Objective To explore the predictive value of a radiomics model based on dual-phase enhanced computed tomography (CT) for progression-free survival (PFS) in patients with small cell lung cancer (SCLC). Methods Clinical data of 148 SCLC patients confirmed by histopathology were collected retrospectively, including 88 cases from Center 1 (served as the training set) and 60 cases from Center 2 (served as the validation set). Three-dimensional volumes of interest (VOIs) were automatically delineated on arterial-phase and venous-phase enhanced CT images respectively. Radiomics features significantly correlated with PFS were extracted and screened, and the radiomics score (Rad-score) was calculated. Cox regression analysis was used to identify independent clinical risk factors affecting PFS. Subsequently, a clinical model, a radiomics model, and a combined model were constructed based on the independent clinical risk factors and Rad-score, and the predictive efficacy of these models was evaluated. Results A total of 5 radiomics features were finally selected. Clinical stage was an independent risk factor for PFS in SCLC patients (P< 0.001, HR=5.058, 95%CI: 2.139-11.960). According to Rad-score, SCLC patients were divided into the high-risk group (Rad-score ≥ 0.17) and the low-risk group (Rad-score<0.17), with a statistically significant difference in survival between the two groups (validation set:P< 0.001,HR=3.002, 95%CI:1.580-5.706). The predictive efficacy of the radiomics model (C-index: 0.826) and the combined model (C-index: 0.828) for PFS in SCLC patients was significantly higher than that of the clinical model (C-index: 0.582). Compared with the clinical model, the radiomics model had a net reclassification index (NRI) of 0.647 (95%CI: 0.419-0.842,P<0.05) and an integrated discrimination improvement index (IDI) of 0.324 (95%CI: 0.165-0.488,P<0.05). There were no significant differences in NRI and IDI between the radiomics model and the combined model (P>0.05). In terms of clinical utility, both the radiomics model and the combined model were superior to the clinical model. Conclusion The radiomics model based on dual-phase enhanced CT shows excellent performance in predicting PFS of SCLC patients, which can provide valuable information for individualized treatment.

Objective To explore the relationship between fat attenuation index (FAI), myocardial bridge (MB) parameters and coronary atherosclerosis in the proximal segment of left anterior descending artery (LAD) myocardial bridge (MB), and to evaluate their value in risk prediction. Methods A total of 243 patients with LAD-MB (confirmed by coronary computed tomography angiography, CCTA) were enrolled. The FAI and MB parameters in the proximal segment of MB were measured. The patients were divided into four groups: simple MB group (Group A, n=127), MB combined with proximal atherosclerosis group (Group B, n=116), Group B1 (coronary atherosclerotic stenosis >50%, n=30) and Group B2 (matched control for Group B1, n=30). Logistic regression analysis was used to identify risk factors, and receiver operating characteristic (ROC) curves were plotted. Results There were significant differences in age, hypertension, smoking history, MB length, myocardial bridge index (MMI), stenosis rate and FAI between Group A and Group B (all P<0.05). Age (OR=1.055), stenosis rate (OR=1.160) and FAI (OR=1.099) were independent risk factors, with their respective area under the ROC curve (AUC) values being 0.675, 0.846 and 0.666. The combined AUC of these three factors increased to 0.889. In Group B1 and Group B2, only FAI was an independent risk factor for coronary atherosclerotic stenosis >50% (OR=1.130, AUC=0.719). Conclusion The combination of age, MB stenosis rate and FAI can predict proximal atherosclerosis (AUC=0.889), and FAI is the only predictor for coronary atherosclerotic stenosis >50% (P<0.05).

Objective To evaluate the added value of clinical and radiomic features in differentiating benign and malignant non-mass enhancement (NME) lesions based on breast multi-parametric magnetic resonance imaging (MRI). Methods A retrospective analysis was performed on 147 NME patients who met the inclusion and exclusion criteria between September 2021 and September 2024. These patients were randomly divided into a training set and a validation set at a ratio of 7∶3. Radiomic features were extracted from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and apparent diffusion coefficient (ADC) images. The least absolute shrinkage and selection operator (LASSO) model was used for feature selection and to construct a radiomic model. For clinical features, a clinical model was established using multivariate Logistic regression analysis. The optimal-performing radiomic model was combined with clinical features to build a visual nomogram model. The performance of the nomogram model was evaluated using calibration curves and decision curve analysis (DCA). Results The results showed that the nomogram model exhibited excellent performance in distinguishing benign and malignant NME lesions. The area under the curve (AUC) was 0.928 in the training set and 0.882 in the validation set. This performance was superior to that of the single clinical model and the single radiomic model. Conclusion sThis study used clinical features combined with multi-parametric MRI radiomics to quantitatively analyze the information contained in images. It was confirmed that the nomogram model provides good benefits for clinical decision-making in differentiating benign and malignant NME lesions, which is conducive to disease treatment decisions.

