study, the features of "High Gray Level Zone Emphasis" and "Large Area Low Gray Level Emphasis" are incorporated into the Gray-Level Size Zone Matrix (GLSZM), while "Large Dependence High Gray Level Emphasis" belongs to the Gray-Level Dependence Matrix (GLDM). The GLSZM primarily describes the characteristics of homogeneous regions [ 17 ] . As a variant of GLSZM, the GLDM focuses on describing the gray-level dependencies within lesion regions [ 18 ] . This category of radiomics features primarily reflects the internal heterogeneity of lesions. The higher the value, the coarser the image texture, and the more pronounced the 3D characteristics of the corresponding matrix, suggesting the presence of extensive and high-density lesion areas. As stated by Scholar Shiri [ 15 ] , when the lungs are infected to varying degrees, highly heterogeneous textures are generated, and the characteristic values of gray-level non-uniformity will increase. As of now, numerous studies have utilized radiomics analysis of CT texture features to achieve auxiliary diagnosis and prediction of pulmonary edema, attaining a high level of diagnostic performance [ 19 , 20 ] . Notably, the findings of this experiment were consistent with previous research findings. The model constructed based on the aforementioned radiomics features demonstrated good performance in distinguishing UP and PI. Although there was no statistically significant difference in the AUC between the radiomics model and the clinical-only model in the external validation set, this might be attributed to an insufficient sample size. DCA demonstrated that within the entire threshold probability interval, the radiomics model offered greater net clinical benefits compared to the clinical-only model, further substantiating the value of radiomics technology. In this study, we utilized a pre-trained 3D-ResNet34 network to extract DL features. This model employed a deep transfer learning approach to address the issue of limited dataset size, endowing the model with better generalization ability and replicability [ 21 ] . Multivariate analysis results revealed that Deepscore was a key factor associated with the occurrence of UP. The DL model demonstrated high diagnostic performance on the training set, with sensitivity, specificity, and accuracy reaching 0.803, 0.845, and 0.828, respectively. The corresponding AUC value was 0.881, indicating strong discriminatory ability. Compared with the clinical-only model, the DL model was able to uncover information invisible to the naked eye, thereby exhibiting higher diagnostic efficiency and clinical practicality. However, in the external validation set, the performance of the DL prediction model was suboptimal, which might be due to its heavy reliance on large datasets, weakening its diagnostic performance with small sample sizes [ 22 ] . In addition, compared to radiomics features, the 9 DL features obtained in this experiment exhibited poorer interpretability. The "black box" effect is one of the primary reasons why DL has not yet been widely adopted on a large scale. To address this issue, researchers have proposed utilizing visualization tools such as attention mechanisms and heatmaps for interpretation, or employing interpretability algorithm models like LIME or SHAP to tackle the aforementioned challenges [ 23 , 24 ] . Consequently, it is feasible to apply DL technology to the diagnosis of UP. The innovation of this study lay in the integration of DL with traditional radiomics features. This integration not only expanded the model's feature set but also ensured the reliability of the data input dimension. In previous studies, radiomics and DL were frequently considered as two independent techniques for constructing models. However, the features extracted by traditional radiomics methods are manually defined, suggesting that radiomics features may not encompass important characteristics that humans have not yet discovered. Numerous studies have demonstrated that DL can achieve autonomous learning from data, thereby overcoming the limitations of manual definition and enhancing the accuracy of image analysis. The DL-based radiomics model (DLR) has achieved remarkable results in disease classification, differential diagnosis, and prognosis evaluation [ 25 , 26 , 27 ] . Therefore, DL serves as a powerful complement to traditional radiomics technology. The synergistic integration of the two has enhanced the model's efficacy. Compared with classification models constructed using single-source features, combined models yield higher overall net benefits. The DLRC model comprised four clinical variables: SBP, CRP, Hb, and NT-proBNP. Among these variables, SBP and CRP had relatively high weights in the nomogram. Elevated SBP has been identified as a major risk factor for CKD-related deaths globally. There is a strong correlation between increased blood pressure and the risks of CKD and ESRD [ 28 ] . An intervention trial targeting SBP in CKD patients has confirmed that intensive antihypertensive treatment can effectively reduce the incidence of cardiovascular and cerebrovascular events. Another 7-year prospective cohort study demonstrated that patients with controlled blood pressure (SBP < 130 mmHg) had a 64.2% lower risk of ESRD and a 30.4% lower all-cause mortality rate than patients with blood pressure below the target level [ 29 ] . Therefore, early and sustained blood pressure management is crucial for achieving triple benefits for the heart, lungs, and kidneys. CRP, an acute-phase protein synthesized by the liver, reflects the body's response to tissue injury through its release. CRP levels can rise under various pathological conditions, such as infections, autoimmune diseases, neurodegenerative diseases, and malignant tumors. During a systemic infection, CRP levels can soar to 1000 times the normal value. Related research has indicated that the diagnostic accuracy for community-acquired pneumonia peaks when the CRP cutoff value exceeds 20 mg/L [ 30 ] . Conversely, during the progression of CKD, there exists a synergistic interaction between the activation of oxidative stress responses and inflammatory reactions. Elevated CRP levels are associated with an increased risk of CKD. Following cardiovascular events, it has been confirmed that CRP levels deviate by approximately 25% from the baseline value [ 31 ] . However, at this time, the magnitude of CRP elevation is relatively low compared to the changes induced by infection. Therefore, as an important diagnostic biomarker, the clinical value of CRP largely depends on the specific clinical settings. Previous researchers [ 32 ] have posited a significant association between the radiographic manifestations of UP conditions and the levels of SCr and blood urea nitrogen (BUN). Higher levels of SCr and BUN corresponded to more severe chest pathological changes, and vice versa. However, Hemyari et al. [ 33 ] argued that there was no significant correlation between the severity of SCr and BUN retention and the imaging manifestations of UP. In this study, there was no statistically significant difference in SCr levels between UP and PI. Therefore, it is currently not feasible to confirm that SCr is an independent risk factor for the occurrence of UP. The specific mechanisms by which various toxins trigger the onset and progression of UP, along with the intrinsic correlations between toxin levels and UP imaging manifestations, require further clinical investigations. The limitations of this study can be summarized as follows: (1) Currently, there is still a lack of a "gold standard" for diagnosing UP, which may affect the accuracy of the experiment. Therefore, it is essential for nephrologists and radiologists to perform a joint assessment based on disease progression, make evaluations, and offer a final diagnosis based on the actual clinical treatment response. (2) Although this study was conducted as a dual-center study, challenges in case collection, a relatively limited sample size, and differences in CT scanner parameters and laboratory testing standards between the two hospitals may have affected the models' performance. Additionally, the feasibility of integrating multiple types of information in practical clinical settings remains uncertain. Subsequent research will pursue multi-center collaborations to analyze larger datasets, aiming to improve the models' effectiveness and generalizability. (3) Considering the retrospective nature of this study, its conclusions are inevitably influenced by the inherent selection bias in historical datasets. It is essential to conduct prospective research in the future, employing a systematic and rigorous study design to collect new data for validating and supplementing existing research findings. In summary, an integrated model we proposed demonstrates strong potential as a highly valuable auxiliary tool for clinicians in the early diagnosis of UP among non-dialysis patients with CKD5. Conclusion In this study, clinical factors, radiomics features, and DL features were integrated to develop a combined model. The DLRC model achieved the early non-invasive diagnosis of UP, providing an objective basis for individualized treatment decisions. Meanwhile, it has substantial reference value for optimizing the management of complications associated with chronic kidney disease. Abbreviations UP Uremic pneumonia PI Pulmonary infection CKD Chronic kidney disease DL Deep learning DLRC Deep learning-radiomics-clinical model Declarations Data availability Availability of data and materials: Due to privacy considerations, the datasets generated during the current study are not publicly available. However, data may be available from the corresponding author (E-mail: [email protected] ) upon reasonable request. Acknowledgements This work was supported by the radiology departments of Ningbo Yinzhou NO.2 Hospital and Ningbo No.2 Hospital, China, which provided comprehensive help. In addition, we are very grateful to my colleagues and friends for their contributions to this paper. Funding This research received no external funding. Author information Authors and Affiliations Department of Radiology, Ningbo Yinzhou No.2 Hospital, Ningbo 315192, Zhejiang, China Yanan Wang & Minghui Yan Department of Radiology, Ningbo No.2 Hospital, Ningbo 315010, Zhejiang, China Jingfeng Zhang & Qi Dai Contributions All authors contributed to the study conception and design. Material preparation and data collection were performed by Yanan Wang, Minghui Yan, and analysis were performed by Jingfeng Zhang and Qi Dai. The first draft of the manuscript was written by Yanan Wang and all authors commented on previous versions of the manuscript. Corresponding author Correspondence to Minghui Yan. Ethics approval and consent to participate This retrospective study was approved by the Medical Ethics Committee of Ningbo Yinzhou NO.2 Hospital (registration number 2025–007, approved on 18 April 2025), and written informed consent was waived. Because the medical records used in this study were obtained from past clinical diagnosis and treatment, this clinical study did not directly involve the subjects, and the results were not used for subject diagnosis. Therefore, it will not have any adverse effects on the subjects. The privacy and personal identity information of the subjects are protected. Furthermore, the study adheres to the Helsinki Declaration. Consent for publication Not applicable. Competing interests The authors declare no competing interests. References Shaik L, Thotamgari SR, Kowtha P, Ranjha S, Shah RN, et al. A Spectrum of pulmonary complications occurring in end-stage renal disease patients on maintenance hemodialysis. Cureus. 2021;13:e15426. 10.7759/cureus.15426 . Banerjee D, Ma JZ, Collins AJ, Herzog CA. Long-term survival of incident hemodialysis patients who are hospitalized for congestive heart failure, pulmonary edema, or fluid overload. Clin J Am Soc Nephrol. 2007;2:1186–90. 10.2215/CJN.01110307 . Yi J, Liu Y, Hu Z. Relation between changes in chest imaging pathology and levels of SCr and BUN in uremic patients. Med J Natl Def Forces Southwest China. 2016;26:863–5. A. G A MGT. Microbiological assay of respiratory infections in kidney diseases: a retrospective study in 2019. Egypt J Chest Dis Tuberc. 2025;74:214–20. 10.4103/ecdt.ecdt_116_22 . Jin Q, Wang Y, Ye C et al. Consensus among experts on blood access used for hemodialyis in China (The 2nd edition). CJBP. (2019), 18: 365–381. Ammirabile A, Cavinato L, Ferro CAP, Fiz F, Savino MS, et al. CT-radiomics and pathological tumor response to systemic therapy: A predictive analysis for colorectal liver metastases. Development and internal validation of a clinical-radiomic model. Eur J Surg Oncol. 2025;51:109557. 10.1016/j.ejso.2024.109557 . Fang X, Li X, Bian Y, Ji X, Lu J. Radiomics nomogram for the prediction of 2019 novel coronavirus pneumonia caused by SARS-CoV-2. Eur Radiol. 2020;30:6888–901. 10.1007/s00330-020-07032-z . Su W, Cheng D, Ni W, Ai Y, Yu X, et al. Multi-omics deep learning for radiation pneumonitis prediction in lung cancer patients underwent volumetric modulated arc therapy. Comput Methods Programs Biomed. 2024;254. 10.1016/j.cmpb.2024.108295 . Ma G, Wang K, Zeng T, Sun B, Yang L. A Joint Classification Method for COVID-19 Lesions Based on Deep Learning and Radiomics. Tomography. 2024;10:1488–500. 10.3390/tomography10090109 . Tian S, Yu Y, Shi K, Jiang Y, Song H, et al. Deep learning radiomics based on ultrasound images for the assisted diagnosis of chronic kidney disease. Nephrol (Carlton). 2024;29:748–57. 10.1111/nep.14376 . Velichko E, Shariaty F, Orooji M, Pavlov V, Pervunina T, Zavjalov S, Khazaei R, et al. Development of computer-aided model to differentiate COVID-19 from pulmonary edema in lung CT scan: EDECOVID-net. Comput Biol Med. 2022;141:105172. 10.1016/j.compbiomed.2021.105172 . Betshrine RR, Nehemiah KH, Marishanjunath CS, et al. Diagnosis of pulmonary edema and Covid-19 from CT slices using squirrel search algorithm, support vector machine and back propagation neural network. JIFS. 2023;44:5633–46. 10.3233/JIFS-222564 . Liu F, Zhang BD, Cheng HS, et al. A radiomics and deep learning nomogram developed and validated for predicting no-collapse survival in patients with osteonecrosis after multiple drilling. BMC Med Inf Decis Mak. 2025;25:26. 10.1186/s12911-025-02859-2 . Wen Z, Gao X, Wu Q, Yang J, Sun J, et al. Baseline [18F] FDG PET/CT radiomics for predicting interim efficacy in follicular lymphoma treated with first-line R-CHOP. BMC Cancer. 2025;25:128. 10.1186/s12885-025-13507-3 . Wu Y, Cao F, Lei H, Zhang S, et al. Interpretable multiphasic CT-based radiomic analysis for preoperatively differentiating benign and malignant solid renal tumors: a multicenter study. Abdom Radiol (NY). 2024;49:3096–106. 10.1007/s00261-024-04351-3 . Piccolo CL, Sarli M, Pileri M, Tommasiello M, Rofena A, et al. Radiomics for Predicting Prognostic Factors in Breast Cancer: Insights from Contrast-Enhanced Mammography (CEM). J Clin Med. 2024;13:6486. 10.3390/jcm13216486 . Wang X, Luo X, Pan H, Wang X, Xu S, Li H, Lin Z. Performance of hippocampal radiomics models based on T2-FLAIR images in mesial temporal lobe epilepsy with hippocampal sclerosis. Eur J Radiol. 2023;167:111082. 10.1016/j.ejrad.2023.111082 . Pisani N, Abate F, Avallone AR, Barone P, Cesarelli M, et al. A radiomics approach to distinguish progressive supranuclear palsy richardson's syndrome from other phenotypes starting from MR images. Comput Methods Programs Biomed. 2025;266:108778. 10.1016/j.cmpb.2025.108778 . Tian X, MA A, Jiang L, et al. CT imaging-based on texture analysis: discrimination of high altitude pulmonary edema and acute cardiogenic pulmonary edema. Radiol Pract. 2020;35(01):45–9. 10.13609/j.cnki.1000-0313.2020.01.009 . Brusasco C, Santori G, Tavazzi G, et al. Second-order grey-scale texture analysis of pleural ultrasound images to differentiate acute respiratory distress syndrome and cardiogenic pulmonary edema. J Clin Monit Comput. 2020;36:1–10. 10.1007/s10877-020-00629-1 . Nijiati M, Tuerdi M, Damola M, Yimit Y, Yang J, et al. A deep learning radiomics model based on CT images for predicting the biological activity of hepatic cystic echinococcosis. Front Physiol. 2024;15:1426468. 10.3389/fphys.2024.1426468 . Xie K, Jiang H, Chen X, Ning Y, Yu Q, et al. Multiparameter MRI-based model integrating radiomics and deep learning for preoperative staging of laryngeal squamous cell carcinoma. Sci Rep. 2025;15:16239. 10.1038/s41598-025-01270-1 . Ganji Z, Nikparast F, Shoeibi N, Shoeibi A, Zare H. Decoding Parkinson's Diagnosis: An OCT-Based Explainable AI with SHAP/LIME Transparency from the Persian Cohort Study. Photodiagnosis Photodyn Ther. 2025;14:104668. 10.1016/j.pdpdt.2025.104668 . Rathore PS, Kumar A, Nandal A, Dhaka A, Sharma AK. A feature explainability-based deep learning technique for diabetic foot ulcer identification. Sci Rep. 2025;15:6758. 10.1038/s41598-025-90780-z . Chen J, Liu S, Lin Y, Hu W, Shi H, Liao N, et al. The Quality and Accuracy of Radiomics Model in Diagnosing Osteoporosis: A Systematic Review and Meta-analysis. Acad Radiol. 2025;32:2863–75. 10.1016/j.acra.2024.11.065 . Yuan Z, Handcrafted Radiomics. Deep learning radiomics in the prediction of radiation pneumonitis for NSCLC patients treated with immunotherapy followed with thoracic radiotherapy. Int J Radiat Oncol Biol Phys. 2023;117:e79–79. 10.1016/J.IJROBP.2023.06.822 . Costa MVL, de Aguiar EJ, Rodrigues LS, Traina C Jr, Traina AJM. DEELE-Rad: exploiting deep radiomics features in deep learning models using COVID-19 chest X-ray images. Health Inf Sci Syst. 2024;13:11. 10.1007/s13755-024-00330-6 . Sakuma H, Matsuki M, Hasebe N, Nakagawa N. Real-world trends in pre-dialysis blood pressure levels of patients undergoing dialysis in Japan using a web-based national database. J Clin Hypertens (Greenwich). 2023;25:1163–71. 10.1111/jch.14736 . Dong B, Zhao Y, Wang J, Lu C, Chen Z, et al. Ren Fail. 2024;46:2403645. 10.1080/0886022X.2024.2403645 . Epidemiological analysis of chronic kidney disease from 1990 to 2019 and predictions to 2030 by Bayesian age-period-cohort analysis. Htun TP, Sun Y, Chua HL, Pang J. Clinical features for diagnosis of pneumonia among adults in primary care setting: A systematic and meta-review. Sci Rep. 2019;9:7600. 10.1038/s41598-019-44145-y . Ali S, Zehra A, Khalid MU, Hassan M, Shah SIA. Role of C-reactive protein in disease progression, diagnosis and management. Discoveries (Craiova). (2023). 11: e179. 10.15190/d.2023.18 Sheng W, Ming L, Xiangyu W, et al. The ratio of NT-proBNP to CysC 1.53 predicts heart failure in patients with chronic kidney disease. Front Cardiovasc Med. 2021;11:731864–731864. 10.3389/fcvm.2021.731864 . Hemyari AB, Gulati R. Uremic Alveoli: A rare case of diffuse alveolar hemorrhage due to severe uremia. J Clin Med Ther. 2018;3:1–3. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-7079017","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":497300444,"identity":"be536064-b547-4865-a423-2dbc29e86522","order_by":0,"name":"Yanan Wang","email":"","orcid":"","institution":"Ningbo Yinzhou NO.2 Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yanan","middleName":"","lastName":"Wang","suffix":""},{"id":497300445,"identity":"6fa8f818-a9a4-4f1d-b8a6-2e65d71529aa","order_by":1,"name":"Jingfeng Zhang","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Jingfeng","middleName":"","lastName":"Zhang","suffix":""},{"id":497300446,"identity":"3164f00d-8f19-4dc5-8d97-8a2100d1ef35","order_by":2,"name":"Qi Dai","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Dai","suffix":""},{"id":497300447,"identity":"e2d8f3d1-771e-4f5d-b536-3832e355fe71","order_by":3,"name":"Minghui Yan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA10lEQVRIiWNgGAWjYPACCTA68MHAxo40LQdnFKQlk2YRM8+HQ4wNhBTyz8gx/FyYY5EnH91jeNjG4AAzA/vhoxvwmn0jx1h65jaJYsM7ZwwO5xjc4WPgSUu7gU+LgUSOgTTvNonEjTNyQFqeMTNI8JgR0mL8G67FwuAwYwMRWszAtswHWneYgRgtEmeelVmDtGyQSCs42GOQlsxGyC/87cmbb/Nuq0ucPyN584cff2zs+NkPH8OrhUEgwwDiwgNQATa8ysHWHH8ApuUbCCodBaNgFIyCkQoAXRhLGclQkMsAAAAASUVORK5CYII=","orcid":"","institution":"Ningbo Yinzhou NO.2 Hospital","correspondingAuthor":true,"prefix":"","firstName":"Minghui","middleName":"","lastName":"Yan","suffix":""}],"badges":[],"createdAt":"2025-07-09 02:38:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7079017/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7079017/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88780949,"identity":"e54529ef-d64a-46c2-8942-7ce6753dd265","added_by":"auto","created_at":"2025-08-11 10:49:44","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":236909,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart for patient enrollment\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/1bfa6aae32e50256db93cd1d.png"},{"id":88780946,"identity":"f8625fb9-be40-4ae1-bdce-ce15e6acbba4","added_by":"auto","created_at":"2025-08-11 10:49:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":63279,"visible":true,"origin":"","legend":"\u003cp\u003eChest CT images and corresponding ROI delineation images of a patient with UP. (a) Original image; (b) Cross sectional hook drawing; (c) Coronal hook diagram; (d) 3D stereoscopic diagram\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/d1c8dc62c2124bc0c7c62520.png"},{"id":88782050,"identity":"afc25584-4d58-4ce6-9729-919b41d56f00","added_by":"auto","created_at":"2025-08-11 10:57:44","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":99290,"visible":true,"origin":"","legend":"\u003cp\u003eConfiguration of ResNet34\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/600cdbde5c91c908026029d5.png"},{"id":88782534,"identity":"d9773eef-b362-4aaa-968d-2a4ba53666c3","added_by":"auto","created_at":"2025-08-11 11:05:45","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":123331,"visible":true,"origin":"","legend":"\u003cp\u003eLASSO regression analysis for screening radiomics and DL features. (a) Plot of the partial likelihood deviance versus log(λ) for radiomics features; (b) LASSO variable shrinkage coefficient plot for the 7 radiomics features; (c) LASSO model regression analysis for radiomics features; (d) Plot of the partial likelihood deviance versus log(λ) for DL features; (e) LASSO variable shrinkage coefficient plot for the 9 DL features; (f) LASSO model regression analysis for DL features\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/83fcfe69f22607daddbe26c7.png"},{"id":88782527,"identity":"d31a59f4-cab6-4cf1-b91f-0aa7a68b7265","added_by":"auto","created_at":"2025-08-11 11:05:44","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":109380,"visible":true,"origin":"","legend":"\u003cp\u003eBox plots of Rad-score and Deepscore in the training set (a and c) and the test set (b and d).\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/ee67eb2ac047281859a4923a.png"},{"id":88780952,"identity":"ae33027c-ca7f-41bf-8df3-9270a8d4c2b7","added_by":"auto","created_at":"2025-08-11 10:49:44","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":118826,"visible":true,"origin":"","legend":"\u003cp\u003eModel evaluation of different models. (a-c) Comparison of ROC curves of four models in the training set, the test set, and the external validation set; (d) DCA of four models; (e) Calibration curves of the model\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/65ace3bf832aafbf1d17da92.png"},{"id":88782528,"identity":"14b4419b-ddf4-4d8d-a03f-387039545095","added_by":"auto","created_at":"2025-08-11 11:05:44","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":120052,"visible":true,"origin":"","legend":"\u003cp\u003eDeep learning radiomics clinical nomogram\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/9c49ec177d89aaa1c2c72c10.png"},{"id":90286057,"identity":"3bd41848-10cd-479e-a4e4-f7193bca1963","added_by":"auto","created_at":"2025-09-01 06:09:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1779342,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7079017/v1/5f7de93a-c477-437d-add2-abcfc0681465.