Objective To evaluate the efficacy of quantitative parameters [iodine concentration (IC) and normalized iodine concentration (nIC)] of spectral CT in identifying lymph node metastasis (LM) in gastric cancer using evidence-based Meta-analysis. Methods A comprehensive search was conducted in six major databases [PubMed, EMBASE, Cochrane Library, Web of Science, China National Knowledge Infrastructure (CNKI), and Wanfang Database] for literature on the diagnosis of gastric cancer LM by spectral CT, published from January 2010 to January 2025. Literature meeting the inclusion criteria was subjected to quality assessment, characteristic data extraction, and heterogeneity evaluation. According to the research objects, the literature was divided into two categories: prediction of LM using spectral CT parameters of lymph nodes, and prediction of LM using spectral CT parameters of gastric tumors. For the most studied parameters in lymph node research [arterial-phase normalized iodine concentration (nICa)] and the most studied parameters in gastric tumor research [venous-phase iodine concentration (ICp) and venous-phase normalized iodine concentration (nICp)], effect sizes were pooled to obtain combined sensitivity, specificity, positive likelihood ratio, and diagnostic odds ratio. Forest plots and summary receiver operating characteristic (SROC) curves were drawn, and the area under the curve (AUC) was calculated. Begg's funnel plot asymmetry test was used to detect publication bias. Results A total of 10 studies meeting the inclusion criteria were included in the Meta-analysis, and they were classified by data type into 3 studies with lymph node data and 7 studies with gastric tumor data. There was publication bias for lymph node nICa, while no publication bias was found for tumor ICp and nICp. The combined AUCs of the three parameters were 0.85 (95%CI: 0.81-0.87), 0.79 (95%CI: 0.76-0.83), and 0.83 (95%CI: 0.80-0.86), respectively. Conclusion Spectral CT is applied in the evaluation of gastric cancer LM due to its advantages of non-invasiveness, convenience, cost-effectiveness, and efficiency. Among its parameters, those related to iodine concentration have certain application value, providing a new approach for preoperative evaluation of gastric cancer LM.

Objective To evaluate the diagnostic efficacy of artificial intelligence (AI)-based non-gated coronary artery calcium score (CACS), and preliminarily explore whether it has the ability to accurately and stably stratify populations with different cardiovascular disease risk factors. Methods A total of 184 patients who underwent low-dose non-gated chest CT plain scan and coronary CT angiography were collected retrospectively. CACS was measured and graded according to the following criteria: CACS = 0 for extremely low risk, 0 < CACS < 100 for low risk, 100 ≤ CACS < 400 for moderate risk, and CACS ≥ 400 for high risk. For the overall population assessment, Spearman correlation coefficient (r), Bland-Altman method, and intraclass correlation coefficient (ICC) were used to evaluate the correlation and consistency between the two methods (non-gated CACS and gated CACS). Weighted Kappa analysis was applied to assess the consistency of coronary artery calcification risk stratification. With electrocardiogram (ECG)-gated calcium score as the gold standard, predicted values were calculated via ordinal Logistic regression, and a multiclass receiver operating characteristic (ROC) curve was plotted using the “one-vs-rest” strategy to evaluate the performance of non-gated CACS. Second-order clustering was used to divide the population into different subgroups based on cardiovascular disease risk factors, with log-likelihood distance metric and Bayesian Information Criterion (BIC) as the bases. In each subgroup, ICC, r, and the diagnostic efficacy for moderate- and high-risk populations (CACS > 100) were further evaluated. A P-value < 0.05 was considered statistically significant. Results For the overall population, the correlation coefficient of non-gated CACS was r = 0.965 (P< 0.001), ICC = 0.970 (P < 0.001), the area under the curve (AUC) for each risk category was > 0.9 (P< 0.001), and the weighted Kappa was 0.854 (P< 0.001). Using second-order clustering, the population was divided into three subgroups. In all subgroups, r > 0.9 (P< 0.001) and ICC > 0.9 (P < 0.001). Additionally, non-gated CACS showed good stratification ability for moderate- and high-risk populations in different subgroups, with AUC > 0.9 in all cases. Conclusion Non-gated calcium score has high reliability and stability for risk grading, and is suitable for coronary heart disease screening in different populations.