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Research on the construction and application of a multi-feature auxiliary diagnostic model for uremic pneumonia based on radiomics and deep learning: a retrospective bi-centre study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eUremic pneumonitis (UP) is a secondary non-infectious pulmonary injury resulting from the accumulation of metabolites and fluid retention in end-stage renal disease (ESRD), also known as uremic pulmonary edema. As a common pulmonary complication of uremia and chronic renal failure, UP has an extremely high incidence rate. Related literature reports that its incidence rate can reach 50%\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. As one of the most urgent symptoms in patients with chronic kidney disease (CKD), UP has a mortality rate of 6.6% and a five-year survival rate of 21.2%\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. In a study involving 36 patients with UP, the misdiagnosis rate reached 58.3%\u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. Due to the low specificity of UP chest imaging and mild initial symptoms, UP is prone to being misdiagnosed as pulmonary infection (PI). Relevant reports indicate that the incidence rate of PI in this population can reach 19%\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. In fact, as the two most common pulmonary diseases among CKD patients, undifferentiated UP and PI share similarities in clinical symptoms and imaging manifestations, yet their treatment approaches differ significantly. Timely initiation of dialysis treatment is the most effective treatment for patients with UP. The \"Chinese Expert Consensus on Vascular Access for Hemodialysis (2nd Edition)\" states that the success rate of emergency dialysis rescue for UP can reach 95%\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e. For patients with PI, targeted anti-inflammatory therapy should be implemented based on specific pathogen types. Given that ESRD patients face a significantly increased cardiovascular risk when UP occurs before dialysis, preliminary diagnosis is of great importance for subsequent treatment. Therefore, early and accurate identification of non-dialysis UP patients is crucial for scientifically and efficiently managing this patient group and reducing the risk of adverse cardiovascular events.\u003c/p\u003e\u003cp\u003eIn recent years, artificial intelligence technologies such as radiomics and deep learning (DL) have been widely applied in various aspects of clinical practice, including auxiliary diagnosis, surgical path planning, lesion segmentation, and measurement. During the COVID-19 pandemic, radiomics and DL models provided technical support for accurate identification and disease monitoring of COVID-19\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. To address the limitations of manually defined features in traditional radiomics, some scholars have integrated DL with radiomics features, demonstrating excellent differential diagnostic capabilities in previous studies\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e. Therefore, integrating these two quantitative techniques into the differential diagnosis of UP is feasible and has clinical value.\u003c/p\u003e\u003cp\u003eThis study aimed to develop a scalable auxiliary diagnostic model that integrates radiomics and DL. The model incorporated clinical manifestations, laboratory examination results, and imaging data of UP to provide a scientific basis for the early identification and intervention of UP in clinical practice.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003eStudy participants\u003c/p\u003e\u003cp\u003e This study strictly adhered to the ethical principles outlined in the Declaration of Helsinki and obtained formal approval from the Ethics Committee of the Ningbo Yinzhou NO.2 Hospital (registration number: 2025-007). In accordance with relevant regulations, the requirement for research participants to sign informed consent forms was waived. A total of 1,373 patients diagnosed with CKD5 who received medical treatment at two hospitals from January 2019 to December 2023 were initially recruited. After applying the inclusion and exclusion criteria, 334 patients were ultimately included in this study. The development cohort (n\u0026thinsp;=\u0026thinsp;241) was randomly divided into a training set (n\u0026thinsp;=\u0026thinsp;169) and a test set (n\u0026thinsp;=\u0026thinsp;72) at a ratio of 7:3. All 93 patients in the external cohort were incorporated into the external validation set. The detailed process of case screening and grouping is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. It is worth noting that in the PI group (n\u0026thinsp;=\u0026thinsp;150) of the development cohort, the distribution of infection types was as follows: 41 cases of bacterial infection, 85 cases of viral infection, 15 cases of fungal infection, 3 cases of mycoplasma infection, and 1 case of chlamydia infection. In the PI group of the external cohort (n\u0026thinsp;=\u0026thinsp;56), there were 16 cases of bacterial infection, 32 cases of viral infection, 7 cases of fungal infection, and 1 case of mycoplasma infection.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe inclusion criteria were as follows: (1) Patients' renal function had progressed to CKD5, which was defined as an estimated glomerular filtration rate (eGFR) of less than 15 mL/(min\u0026middot;1.73 m\u0026sup2;). (2) Patients who had been diagnosed with newly-developed exudative pulmonary lesions confirmed by standard chest CT, and UP had been confirmed through pathological examination and clinical assessment. (3) Patients had been diagnosed with PI based on respiratory symptoms and pathogen detection results (e.g., positive culture results from sputum, blood, pleural effusion, or bronchoalveolar lavage fluid).\u003c/p\u003e\u003cp\u003eThe exclusion criteria were as follows: (1) Patients who were undergoing hemodialysis or peritoneal dialysis. (2) Patients with a history of tuberculosis, Goodpasture's syndrome, pulmonary embolism, pulmonary tumors, pulmonary trauma, pulmonary surgery, cardiogenic diseases, etc. (3) Patients with both UP and PI. (4) Patients with poor-quality imaging and incomplete clinical data.\u003c/p\u003e\u003cp\u003eThe baseline clinical data included variables such as gender, age, chest tightness, edema, cough, expectoration, body temperature (T), systolic blood pressure (SBP), diastolic blood pressure (DBP), C-reactive protein level (CRP), white blood cell count (WBC), hemoglobin (Hb), serum creatinine level (SCr), procalcitonin (PCT), N-terminal pro-B-type natriuretic peptide (NT-proBNP), arterial blood pH value (pH), and blood oxygen saturation (SpO₂).\u003c/p\u003e\u003cp\u003eCT imaging and image segmentation\u003c/p\u003e\u003cp\u003eThe CT scans in this study were performed using Kangda CT (China), GE optima 620 (USA), and Siemens Somatom go.Top (Germany) scanners. The CT scanning parameters were set as follows: a tube voltage of 120 kV, a tube current of 150\u0026ndash;300 mA, a layer thickness of 1-1.25 mm, and a matrix size of 512 \u0026times; 512.\u003c/p\u003e\u003cp\u003eThis study utilized the open-source software 3D Slicer (version 5.2.2, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.slicer.org/\u003c/span\u003e\u003cspan address=\"https://www.slicer.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) to carry out semi-automatic image segmentation. The image analysis was conducted by two radiologists with extensive experience in chest imaging diagnosis (5 years and 7 years, respectively). They conducted the image analysis without knowledge of the experimental results. Firstly, the CT images reconstructed by maximum intensity projection (MIP) were resampled to achieve voxel sizes of 1 mm \u0026times; 1 mm \u0026times; 1 mm and then imported into the 3D Slicer software (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). Subsequently, a junior physician utilized the \"Interactive Lobe Segmentation\" plugin to perform semi-automatic image segmentation based on thresholds to determine the approximate extent of the lungs. After that, the doctor manually adjusted to accurately depict the region of interest (ROI) encompassing the entire lung tissue (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb-c). Ultimately, the volume of the 3D region of interest (VOI) was calculated (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed). To ensure the accuracy and reliability of the ROI, a senior physician reviewed and modified the delineation results. If there were significant differences in the delineated lesion areas, the physician redefined the boundaries and made corresponding adjustments.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eExtraction of radiomics and deep learning features\u003c/p\u003e\u003cp\u003eUsing the 3D Slicer software, a total of 851 radiomics features were extracted from each ROI. These features included first-order features, 2D shape features, texture features, and transformed features. By cropping and normalizing the ROI into a 224 \u0026times; 224 pixel 3D image block, DL features were calculated and then input into the model. Transfer learning techniques were applied to compensate for insufficient data, and a moderately sized ResNet34 architecture was used to extract DL features, yielding a total of 512 features. The ResNet34 model comprised 34 learnable layers, and its specific architecture was as follows: (1) An initial feature extraction layer with 64 channels and a 7 \u0026times; 7 convolutional kernel. (2) A 3 \u0026times; 3 max-pooling layer (with a stride of 2) that reduced the dimensionality of the feature map to 56 \u0026times; 56. (3) Residual blocks with skip connections were utilized to gradually extract deep-level features and compress the feature map to 7 \u0026times; 7; (4) A 512-channel feature map was processed by average pooling to output a 512-dimensional feature vector\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eTo ensure the reproducibility of feature extraction, the two radiologists previously mentioned conducted reliability assessments on a randomly selected sample of 30 cases. For the intra-rater reliability assessment, the first radiologist obtained radiomics and DL features following specific procedures. The same methodology and steps were repeated two weeks later. Then, intra-rater correlation analyses were performed on the features extracted from these two sets. As for the inter-rater reliability assessment, the second radiologist employed the same methodologies and procedures as the first one. Subsequently, the features obtained by the second radiologist were compared and analyzed with those extracted by the first radiologist. The results demonstrated that both the intra-rater and inter-rater correlation coefficients exceeded 0.75 (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05), indicating a high level of consistency in the extracted features.\u003c/p\u003e\u003cp\u003eSelection of radiomics and deep learning features\u003c/p\u003e\u003cp\u003eFirstly, the minimum redundancy maximum relevance (mRMR) algorithm was utilized to eliminate redundant and irrelevant features. After that, the least absolute shrinkage and selection operator (LASSO) algorithm was employed to pick out the most predictive features. In the process of LASSO regression, the regularization parameter (λ) was optimized through ten-fold cross-validation to achieve precise feature selection. Finally, the selected features were combined linearly with their respective coefficients obtained from the LASSO regression. This linear combination led to the calculation of the radiomics score (Rad-score) and the DL score (Deepscore).\u003c/p\u003e\u003cp\u003eConstruction and evaluation of models\u003c/p\u003e\u003cp\u003eIn this study, logistic regression analysis was employed to construct the following four models. These models were constructed based on clinical risk factors, the Rad-score, and the Deepscore: a clinical model, a radiomics model, a DL model, and a combined deep learning-radiomics-clinical (DLRC) model.\u003c/p\u003e\u003cp\u003eA multi-dimensional evaluation framework was adopted to assess the performance of these models: (1) The ability of the four models to distinguish between UP and PI was quantified using the receiver operating characteristic (ROC) curve and its area under the curve (AUC) value. Inter-group comparisons were conducted using the DeLong test. (2) The generalization performance was evaluated using an external validation dataset. (3) A comprehensive evaluation was performed by integrating multiple metrics, including sensitivity, specificity, and accuracy. (4) The calibration of the models was validated by plotting calibration curves. (5) Decision curve analysis (DCA) was used to quantify the clinical utility of the four models. (6) The final results were presented visually through nomogram.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eIn this study, statistical analysis was conducted using the R language (version 4.3.2, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.r-project.org/\u003c/span\u003e\u003cspan address=\"https://www.r-project.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). For categorical variables like gender and clinical symptoms, either the chi-square test or Fisher's exact test was employed. For continuous variables like age and laboratory indicators, once normality testing had been conducted, the t-test was applied to variables that were found to conform to a normal distribution. Meanwhile, non-parametric tests were used for those variables that did not meet the normality criterion. The results were presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation or median (interquartile range). In-depth analyses were conducted using specialized packages within the R software. Feature selection was performed through the mRMR algorithm using the \"survcomp\" package. LASSO regression was conducted by turning to the \"glmnet\" package. Both nomograms and calibration curves, which were crucial for visualizing and validating the models, were constructed by applying the \"rms\" package. DCA was carried out by utilizing the \"rmda\" package. The ROC curves were plotted using the \"pROC\" package. The significance level was set at \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eClinical characteristics\u003c/p\u003e\u003cp\u003eThe characteristics of patients in the training set, the test set, and the external validation set are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Univariate and multivariate logistic regression analyses were carried out on the basic information, clinical symptoms, and laboratory examination results of the patients in the training set. The analysis revealed significant differences in SBP, NT-proBNP, Hb, and CRP between the two groups (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eClinical laboratory results and statistical outcomes of the patients\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"10\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eCharacteristic\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eTraining set\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;169)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u003cp\u003eTesting set\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;72)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e\u003cp\u003eExternal validation set (n\u0026thinsp;=\u0026thinsp;93)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c10\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eUP\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;63)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePI\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;106)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eUP\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;27)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003ePI\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;45)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003eUP\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;37)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e\u003cp\u003ePI\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;56)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.870\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.205\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.668\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e43.0 (68.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e68.0 (64.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e15.0 (55.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e16.0 (35.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e20.0 (54.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e34.0 (60.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e23.0 (31.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e38.0 (35.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e12.0 (44.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e29.0 (64.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e17.0 (45.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e22.0 (39.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge, M(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e60.5 (24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e64.0 (20.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.070\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e66.0 (28.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e65.0 (18.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.492\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e61.5 (17.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e72.0 (20.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEdema, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.426\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.795\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.119\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e33.0 (52.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e63.0(59.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e18.0 (66.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e32.0 (71.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e21.0 (56.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e41.0 (73.2)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e30.0 (47.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e43.0(40.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e9.0 (33.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e13.0 (28.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e16.0 (43.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e15.0 (26.8)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eChest tightness,\u003c/p\u003e\u003cp\u003en (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.207\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.805\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.396\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e25.0 (39.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e55.0 (51.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e17.0 (60.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e26.0 (57.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e19.0 (51.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e23.0 (41.1)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e38.0 (60.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e51.0 (48.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10.0 (39.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e19.0 (42.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e18.0 (48.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e33.0 (58.9)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCough, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.265\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.316\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.026\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e21.0 (33.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e47.0 (44.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e17.0 (60.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e19.0 (42.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e17.0 (45.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e13.0 (23.2)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e42.0 (66.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e59.0 (55.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e11.0 (39.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e26.0 (57.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e20.0 (54.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e43.0 (76.8)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eExpectoration,\u003c/p\u003e\u003cp\u003en (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.431\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.803\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.003\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e37.0 (58.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e54.0 (50.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e16.0 (59.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e28.0 (62.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e23.0 (62.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e17.0 (30.3)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e27.0 (41.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e52.0 (49.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e11.0 (40.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e17.0 (37.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e14.0 (37.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e39.0 (69.7)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT, M (Q₃-Q₁)\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e36.6 (0.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e36.8 (0.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.009\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e36.7 (0.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e36.8 (0.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.545\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e36.5 (0.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e36.8 (0.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.005\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSBP, Mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e158.2\u0026thinsp;\u0026plusmn;\u0026thinsp;19.7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e140.6\u0026thinsp;\u0026plusmn;\u0026thinsp;23.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e157.7\u0026thinsp;\u0026plusmn;\u0026thinsp;21.3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e140.4\u0026thinsp;\u0026plusmn;\u0026thinsp;23.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.004\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e156.3\u0026thinsp;\u0026plusmn;\u0026thinsp;27.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e143.6\u0026thinsp;\u0026plusmn;\u0026thinsp;24.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.022\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDBP,\u003c/p\u003e\u003cp\u003eM(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e85.0 (17.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e80.0 (13.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.017\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e86.5 (25.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e80.0 (18.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.040\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e82.0 (16.25)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e72.0 (20.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.005\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRP,\u003c/p\u003e\u003cp\u003eM(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.6 (8.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e10.9 (39.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2.2 (3.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e11.4 (33.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e9.7 (18.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e71.9 (87.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWBC,\u003c/p\u003e\u003cp\u003eM(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e6.2 (3.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6.8 (4.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.280\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e5.9 (2.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e7.0 (4.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.017\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e6.5 (3.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e7.4 (5.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.289\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHb, M(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e88.3\u0026thinsp;\u0026plusmn;\u0026thinsp;22.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e96.9\u0026thinsp;\u0026plusmn;\u0026thinsp;18.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.008\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e80.5\u0026thinsp;\u0026plusmn;\u0026thinsp;28.