Objective To investigate alterations in cerebral blood flow (CBF) and cerebral blood flow network under different Alzheimer's disease (AD)-related pathological burdens and cognitive states, and to deepen the understanding of neurobiological mechanisms in the AD disease spectrum. Methods Data were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, including 29 Aβ-negative cognitively normal subjects (Aβ-CN), 34 Aβ-positive cognitively normal subjects (Aβ+ CN), and 39 Aβ-positive cognitively impaired subjects (Aβ+ CI). Demographic data, neuropsychological test results, and imaging data of the subjects were collected. CBF information was obtained from arterial spin labeling (ASL) imaging. Voxel-based morphometry was used to compare and identify brain regions with CBF differences among the three groups, and partial correlation analysis was conducted between CBF in these regions and scores of neuropsychological assessment scales. A cerebral blood flow network based on CBF values was constructed, and graph theory analysis was applied to explore the changing characteristics of global network properties and nodal properties. Results Significant differences in CBF among the three groups were observed in brain regions including the frontal lobe, temporal lobe, and parietal lobe (GRF correction, voxel-level P < 0.001, cluster-level P< 0.05). Compared with the Aβ-CN group, the Aβ+ CN group showed significantly decreased CBF in the bilateral anterior cingulate and paracingulate gyri, left inferior parietal angular gyrus, left middle temporal gyrus, and left inferior temporal gyrus. Compared with the Aβ+ CN group, the Aβ+ CI group had significantly decreased CBF in the right orbital part of the middle frontal gyrus and left middle temporal gyrus; compared with the Aβ-CN group, the Aβ+ CI group showed significantly decreased CBF in all the above-mentioned brain regions with differences (all P<0.05). Correlation analysis indicated that CBF in the right orbital part of the middle frontal gyrus was negatively correlated with the score of Alzheimer's Disease Assessment Scale-Cognitive Subscale (ADAS-13) (r = -0.338, P= 0.011, FDR correction); CBF in the left inferior temporal gyrus was positively correlated with the score of Montreal Cognitive Assessment (MoCA) (r = 0.329, P = 0.011, FDR correction); CBF in the right anterior cingulate and paracingulate gyri was positively correlated with the MoCA score (r = 0.280, P= 0.044, FDR correction). Graph theory analysis results showed that compared with the Aβ+ CN group, the Aβ+ CI group had a significantly decreased global efficiency (P= 0.039); compared with the Aβ- CN group, the Aβ+ CI group had significantly increased small-worldness(P= 0.034) and local efficiency (P= 0.032), global efficiency was significantly decreased(P=0.005). Conclusion With the progression of Aβ pathology and the decline of cognitive function, CBF and cerebral blood flow network patterns in regions such as the frontal lobe, temporal lobe, and parietal lobe undergo significant changes. These findings provide a certain theoretical basis for exploring the pathological and neurobiological mechanisms of the AD disease spectrum from the perspective of cerebral blood flow.

Objective To explore changes in the resting-state brain functional network of patients with diabetic retinopathy (DR) using degree centrality (DC) and functional connectivity (FC) methods. Methods A total of 26 DR patients, 30 diabetic patients without retinopathy (NDR group), and 30 healthy controls (HC group) were enrolled. The differences in DC values across various brain regions were compared among the three groups. Brain regions with intergroup DC differences were used as seed points to analyze whether there were abnormalities in whole-brain FC values. Partial correlation analysis was applied to evaluate the correlations between DC values, FC values, scores of neuropsychological assessment scales, and clinical biochemical indicators. Results There were statistically significant differences in DC values among the three groups in the right postcentral gyrus, right precentral gyrus, left lingual gyrus, and right superior temporal gyrus (P<0.05). Compared with the HC group, the DR group had significantly decreased DC values in the right postcentral gyrus, right precentral gyrus, and left lingual gyrus; the NDR group had significantly decreased DC value in the left lingual gyrus and significantly increased DC value in the right superior temporal gyrus. Compared with the NDR group, the DR group had significantly decreased DC values in the right postcentral gyrus, right precentral gyrus, left lingual gyrus, and right superior temporal gyrus. Functional connectivity analysis among the three groups showed that when the left lingual gyrus was used as the seed point, there were statistically significant differences in FC values between the left lingual gyrus and the right lingual gyrus, as well as between the left lingual gyrus and the right cuneus. Partial correlation analysis revealed that in the DR group, the DC value of the right precentral gyrus was significantly positively correlated with the high-density lipoprotein cholesterol (HDL-C) level (r=0.45, P=0.037). Conclusion DR patients have abnormal brain functional activity in multiple brain regions, including the right postcentral gyrus, right precentral gyrus, left lingual gyrus, and right superior temporal gyrus. These abnormal regions may be related to functions such as vision, somatosensation, and cognition.

Objective To analyze the risk factors of clinically significant prostate cancer (csPCa) using LASSO and multivariate Logistic regression, construct a nomogram prediction model, and explore the diagnostic value of PI-RADSv2.1 score combined with PSA-derived parameters for csPCa. Methods A total of 230 patients with pathologically confirmed prostate cancer or benign prostatic hyperplasia were included retrospectively. They were randomly divided into a training set (n=160) and a validation set (n=70) at a ratio of 7∶3. According to pathological results, patients were classified into the csPCa group (clinically significant prostate cancer) and the no-csPCa group (non-clinically significant prostate cancer). MRI data were analyzed to conduct PI-RADSv2.1 scoring. Prostate-specific antigen density (PSAD), prostate-specific antigen transition zone density (PSAT), free PSA ratio/prostate-specific antigen density [(F/T)/PSAD], and free PSA ratio/prostate-specific antigen transition zone density [(F/T)/PSAT] were calculated. Differences in each parameter between the two groups were compared, and correlation coefficient heatmap analysis was performed. The LASSO regression model was used to screen the most valuable parameters, and multivariate Logistic regression analysis was applied to construct a prediction model and a nomogram. Receiver operating characteristic (ROC) curve, calibration curve, and decision curve were used to evaluate the predictive efficacy of the model. Results In both the training set and validation set, PSAT, PSAD, and PI-RADSv2.1 scores in the csPCa group were significantly higher than those in the no-csPCa group, while (F/T)/PSAT and (F/T)/PSAD were significantly lower than those in the no-csPCa group; the differences in all parameters between the two groups were statistically significant (all P<0.001). Three most valuable parameters were screened out by LASSO regression, including age, PI-RADSv2.1 score, and (F/T)/PSAT. Multivariate Logistic regression analysis showed that age, PI-RADSv2.1 score, and (F/T)/PSAT were independent predictors of csPCa. A nomogram model was established based on these factors. The calibration curve fitted well with the ideal curve. The area under the ROC curve (AUC) of the model was 0.926, with a sensitivity of 88.7% and a specificity of 86.0%. Conclusion This nomogram model has high predictive value for csPCa and provides great clinical benefits.

Objective To explore the value of dual-energy computed tomography (DECT) quantitative parameters combined with traditional imaging features in predicting the invasiveness of pulmonary adenocarcinoma in pure ground-glass nodules (pGGNs). Methods A total of 124 pGGNs from patients who underwent enhanced chest DECT scans were collected retrospectively, including 64 cases in the minimally invasive adenocarcinoma group and 60 cases in the invasive adenocarcinoma group. Clinical features (gender, age), traditional CT features (diameter, CT value, shape, margin, lobulation, pleural indentation sign, air bronchogram sign, vacuolar sign, and vascular convergence sign), and DECT quantitative parameters [arterial-phase and venous-phase electron density (Rho_A, Rho_V), iodine concentration (IC_A, IC_V), normalized iodine concentration (NIC_A, NIC_V), and spectral curve slope (λ_HU_A, λ_HU_V)] were compared between the two groups. Multivariate Logistic regression was used to construct predictive models, and the diagnostic efficacy of the models was evaluated. Results There were statistically significant differences in margin, pleural indentation sign, and air bronchogram sign between the two groups (all P < 0.05). The age of patients, pGGN diameter, CT value, Rho_A, Rho_V, IC_V, and NIC_V in the invasive adenocarcinoma group were significantly higher than those in the minimally invasive adenocarcinoma group (all P< 0.05). Based on the above features, three models were constructed: the traditional CT model (features: diameter and CT value), the clinical-DECT model (features: Rho_A, NIC_V, and age), and the combined CT model (features: diameter, Rho_A, and NIC_V). The areas under the curve (AUCs) of the combined CT model, clinical-DECT model, and traditional CT model were 0.873 (95%CI: 0.802-0.926), 0.795 (95%CI: 0.714-0.862), and 0.835 (95%CI: 0.758-0.895), respectively. The AUC of the combined CT model was significantly higher than that of the traditional CT model (Z=2.090, P=0.037) and the clinical-DECT model (Z=2.117, P=0.034). Conclusion The combined CT model constructed with diameter, Rho_A, and NIC_V may be helpful in predicting the invasiveness of pulmonary adenocarcinoma in pGGNs.

Objective To investigate the application value of computed tomography (CT)-derived extracellular volume fraction (ECV) in preoperative prediction of benign and malignant gastrointestinal stromal tumors (GISTs). Methods A total of 110 GIST patients confirmed by surgery and pathology from January 2020 to January 2025 were included retrospectively, including 8 cases in the very low-risk group, 30 cases in the low-risk group, 20 cases in the intermediate-risk group, and 52 cases in the high-risk group. Among them, the very low-risk and low-risk groups were classified as the benign group (38 cases in total), and the intermediate-risk and high-risk groups were classified as the malignant group (72 cases in total). All patients underwent multi-phase dynamic enhanced CT scan before surgery. The tumor ECV (%) was measured through plain scan and equilibrium phase images, and clinical and imaging data were collected. Univariate and multivariate Logistic regression analyses were used to identify independent risk factors and construct individual and combined diagnostic models. Receiver operating characteristic (ROC) curves were used to evaluate diagnostic efficacy, and DeLong test was applied to compare differences in area under the curve (AUC). Results Multivariate Logistic regression analysis showed that ECV (%) and maximum tumor diameter were independent predictors of GIST malignancy (both P<0.05). The AUCs of ECV (%), maximum tumor diameter, and the combined model were 0.882 (95%CI: 0.819-0.945), 0.845 (95%CI: 0.765-0.925), and 0.916 (95%CI: 0.862-0.970), respectively. The AUC of the combined model was higher than that of ECV (%) alone. Conclusion The ECV parameter of lesions can effectively quantify the degree of extracellular matrix remodeling in GISTs, significantly improve the preoperative predictive value for benign and malignant gastrointestinal stromal tumors, and provide a non-invasive evaluation index for formulating individualized treatment plans.