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e106.0\u0026thinsp;\u0026plusmn;\u0026thinsp;31.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e83.5\u0026thinsp;\u0026plusmn;\u0026thinsp;26.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e85.5\u0026thinsp;\u0026plusmn;\u0026thinsp;26.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.287\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSCr, M(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e631.0 (378.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e630.0 (403.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.805\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e591.5\u003c/p\u003e\u003cp\u003e(536.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e649.0 (369.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.455\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e639.0\u003c/p\u003e\u003cp\u003e(331.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e499.5\u003c/p\u003e\u003cp\u003e(303.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.068\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNT-proBNP,\u003c/p\u003e\u003cp\u003eM(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e18750.0 (20252.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7090.0\u003c/p\u003e\u003cp\u003e(14670.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e20720.0 (27157.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e7205.0 (18989.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.004\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e35000\u003c/p\u003e\u003cp\u003e(18610.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e9638.5 (19384.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePCT,\u003c/p\u003e\u003cp\u003eM(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.1 (0.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.2 (0.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.335\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.0 (0.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.1 (0.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.002\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.5 (0.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e0.5 (1.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.133\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePH, M(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7.4 (0.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7.4 (0.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.108\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e7.4 (0.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e7.4 (0.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.117\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e7.4 (0.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e7.4 (0.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.295\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpO2,\u003c/p\u003e\u003cp\u003eM(Q₃-Q₁)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e95.9 (4.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e97.5 (2.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.789\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e96.9 (3.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e97.5 (3.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.962\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e94.9 (5.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e94.2 (9.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.689\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"10\"\u003e\u003csup\u003ea\u003c/sup\u003eM: Median, Q₁: 1st Quartile, Q₃: 3st Quartile\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eResults of multivariate logistic regression analysis\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariable\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOR\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e95% CI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSBP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.032\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.011\u0026ndash;1.053\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.003\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.973\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.955\u0026ndash;0.991\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.003\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHb\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.978\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.958\u0026ndash;0.998\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.033\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNT-proBNP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.000\u0026ndash;1.000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eDevelopment of radiomics and deep learning models\u003c/p\u003e\u003cp\u003eA total of 851 radiomics features and 512 DL features were obtained in this study. By applying the mRMR algorithm and LASSO regression, 7 radiomics features and 9 DL features were ultimately selected (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe Rad-score was constructed using the 7 selected features and their corresponding regression coefficients. The specific formula is as follows:\u003c/p\u003e\u003cp\u003eRad-score\u0026thinsp;=\u0026thinsp;1.009 \u0026times; 10Percentile + (-0.35) \u0026times; High Gray Level Zone Emphasis + (-0.383) \u0026times; Flatness + (-0.39) \u0026times; Large Area Low Gray Level Emphasis\u0026thinsp;+\u0026thinsp;0.46 \u0026times; Major Axis Length + (-0.066) \u0026times; Large Dependence High Gray Level Emphasis + (-0.252) \u0026times; Maximum2D Diameter Row + (-0.76)\u003c/p\u003e\u003cp\u003eThe Deepscore was constructed using the 9 selected features and their corresponding regression coefficients. The specific formula is as follows:\u003c/p\u003e\u003cp\u003eDeepscore = (-0.5018) \u0026times; v_394 + (-0.736) \u0026times; v_273 + (-0.585) \u0026times; v_255\u0026thinsp;+\u0026thinsp;0.521 \u0026times; v_357 + (-0.195) \u0026times; v_17\u0026thinsp;+\u0026thinsp;0.121 \u0026times; v_201 + (-0.2) \u0026times; v_187 + (-0.263) \u0026times; v_232 + (-0.209) \u0026times; v_10 + (-0.986)\u003c/p\u003e\u003cp\u003eIn both the training set (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea, c) and the test set (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb, d), the Rad-score and Deepscore of the UP group were significantly higher than those of the PI group (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The AUC values for these scores are as follows: 0.862 (95% CI: 0.808\u0026ndash;0.917) and 0.881 (95% CI: 0.827\u0026ndash;0.935) for the training set; 0.813 (95% CI: 0.705\u0026ndash;0.921) and 0.855 (95% CI: 0.760\u0026ndash;0.950) for the test set; and 0.828 (95% CI: 0.739\u0026ndash;0.917) and 0.78 (95% CI: 0.675\u0026ndash;0.886) for the external validation set, respectively.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eEvaluation and validation of models\u003c/p\u003e\u003cp\u003eThe ability of the four models to distinguish between UP and PI was quantitatively evaluated using the ROC curves and the AUC values from the training set (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea), the test set (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb), and the external validation set (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ec), respectively. The accuracy (ACC), sensitivity (SEN), specificity (SPE), positive predictive value (PPV), and negative predictive value (NPV) of each model were calculated according to the confusion matrix (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The DeLong test results revealed statistically significant differences (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) in the AUC values between the combined DLRC model and each single-feature model. However, in the external validation set, there were no statistically significant differences in the ROC curves between the clinical model and the radiomics model (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.763). Evidently, the combined DLRC model demonstrated robust discriminative ability, with AUC values of 0.944 (95% CI: 0.911\u0026ndash;0.978), 0.930 (95% CI: 0.873\u0026ndash;0.986), and 0.885 (95% CI: 0.817\u0026ndash;0.952) in the training set, the test set, and the external validation set, respectively. Their sensitivities were 0.894, 0.833, and 0.892, respectively; specificities\u0026zwnj; were 0.883, 0.896, and 0.751, respectively; and accuracies were 0.888, 0.875, and 0.806, respectively. DCA (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ed) indicated that the DLRC model showed better clinical utility when the risk threshold was below 90%. Additionally, the calibration curve of the combined DLRC model (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ee) indicated good consistency between the predicted probabilities and the actual observed probabilities (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). By constructing a deep learning radiomics clinical nomogram (DLRCN), complex regression equations were converted into visual graphs to more intuitively present the diagnostic efficacy of the DLRC model for UP (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance of the four models in the training set, the test set, and the external validation set\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"9\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCohort\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAUC\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e95%CI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eACC\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSEN\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eSPE\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003ePPV\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e\u003cp\u003eNPV\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eClinical\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.828\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.766\u0026ndash;0.891\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.787\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.667\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.864\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.759\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.802\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.894\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.821\u0026ndash;0.967\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.833\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.833\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.833\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.714\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.909\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal validation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.809\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.716\u0026ndash;0.902\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.796\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.811\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.786\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.714\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.863\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eRadiomics\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.862\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.808\u0026ndash;0.917\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.798\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.788\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.806\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.722\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.856\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.813\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.705\u0026ndash;0.921\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.722\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.875\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.646\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.553\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.912\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal validation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.828\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.739\u0026ndash;0.917\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.774\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.784\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.768\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.690\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.843\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eDeep learning\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.881\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.827\u0026ndash;0.935\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.828\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.803\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.845\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.768\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.870\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.855\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.760\u0026ndash;0.950\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.806\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.792\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.813\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.679\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.886\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal validation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.780\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.675\u0026ndash;0.886\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.774\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.703\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.821\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.722\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.807\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eDLRC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.944\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.911\u0026ndash;0.978\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.888\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.894\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.883\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.831\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.929\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.930\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.873\u0026ndash;0.988\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.875\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.833\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.896\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.800\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.915\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal validation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.885\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.817\u0026ndash;0.952\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.806\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.892\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.751\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.702\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.913\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThrough a comprehensive systematic review of relevant domestic and international literature, this study is the first to integrate radiomics with DL for the diagnosis of UP, which showcases notable methodological innovation. Previous studies have utilized DL to differentiate generalized pulmonary edema from various types of pneumonia. Velichko E et al.\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e developed an auxiliary model named EDECOVID-net. This was the first method designed to distinguish between pulmonary edema and COVID-19, and it achieved an outstanding accuracy rate of 0.98. Similarly, Rachel et al.\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e integrated the squirrel search algorithm (SSA) with the backpropagation neural network (BPNN) for the differential diagnosis of pulmonary edema and COVID-19, and the study also exhibited excellent performance, attaining an accuracy rate of 0.88. However, these studies only used imaging data, did not incorporate other clinical factors, and failed to integrate radiomics and DL quantitative techniques. In contrast, this study established a combined DLRC model by integrating clinical factors, Rad-score, and Deepscore, exhibiting excellent diagnostic performance. In the training set, the AUC of the model for predicting UP reached 0.944, which was significantly higher than that of any single-feature model (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05). In the external validation set, when compared with other models in the same group, the DLRC model (AUC\u0026thinsp;=\u0026thinsp;0.885) exhibited superior discriminative ability, indicating that the combined model possessed good generalization ability and stability. In terms of accuracy, sensitivity, and specificity, the DLRC model demonstrated outstanding performance, guaranteeing high diagnostic accuracy and sensitivity for identifying UP cases, as well as robust discriminatory ability for non-target cases. In prior research on solid tumors, the DLRC model has been validated to yield comparable outcomes\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e. The current experimental findings offer crucial evidence-based support for the potential future application of this model in the field of diffuse lesions.\u003c/p\u003e\u003cp\u003eTo date, research on applying radiomics technology to address the differential diagnosis between UP and PI remains scarce. In recent years, the application of radiomics technology in the realm of pulmonary infectious diseases has become increasingly prevalent. Particularly after the onset of the COVID-19, radiomics has offered robust support for clinical decision-making and demonstrated its feasibility in identifying diffuse pulmonary lesions. This study extracted a total of 7 significant radiomics features, including 1 first-order feature, 3 shape features, and 3 texture features. The first-order feature, serving as the foundational level in radiomics analysis, is a feature value directly calculated based on the pixel intensity distribution of the original image\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. The first-order feature \"10Percentile\" represents the gray-level value of the low-intensity 10th percentile within the ROI. The gray-level value in medical imaging is closely correlated with the composition of pulmonary exudate. In cases of UP, due to elevated pressure, pulmonary tissue fluid leaks out and accumulates in the alveoli and interstitium. When PI occurs, the exudate comprises epithelial cells, purulent exudate, inflammatory cells, and its composition is relatively complex. The difference in the composition of pulmonary exudate between UP and PI results in alterations in gray-level values, which is consistent with the pathological manifestations.\u003c/p\u003e\u003cp\u003eShape features are employed to quantitatively describe the geometric properties of the ROI, primarily representing morphological details such as the shape, size, and volume of lesions. These features assist in evaluating the growth rate and spatial distribution of lesions\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. In this study, the values of \"Flatness\" and \"Maximum 2D Diameter Row\" in the UP group were both negative, indicating that the morphology of UP lesions was relatively flat. This finding was associated with the relatively limited extent of pulmonary involvement. It was hypothesized that this phenomenon might be related to the relative reduction in lung volume resulting from decreased lung function and alveolar collapse.\u003c/p\u003e\u003cp\u003eAs the third category of radiomics features in the results of this study, texture features reflect the microstructure and functional status of tissues by analyzing the 3D spatial arrangement of pixels, such as contrast, coarseness, and homogeneity\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. In this study, the features of \"High Gray Level Zone Emphasis\" and \"Large Area Low Gray Level Emphasis\" are incorporated into the Gray-Level Size Zone Matrix (GLSZM), while \"Large Dependence High Gray Level Emphasis\" belongs to the Gray-Level Dependence Matrix (GLDM). The GLSZM primarily describes the characteristics of homogeneous regions\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e. As a variant of GLSZM, the GLDM focuses on describing the gray-level dependencies within lesion regions\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. This category of radiomics features primarily reflects the internal heterogeneity of lesions. The higher the value, the coarser the image texture, and the more pronounced the 3D characteristics of the corresponding matrix, suggesting the presence of extensive and high-density lesion areas. As stated by Scholar Shiri\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e, when the lungs are infected to varying degrees, highly heterogeneous textures are generated, and the characteristic values of gray-level non-uniformity will increase. As of now, numerous studies have utilized radiomics analysis of CT texture features to achieve auxiliary diagnosis and prediction of pulmonary edema, attaining a high level of diagnostic performance\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e. Notably, the findings of this experiment were consistent with previous research findings. The model constructed based on the aforementioned radiomics features demonstrated good performance in distinguishing UP and PI. Although there was no statistically significant difference in the AUC between the radiomics model and the clinical-only model in the external validation set, this might be attributed to an insufficient sample size. DCA demonstrated that within the entire threshold probability interval, the radiomics model offered greater net clinical benefits compared to the clinical-only model, further substantiating the value of radiomics technology.\u003c/p\u003e\u003cp\u003eIn this study, we utilized a pre-trained 3D-ResNet34 network to extract DL features. This model employed a deep transfer learning approach to address the issue of limited dataset size, endowing the model with better generalization ability and replicability\u003csup\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e. Multivariate analysis results revealed that Deepscore was a key factor associated with the occurrence of UP. The DL model demonstrated high diagnostic performance on the training set, with sensitivity, specificity, and accuracy reaching 0.803, 0.845, and 0.828, respectively. The corresponding AUC value was 0.881, indicating strong discriminatory ability. Compared with the clinical-only model, the DL model was able to uncover information invisible to the naked eye, thereby exhibiting higher diagnostic efficiency and clinical practicality. However, in the external validation set, the performance of the DL prediction model was suboptimal, which might be due to its heavy reliance on large datasets, weakening its diagnostic performance with small sample sizes\u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e. In addition, compared to radiomics features, the 9 DL features obtained in this experiment exhibited poorer interpretability. The \"black box\" effect is one of the primary reasons why DL has not yet been widely adopted on a large scale. To address this issue, researchers have proposed utilizing visualization tools such as attention mechanisms and heatmaps for interpretation, or employing interpretability algorithm models like LIME or SHAP to tackle the aforementioned challenges\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e. Consequently, it is feasible to apply DL technology to the diagnosis of UP.