Objective To explore the feasibility of constructing a prediction model for mild renal insufficiency based on non-contrast renal computed tomography (CT) radiomics. Methods A retrospective analysis was performed on 693 hospitalized patients who had abdominal non-contrast CT images within 1 week before or after renal function testing. According to the estimated glomerular filtration rate (eGFR), patients were divided into two groups: (1) Normal Renal Function Group: 90 ≤ eGFR < 120 ml/(min·1.73m²); (2) Mild Renal Insufficiency Group: 60 ≤ eGFR < 90 ml/(min·1.73m²). A 3D U-Net deep learning technique was used to train an automatic kidney segmentation model for delineating the region of interest (ROI). LASSO regression analysis was applied to screen features and parameters associated with mild renal insufficiency, and radiomics prediction models were established based on Logistic regression, support vector machine, and decision tree. The efficacy of the models was evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC). The model with the best efficacy was presented as a radiomics nomogram, and the performance of the nomogram in predicting mild renal insufficiency was assessed using calibration curves. Results Among the radiomics prediction models, the one established by logistic regression showed the best efficacy, with AUC values of 0.849 in the training set and 0.782 in the test set. The nomogram of this prediction model exhibited good performance in predicting mild renal insufficiency, as demonstrated by waterfall plots and calibration curves. Conclusion It is feasible to construct a radiomics prediction model for mild renal insufficiency based on non-contrast CT images, which can remind clinicians to timely assess potential renal function abnormalities in patients.

Objective To investigate the predictive value of three-dimensional (3D) CT morphological parameters combined with radiomics features for risk stratification of thymoma. Methods Clinical and CT imaging data of 112 patients with thymoma confirmed by surgery and pathology from four tertiary hospitals in China were retrospectively analyzed. According to postoperative pathology and WHO classification, the patients were divided into the low-risk thymoma group (n=67) and the high-risk thymoma group (n=45). Clinical data and CT images were collected. Whole-tumor regions of interest (ROIs) were delineated using 3D Slicer software. ImageJ software was used to calculate 3D morphological parameters, including volume, surface area, sphericity, ellipsoidness, and compactness. SPSS statistical software was applied to analyze clinical features, imaging features, and 3D morphological features. PyRadiomics software was used to extract radiomics features. Three predictive models were constructed, namely the clinicoradiological-3D morphological model, the radiomics model, and the combined model. The diagnostic performance of each model was evaluated by calculating the area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and Brier score. Results There were statistically significant differences between the low-risk and high-risk thymoma groups in tumor detection mode, tumor margin, longest diameter, volume, sphericity, ellipsoidness, and compactness (all P < 0.05). Among the three predictive models, the clinicoradiological-3D morphological model had an AUC of 0.489, the radiomics model had an AUC of 0.712, and the combined model had an AUC of 0.795. Conclusion The combined model integrating clinicoradiological-3D morphological features and radiomics features shows superior predictive value for thymoma risk stratification compared with the single clinicoradiological-3D morphological model or radiomics model.

Objective To explore the diagnostic value of CT lymphangiography (CTL) and MR lymphangiography (MRL) in lymphatic plastic bronchitis (PB). Methods The clinical and imaging data of 27 patients with clinically confirmed lymphatic plastic bronchitis were analyzed retrospectively. All patients underwent both CTL and MRL examinations. According to the distribution of abnormal lymphatic vessels in the neck and chest on MRL, the disease was classified into four types: TypeⅠ showed tiny abnormal lymphatic vessels in the supraclavicular region and mediastinum; Type Ⅱ showed increased abnormal lymphatic vessels in the supraclavicular region without extension to the mediastinum; Type Ⅲ showed increased abnormal lymphatic vessels in the supraclavicular region with extension to the mediastinum; Type Ⅳshowed abnormal lymphatic vessels in the supraclavicular region with extension to the mediastinum, lung parenchyma, and interstitium. Patients with TypeⅠ and Type Ⅱ were classified into the mild group, while those with Type Ⅲ and Type Ⅳ were classified into the severe group. The CTL imaging findings of each group were recorded, and the CTL imaging indicators included abnormal contrast medium deposition in the lungs, mediastinum, abdominal-pelvic cavity, thoracic duct, and its tributaries. Statistical analysis was performed on the CTL imaging indicators of each group, with a P-value < 0.05 considered statistically significant. Results Based on the range of abnormal lymphatic vessels in the neck and chest on MRL, 27 cases of lymphatic plastic bronchitis were classified into the mild group (10 cases) and the severe group (17 cases). There were no statistically significant differences between the two groups in gender, age of onset, disease course, clinical symptoms, or complicated chylous effusion (all P > 0.05), while there was a statistically significant difference in complicated lymphatic malformation (P = 0.018). The differences in patchy ground-glass opacity, large grid shadow, and thickening of bronchovascular bundles between the two groups were statistically significant (all P < 0.05), with the incidence of these signs higher in the severe group than in the mild group. The differences in abnormal contrast medium deposition around the pericardium, subcarina, pulmonary hilum, and bronchovascular bundles between the two groups were also statistically significant (all P < 0.05), and the incidence was higher in the severe group than in the mild group. Conclusion MRL is helpful for displaying the dilation and range of central lymphatic vessels in the neck and chest, while CTL can show abnormal signs in the lungs as well as the location and degree of systemic lymphatic vessel abnormalities. Both are of great value for the diagnosis and classification of lymphogenic PB, and provide an important imaging basis for the diagnosis and treatment of lymphogenic PB.

Objective To study the correlation between brain functional network characteristics and neurobehavioral development at 1 year of age in high-risk preterm infants based on resting-state functional magnetic resonance imaging (rs-fMRI) and graph theory methods. Methods Preterm infants admitted were selected as research objects. According to the presence of clinical high-risk factors, they were divided into the high-risk group (21 cases) and the low-risk group (14 cases). All subjects underwent rs-fMRI scanning and Neonatal Behavioral Neurological Assessment (NBNA) at a corrected gestational age of 40-44 weeks. Graph theory algorithms were used to compare and analyze differences in brain network parameters, and long-term follow-up was conducted. A total of 24 subjects completed the Gesell Development Scale assessment at 1 year of corrected age to explore the association between early brain network parameters and neurodevelopment at 1 year of age. Results (1) Both groups had small-world topological characteristics, and there were no significant differences in global topological parameters between the two groups (P>0.05). (2) In the high-risk group, the betweenness centrality of nodes such as the left precentral gyrus and right middle frontal gyrus increased; the degree centrality of brain regions such as the right postcentral gyrus and right superior temporal gyrus decreased; the nodal clustering coefficient and nodal local efficiency of the left medial superior frontal gyrus increased; and the nodal efficiency of the right postcentral gyrus decreased (all P<0.05). (3) The connectivity between the prefrontal lobe and frontal lobe in the high-risk group was weaker than that in the low-risk group (P<0.05). (4) The NBNA score of the high-risk group was lower than that of the low-risk group (P<0.05), but there was no correlation between the NBNA score and abnormal nodal indicators (P>0.05). (5) The Gesell Development Scale assessment at 1 year of age showed differences in fine motor and personal-social function scores between the two groups (P<0.05). Correlation analysis revealed that language function was positively correlated with the local efficiency of the left medial superior frontal gyrus node (r=0.405, P=0.049), and the average developmental quotient was positively correlated with the betweenness centrality of the left precentral gyrus (r=0.473, P=0.020). Conclusion High-risk preterm infants have neurobehavioral developmental delays. Rs-fMRI based on graph theory algorithms can detect abnormalities in local topological properties at an early stage. Some abnormal brain network indicators (such as the local efficiency of the left medial superior frontal gyrus and the betweenness centrality of the left precentral gyrus) are positively correlated with language and overall development at 1 year of age, but their predictive value should be interpreted with caution.

Objective To explore the application value of radiomics technology based on contrast-enhanced computed tomography (CECT) habitat in the differential diagnosis of fat-poor renal angiomyolipoma (fp-AML) and small renal cell carcinoma (sRCC). Methods A total of 137 patients with pathologically confirmed small renal masses (tumor diameter ≤ 4 cm) were included in this retrospective study, including 89 cases of sRCC and 48 cases of fp-AML. The K-means clustering algorithm was used to partition the tumor region, generating 2 to 5 different clustered sub-regions. The Calinski-Harabasz score (CH score) was applied to evaluate the clustering level of different cluster numbers, and the optimal cluster number was selected. Radiomics features were extracted based on each sub-region, and a support vector machine (SVM) was used to construct radiomics models for different regions. Univariate and multivariate Logistic regression analyses were performed to screen clinical and CT imaging features, and a clinical model was established. The area under the receiver operating characteristic (ROC) curve (AUC) was used to evaluate model performance, calibration curves were used to verify the calibration ability of the models, and decision curve analysis was used to compare the clinical practical value of the models. Results Independent risk factors for differentiating fp-AML from sRCC included pseudocapsule and enhancement pattern, which were used to construct the clinical model. Cluster analysis showed that the optimal number of clusters was 2. Among all models, the combined model (integrating habitat-based radiomics features and clinical features) had the best performance, with AUC values of 0.941 in the training set and 0.931 in the test set, respectively. The decision curve indicated that the combined model had better clinical practical value than other models, and the calibration curve showed good fitting of the model. Conclusion The radiomics model constructed based on habitat has good efficacy in the differential diagnosis of sRCC and fp-AML. By finely partitioning different heterogeneous sub-regions of the tumor and extracting features, tumor heterogeneity can be better captured.