\u003c/p\u003e\u003cp\u003eThe innovation of this study lay in the integration of DL with traditional radiomics features. This integration not only expanded the model's feature set but also ensured the reliability of the data input dimension. In previous studies, radiomics and DL were frequently considered as two independent techniques for constructing models. However, the features extracted by traditional radiomics methods are manually defined, suggesting that radiomics features may not encompass important characteristics that humans have not yet discovered. Numerous studies have demonstrated that DL can achieve autonomous learning from data, thereby overcoming the limitations of manual definition and enhancing the accuracy of image analysis. The DL-based radiomics model (DLR) has achieved remarkable results in disease classification, differential diagnosis, and prognosis evaluation\u003csup\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e. Therefore, DL serves as a powerful complement to traditional radiomics technology. The synergistic integration of the two has enhanced the model's efficacy. Compared with classification models constructed using single-source features, combined models yield higher overall net benefits.\u003c/p\u003e\u003cp\u003eThe DLRC model comprised four clinical variables: SBP, CRP, Hb, and NT-proBNP. Among these variables, SBP and CRP had relatively high weights in the nomogram. Elevated SBP has been identified as a major risk factor for CKD-related deaths globally. There is a strong correlation between increased blood pressure and the risks of CKD and ESRD\u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e. An intervention trial targeting SBP in CKD patients has confirmed that intensive antihypertensive treatment can effectively reduce the incidence of cardiovascular and cerebrovascular events. Another 7-year prospective cohort study demonstrated that patients with controlled blood pressure (SBP\u0026thinsp;\u0026lt;\u0026thinsp;130 mmHg) had a 64.2% lower risk of ESRD and a 30.4% lower all-cause mortality rate than patients with blood pressure below the target level\u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e. Therefore, early and sustained blood pressure management is crucial for achieving triple benefits for the heart, lungs, and kidneys.\u003c/p\u003e\u003cp\u003eCRP, an acute-phase protein synthesized by the liver, reflects the body's response to tissue injury through its release. CRP levels can rise under various pathological conditions, such as infections, autoimmune diseases, neurodegenerative diseases, and malignant tumors. During a systemic infection, CRP levels can soar to 1000 times the normal value. Related research has indicated that the diagnostic accuracy for community-acquired pneumonia peaks when the CRP cutoff value exceeds 20 mg/L\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e. Conversely, during the progression of CKD, there exists a synergistic interaction between the activation of oxidative stress responses and inflammatory reactions. Elevated CRP levels are associated with an increased risk of CKD. Following cardiovascular events, it has been confirmed that CRP levels deviate by approximately 25% from the baseline value\u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e. However, at this time, the magnitude of CRP elevation is relatively low compared to the changes induced by infection. Therefore, as an important diagnostic biomarker, the clinical value of CRP largely depends on the specific clinical settings.\u003c/p\u003e\u003cp\u003ePrevious researchers\u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e have posited a significant association between the radiographic manifestations of UP conditions and the levels of SCr and blood urea nitrogen (BUN). Higher levels of SCr and BUN corresponded to more severe chest pathological changes, and vice versa. However, Hemyari et al.\u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e argued that there was no significant correlation between the severity of SCr and BUN retention and the imaging manifestations of UP. In this study, there was no statistically significant difference in SCr levels between UP and PI. Therefore, it is currently not feasible to confirm that SCr is an independent risk factor for the occurrence of UP. The specific mechanisms by which various toxins trigger the onset and progression of UP, along with the intrinsic correlations between toxin levels and UP imaging manifestations, require further clinical investigations.\u003c/p\u003e\u003cp\u003eThe limitations of this study can be summarized as follows: (1) Currently, there is still a lack of a \"gold standard\" for diagnosing UP, which may affect the accuracy of the experiment. Therefore, it is essential for nephrologists and radiologists to perform a joint assessment based on disease progression, make evaluations, and offer a final diagnosis based on the actual clinical treatment response. (2) Although this study was conducted as a dual-center study, challenges in case collection, a relatively limited sample size, and differences in CT scanner parameters and laboratory testing standards between the two hospitals may have affected the models' performance. Additionally, the feasibility of integrating multiple types of information in practical clinical settings remains uncertain. Subsequent research will pursue multi-center collaborations to analyze larger datasets, aiming to improve the models' effectiveness and generalizability. (3) Considering the retrospective nature of this study, its conclusions are inevitably influenced by the inherent selection bias in historical datasets. It is essential to conduct prospective research in the future, employing a systematic and rigorous study design to collect new data for validating and supplementing existing research findings.\u003c/p\u003e\u003cp\u003eIn summary, an integrated model we proposed demonstrates strong potential as a highly valuable auxiliary tool for clinicians in the early diagnosis of UP among non-dialysis patients with CKD5.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this study, clinical factors, radiomics features, and DL features were integrated to develop a combined model. The DLRC model achieved the early non-invasive diagnosis of UP, providing an objective basis for individualized treatment decisions. Meanwhile, it has substantial reference value for optimizing the management of complications associated with chronic kidney disease.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cem\u003eUP\u003c/em\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eUremic pneumonia\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cem\u003ePI\u003c/em\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ePulmonary infection\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cem\u003eCKD\u003c/em\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eChronic kidney disease\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cem\u003eDL\u003c/em\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eDeep learning\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003e\u003cem\u003eDLRC\u003c/em\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eDeep learning-radiomics-clinical model\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials: Due to privacy considerations, the datasets generated during the current study are not publicly available. However, data may be available from the corresponding author (E-mail: [email protected]) upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the\u0026nbsp;radiology departments of Ningbo Yinzhou NO.2 Hospital and Ningbo No.2 Hospital, China,\u0026nbsp;which provided comprehensive help. In addition, we are very grateful to my colleagues and friends for their contributions to this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research received no external funding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors and Affiliations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDepartment of Radiology, Ningbo Yinzhou No.2 Hospital, Ningbo 315192, Zhejiang, China\u003c/p\u003e\n\u003cp\u003eYanan Wang \u0026amp; Minghui Yan\u003c/p\u003e\n\u003cp\u003eDepartment of Radiology, Ningbo No.2 Hospital, Ningbo 315010, Zhejiang, China\u003c/p\u003e\n\u003cp\u003eJingfeng Zhang \u0026amp; Qi Dai\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors contributed to the study conception and design. Material preparation and data collection were performed by Yanan Wang, Minghui Yan, and analysis were performed by Jingfeng Zhang and Qi Dai. The first draft of the manuscript was written by Yanan Wang and all authors commented on previous versions of the manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCorresponding author\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrespondence to Minghui Yan.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis retrospective study was approved by the Medical Ethics Committee of Ningbo Yinzhou NO.2 Hospital (registration number 2025\u0026ndash;007, approved on 18 April 2025), and written informed consent was waived. Because the medical records used in this study were obtained from past clinical diagnosis and treatment, this clinical study did not directly involve the subjects, and the results\u0026nbsp;were not used for subject diagnosis. Therefore, it will not have any adverse effects on the subjects. The privacy and personal identity information of the subjects are protected. Furthermore, the study adheres to the Helsinki Declaration.\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\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eShaik L, Thotamgari SR, Kowtha P, Ranjha S, Shah RN, et al. A Spectrum of pulmonary complications occurring in end-stage renal disease patients on maintenance hemodialysis. Cureus. 2021;13:e15426. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7759/cureus.15426\u003c/span\u003e\u003cspan address=\"10.7759/cureus.15426\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBanerjee D, Ma JZ, Collins AJ, Herzog CA. Long-term survival of incident hemodialysis patients who are hospitalized for congestive heart failure, pulmonary edema, or fluid overload. Clin J Am Soc Nephrol. 2007;2:1186\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2215/CJN.01110307\u003c/span\u003e\u003cspan address=\"10.2215/CJN.01110307\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYi J, Liu Y, Hu Z. Relation between changes in chest imaging pathology and levels of SCr and BUN in uremic patients. Med J Natl Def Forces Southwest China. 2016;26:863\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eA. G A MGT. Microbiological assay of respiratory infections in kidney diseases: a retrospective study in 2019. Egypt J Chest Dis Tuberc. 2025;74:214\u0026ndash;20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.4103/ecdt.ecdt_116_22\u003c/span\u003e\u003cspan address=\"10.4103/ecdt.ecdt_116_22\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJin Q, Wang Y, Ye C et al. Consensus among experts on blood access used for hemodialyis in China (The 2nd edition). CJBP. (2019), 18: 365\u0026ndash;381.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAmmirabile A, Cavinato L, Ferro CAP, Fiz F, Savino MS, et al. CT-radiomics and pathological tumor response to systemic therapy: A predictive analysis for colorectal liver metastases. Development and internal validation of a clinical-radiomic model. Eur J Surg Oncol. 2025;51:109557. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ejso.2024.109557\u003c/span\u003e\u003cspan address=\"10.1016/j.ejso.2024.109557\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFang X, Li X, Bian Y, Ji X, Lu J. Radiomics nomogram for the prediction of 2019 novel coronavirus pneumonia caused by SARS-CoV-2. Eur Radiol. 2020;30:6888\u0026ndash;901. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00330-020-07032-z\u003c/span\u003e\u003cspan address=\"10.1007/s00330-020-07032-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSu W, Cheng D, Ni W, Ai Y, Yu X, et al. Multi-omics deep learning for radiation pneumonitis prediction in lung cancer patients underwent volumetric modulated arc therapy. Comput Methods Programs Biomed. 2024;254. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cmpb.2024.108295\u003c/span\u003e\u003cspan address=\"10.1016/j.cmpb.2024.108295\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa G, Wang K, Zeng T, Sun B, Yang L. A Joint Classification Method for COVID-19 Lesions Based on Deep Learning and Radiomics. Tomography. 2024;10:1488\u0026ndash;500. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/tomography10090109\u003c/span\u003e\u003cspan address=\"10.3390/tomography10090109\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTian S, Yu Y, Shi K, Jiang Y, Song H, et al. Deep learning radiomics based on ultrasound images for the assisted diagnosis of chronic kidney disease. Nephrol (Carlton). 