Objective To construct a predictive model for thymoma risk stratification based on quantitative and qualitative features of preoperative enhanced computed tomography (CT), and to evaluate the model’s efficacy in distinguishing low-risk from high-risk thymoma. Methods Clinical data of 167 thymoma patients who underwent preoperative enhanced CT scans between January 2018 and June 2024 were analyzed retrospectively. According to pathological results and WHO histological classification, the patients were divided into the low-risk group (n=126) and the high-risk group (n=41). The included quantitative features were age, maximum diameter, non-contrast CT value, maximum enhanced CT value, and maximum enhancement rate. The qualitative features included location, morphology, boundary, calcification, cystic necrosis, adjacent organ involvement, mediastinal lymphadenopathy, pleural effusion, and pericardial effusion. Univariate and multivariate Logistic regression analyses were used to screen independent predictors and construct models. Receiver operating characteristic (ROC) curves were used to compare the predictive efficacy of single-feature models and the combined-feature model. Calibration curves and decision curves were applied for the verification of the nomogram model and the evaluation of clinical net benefits, respectively. Results Multivariate Logistic regression analysis showed that calcification (OR=5.863, 95%CI: 1.356-25.338, P=0.018), maximum enhanced CT value (OR=0.930, 95%CI: 0.890-0.972, P=0.001), and maximum enhancement rate (OR=0.826, 95%CI: 0.764-0.894, P<0.001) were independent predictors for diagnosing high-risk thymoma. Single-feature predictive models and a combined-feature model were constructed using these independent predictors. The ROC curve showed that the area under the curve (AUC) of the combined-feature model was 0.958 (95%CI: 0.928-0.988), with a diagnostic sensitivity of 82.93% and specificity of 96.03%. Delong tests among the models indicated that the combined-feature predictive model was superior to single-feature predictive models, and the combined-feature model had high diagnostic efficacy (P<0.05). A nomogram was constructed based on the combined features. The calibration curve showed that the predicted values of the nomogram were close to the actual values, and decision curve analysis demonstrated good net benefits of the model. Conclusion Establishing a combined predictive model based on preoperative enhanced CT features can effectively predict the high-risk of thymoma, providing an accurate imaging basis for formulating individualized surgical plans.

Objective To investigate the predictive factors for plastic bronchitis (PB) complicating consolidation in children with Mycoplasma pneumoniae pneumonia (MPP). Methods A total of 200 children with MPP and consolidation admitted were enrolled retrospectively, including 95 cases in the PB group and 105 cases in the non-PB group. Clinical data, laboratory test results, and chest computed tomography (CT) imaging features of the children were collected. Differences in each indicator between the two groups were compared, a Logistic regression model was established to analyze factors influencing the occurrence of PB, and receiver operating characteristic (ROC) curves were plotted to evaluate the predictive efficacy of these factors for PB. Results The fever duration and levels of neutrophil percentage (N%), C-reactive protein (CRP), alanine aminotransferase (ALT), aspartate aminotransferase (AST), lactate dehydrogenase (LDH), D-dimer (D-D), and procalcitonin (PCT) in the PB group were higher than those in the non-PB group, while the lymphocyte percentage (L%) was lower (all P < 0.05). Compared with the non-PB group, the PB group had significantly more consolidated lung segments, higher mean CT value of consolidation areas, larger maximum cross-sectional area of consolidation areas, and a higher proportion of complicated pleural effusion, but significantly fewer air bronchogram signs in consolidation areas (all P < 0.05). The length of hospital stay in the PB group was significantly longer than that in the non-PB group (P < 0.05). Logistic regression analysis showed that LDH and mean CT value of consolidation areas were risk factors for PB, while the number of air bronchogram signs in consolidation areas was a protective factor. ROC curve analysis indicated that LDH, mean CT value of consolidation areas, and number of air bronchogram signs in consolidation areas could all be used to predict PB formation. The combined use of the three factors had the highest predictive value, with an area under the ROC curve (AUC) of 0.919, a sensitivity of 94.6%, and a specificity of 78.8%. Conclusion The higher the LDH level and mean CT value of consolidation areas, and the fewer the number of air bronchogram signs in consolidation areas, the greater the possibility of complicated PB. The combination of these three factors has the highest predictive efficacy for PB complicating MPP-related consolidation.