2024;29:748\u0026ndash;57. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/nep.14376\u003c/span\u003e\u003cspan address=\"10.1111/nep.14376\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVelichko E, Shariaty F, Orooji M, Pavlov V, Pervunina T, Zavjalov S, Khazaei R, et al. Development of computer-aided model to differentiate COVID-19 from pulmonary edema in lung CT scan: EDECOVID-net. Comput Biol Med. 2022;141:105172. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2021.105172\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2021.105172\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBetshrine RR, Nehemiah KH, Marishanjunath CS, et al. Diagnosis of pulmonary edema and Covid-19 from CT slices using squirrel search algorithm, support vector machine and back propagation neural network. JIFS. 2023;44:5633\u0026ndash;46. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3233/JIFS-222564\u003c/span\u003e\u003cspan address=\"10.3233/JIFS-222564\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu F, Zhang BD, Cheng HS, et al. A radiomics and deep learning nomogram developed and validated for predicting no-collapse survival in patients with osteonecrosis after multiple drilling. BMC Med Inf Decis Mak. 2025;25:26. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12911-025-02859-2\u003c/span\u003e\u003cspan address=\"10.1186/s12911-025-02859-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWen Z, Gao X, Wu Q, Yang J, Sun J, et al. Baseline \u003csup\u003e[18F]\u003c/sup\u003eFDG PET/CT radiomics for predicting interim efficacy in follicular lymphoma treated with first-line R-CHOP. BMC Cancer. 2025;25:128. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12885-025-13507-3\u003c/span\u003e\u003cspan address=\"10.1186/s12885-025-13507-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu Y, Cao F, Lei H, Zhang S, et al. Interpretable multiphasic CT-based radiomic analysis for preoperatively differentiating benign and malignant solid renal tumors: a multicenter study. Abdom Radiol (NY). 2024;49:3096\u0026ndash;106. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00261-024-04351-3\u003c/span\u003e\u003cspan address=\"10.1007/s00261-024-04351-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePiccolo CL, Sarli M, Pileri M, Tommasiello M, Rofena A, et al. Radiomics for Predicting Prognostic Factors in Breast Cancer: Insights from Contrast-Enhanced Mammography (CEM). J Clin Med. 2024;13:6486. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/jcm13216486\u003c/span\u003e\u003cspan address=\"10.3390/jcm13216486\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang X, Luo X, Pan H, Wang X, Xu S, Li H, Lin Z. Performance of hippocampal radiomics models based on T2-FLAIR images in mesial temporal lobe epilepsy with hippocampal sclerosis. Eur J Radiol. 2023;167:111082. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ejrad.2023.111082\u003c/span\u003e\u003cspan address=\"10.1016/j.ejrad.2023.111082\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePisani N, Abate F, Avallone AR, Barone P, Cesarelli M, et al. A radiomics approach to distinguish progressive supranuclear palsy richardson's syndrome from other phenotypes starting from MR images. Comput Methods Programs Biomed. 2025;266:108778. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cmpb.2025.108778\u003c/span\u003e\u003cspan address=\"10.1016/j.cmpb.2025.108778\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTian X, MA A, Jiang L, et al. CT imaging-based on texture analysis: discrimination of high altitude pulmonary edema and acute cardiogenic pulmonary edema. Radiol Pract. 2020;35(01):45\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.13609/j.cnki.1000-0313.2020.01.009\u003c/span\u003e\u003cspan address=\"10.13609/j.cnki.1000-0313.2020.01.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrusasco C, Santori G, Tavazzi G, et al. Second-order grey-scale texture analysis of pleural ultrasound images to differentiate acute respiratory distress syndrome and cardiogenic pulmonary edema. J Clin Monit Comput. 2020;36:1\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10877-020-00629-1\u003c/span\u003e\u003cspan address=\"10.1007/s10877-020-00629-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNijiati M, Tuerdi M, Damola M, Yimit Y, Yang J, et al. A deep learning radiomics model based on CT images for predicting the biological activity of hepatic cystic echinococcosis. Front Physiol. 2024;15:1426468. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fphys.2024.1426468\u003c/span\u003e\u003cspan address=\"10.3389/fphys.2024.1426468\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXie K, Jiang H, Chen X, Ning Y, Yu Q, et al. Multiparameter MRI-based model integrating radiomics and deep learning for preoperative staging of laryngeal squamous cell carcinoma. Sci Rep. 2025;15:16239. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-025-01270-1\u003c/span\u003e\u003cspan address=\"10.1038/s41598-025-01270-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGanji Z, Nikparast F, Shoeibi N, Shoeibi A, Zare H. Decoding Parkinson's Diagnosis: An OCT-Based Explainable AI with SHAP/LIME Transparency from the Persian Cohort Study. Photodiagnosis Photodyn Ther. 2025;14:104668. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.pdpdt.2025.104668\u003c/span\u003e\u003cspan address=\"10.1016/j.pdpdt.2025.104668\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRathore PS, Kumar A, Nandal A, Dhaka A, Sharma AK. A feature explainability-based deep learning technique for diabetic foot ulcer identification. Sci Rep. 2025;15:6758. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-025-90780-z\u003c/span\u003e\u003cspan address=\"10.1038/s41598-025-90780-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen J, Liu S, Lin Y, Hu W, Shi H, Liao N, et al. The Quality and Accuracy of Radiomics Model in Diagnosing Osteoporosis: A Systematic Review and Meta-analysis. Acad Radiol. 2025;32:2863\u0026ndash;75. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.acra.2024.11.065\u003c/span\u003e\u003cspan address=\"10.1016/j.acra.2024.11.065\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYuan Z, Handcrafted Radiomics. Deep learning radiomics in the prediction of radiation pneumonitis for NSCLC patients treated with immunotherapy followed with thoracic radiotherapy. Int J Radiat Oncol Biol Phys. 2023;117:e79\u0026ndash;79. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/J.IJROBP.2023.06.822\u003c/span\u003e\u003cspan address=\"10.1016/J.IJROBP.2023.06.822\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCosta MVL, de Aguiar EJ, Rodrigues LS, Traina C Jr, Traina AJM. DEELE-Rad: exploiting deep radiomics features in deep learning models using COVID-19 chest X-ray images. Health Inf Sci Syst. 2024;13:11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s13755-024-00330-6\u003c/span\u003e\u003cspan address=\"10.1007/s13755-024-00330-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSakuma H, Matsuki M, Hasebe N, Nakagawa N. Real-world trends in pre-dialysis blood pressure levels of patients undergoing dialysis in Japan using a web-based national database. J Clin Hypertens (Greenwich). 2023;25:1163\u0026ndash;71. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/jch.14736\u003c/span\u003e\u003cspan address=\"10.1111/jch.14736\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDong B, Zhao Y, Wang J, Lu C, Chen Z, et al. Ren Fail. 2024;46:2403645. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/0886022X.2024.2403645\u003c/span\u003e\u003cspan address=\"10.1080/0886022X.2024.2403645\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epidemiological analysis of chronic kidney disease from 1990 to 2019 and predictions to 2030 by Bayesian age-period-cohort analysis.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHtun TP, Sun Y, Chua HL, Pang J. Clinical features for diagnosis of pneumonia among adults in primary care setting: A systematic and meta-review. Sci Rep. 2019;9:7600. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-019-44145-y\u003c/span\u003e\u003cspan address=\"10.1038/s41598-019-44145-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAli S, Zehra A, Khalid MU, Hassan M, Shah SIA. Role of C-reactive protein in disease progression, diagnosis and management. Discoveries (Craiova). (2023). 11: e179. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.15190/d.2023.18\u003c/span\u003e\u003cspan address=\"10.15190/d.2023.18\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSheng W, Ming L, Xiangyu W, et al. The ratio of NT-proBNP to CysC\u003csup\u003e1.53\u003c/sup\u003e predicts heart failure in patients with chronic kidney disease. Front Cardiovasc Med. 2021;11:731864\u0026ndash;731864. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcvm.2021.731864\u003c/span\u003e\u003cspan address=\"10.3389/fcvm.2021.731864\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHemyari AB, Gulati R. Uremic Alveoli: A rare case of diffuse alveolar hemorrhage due to severe uremia. J Clin Med Ther. 2018;3:1\u0026ndash;3.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Uremic pneumonitis, Pulmonary infection, Computed tomography, Radiomics, Deep learning","lastPublishedDoi":"10.21203/rs.3.rs-7079017/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7079017/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e\u003cp\u003eThis study aimed to develop and validate a diagnostic model that integrates radiomics and deep learning (DL) for the early differentiation of uremic pneumonia (UP) from pulmonary infection (PI) in non-dialysis patients.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eThis study retrospectively analyzed the clinical data and non-contrast CT images of 334 non-dialysis patients with stage 5 chronic kidney disease (CKD). In this study, univariate and multivariate logistic regression analysis methods were employed to screen for clinical risk factors. Radiomics and DL features of lesions were extracted based on non-contrast chest CT images. Clinical model, radiomics model, and DL model were constructed separately, and a combined deep learning-radiomics-clinical (DLRC) model was established using feature fusion methods. The performance of each model was evaluated using the receiver operating characteristic (ROC) curve, the area under the curve (AUC), the calibration curve, and decision curve analysis (DCA). In addition, the generalization ability of each model was validated using an external validation cohort (n\u0026thinsp;=\u0026thinsp;93).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eThe DLRC model demonstrated the highest diagnostic performance, with AUC values of 0.944 (95% CI: 0.911\u0026ndash;0.978) in the training set and 0.930 (95% CI: 0.873\u0026ndash;0.985) in the test set. The AUC values of each single-feature model in the training set and the test set were as follows: clinical model (0.828 and 0.894), radiomics model (0.862 and 0.813), and DL model (0.881 and 0.855). In the external validation set, the DLRC model continued to show stable diagnostic performance (AUC\u0026thinsp;=\u0026thinsp;0.885, 95% CI: 0.817\u0026ndash;0.952).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eThe DLRC model demonstrated outstanding performance in differentiating between UP and PI, which facilitated the early identification of UP and the formulation of individualized diagnosis and treatment plans.\u003c/p\u003e","manuscriptTitle":"Research on the construction and application of a multi-feature auxiliary diagnostic model for uremic pneumonia based on radiomics and deep learning: a retrospective bi-centre study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-11 10:49:39","doi":"10.21203/rs.3.rs-7079017/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"dd4df2c3-bead-477d-8ef3-58aa54105538","owner":[],"postedDate":"August 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-01T06:09:22+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-11 10:49:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7079017","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7079017","identity":"rs-7079017","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}