Objective This study aims to explore the feasibility of high b-value diffusion-weighted imaging (DWI) based on the deep learning reconstruction (DLR) algorithm in assessing seminal vesicle invasion (SVI) in prostate cancer, so as to improve the diagnostic efficacy of SVI. Methods A total of 200 patients with prostate cancer admitted between May 2022 and July 2024 were enrolled. All patients completed preoperative prostate magnetic resonance imaging (MRI) examinations and underwent radical prostatectomy. The DLR algorithm based on the denoising diffusion probabilistic model (DDPM) was used to construct a high b-value DWI image generation model. By inputting DWI images with b=1400 s/mm², corresponding high b-value DWI and apparent diffusion coefficient (ADC) images were generated. Two radiologists independently evaluated the patients' prostate MRI, adopted the Likert-5 scale to assess SVI, and compared the diagnostic efficacy of conventional DWI and DLR-DWI using multi-reader multi-case receiver operating characteristic (MRMC-ROC) curves and precision-recall (PR) curves. A P-value < 0.05 was considered statistically significant. Results Histopathological analysis showed that among 400 seminal vesicles, 60 (15%) were SVI-positive. At the seminal vesicle level, the area under the curve (AUC) of DLR-DWI for diagnosing SVI was 0.938 (95%CI: 0.910-0.960), which was significantly higher than that of conventional DWI (0.759, 95%CI: 0.714-0.800), P=0.027; the area under the PR curve of DLR-DWI was 0.768 (95%CI: 0.645-0.858), which was significantly higher than that of conventional DWI (0.450, 95%CI: 0.329-0.576), P<0.001. At the patient level, the AUC of DLR-DWI was 0.922 (95%CI: 0.876-0.955), significantly higher than that of conventional DWI (0.678, 95%CI: 0.608-0.742), P=0.027; the area under the PR curve of DLR-DWI was 0.812 (95%CI: 0.680-0.898), significantly higher than that of conventional DWI (0.497, 95%CI: 0.362-0.633), P<0.001. The sensitivity and specificity of DLR-DWI at the patient level were 80% and 97%, respectively; at the seminal vesicle level, they were 84% and 98%, respectively. Conclusion High b-value DWI based on DLR can significantly improve the diagnostic efficacy of SVI in prostate cancer, and is superior to conventional DWI.

Objective To investigate the preoperative diagnostic value of extracellular volume fraction (ECV) based on enhanced computed tomography (CT) in differentiating renal fat-poor angiomyolipoma (AMLmf) from renal clear cell carcinoma (ccRCC). Methods This retrospective study included 87 patients with pathologically confirmed ccRCC and 38 patients with pathologically confirmed AMLmf. Clinical data and conventional imaging features of the patients were collected. The systemic immune-inflammation index (SII) and preoperative tumor ECV of all patients were calculated. Univariate analysis of the two groups of data was performed using independent samples t-test, Mann-Whitney U test, Fisher's exact test, and Chi-Square test. Indicators with statistical significance in univariate analysis were included in multivariate Logistic regression analysis to identify independent clinical predictors for differentiating ccRCC from AMLmf. The area under the receiver operating characteristic (ROC) curve (AUC) was used to evaluate the diagnostic performance of each indicator alone and the combined diagnosis of multiple indicators. Results Univariate analysis showed that ECV, SII, gender, angled margin, cystic necrosis, pseudocapsule, net enhanced CT value of tumor in cortical phase, net enhanced CT value of tumor in parenchymal phase, tumor enhancement rate in cortical phase, and tumor enhancement rate in parenchymal phase had statistical differences in differentiating the two diseases (all P<0.05). Multivariate Logistic regression analysis showed that ECV, SII, angled margin, cystic change, and net enhanced CT value of tumor in cortical phase were independent clinical predictors for differentiating the two diseases (all P<0.05). ROC curve analysis showed that the AUC of the combined diagnosis of the above five indicators was 0.944, which was significantly higher than that of any single indicator. Conclusion ECV, SII, angled margin, cystic necrosis, and net enhanced CT value of tumor in corticomedullary phase based on enhanced CT can accurately differentiate AMLmf from ccRCC, and the diagnostic efficacy of combining these five imaging features is higher.

Objective To study the application value of three-dimensional Sampling Perfection With Application-optimized Contrasts By Using Different Flip Angle Evolutions (3D SPACE) sequence combined with compressed sensing (CS) technology in non-contrast-enhanced portal venous magnetic resonance venography (MRV). Methods Thirty patients underwent non-contrast-enhanced portal venous MRV. The scanning sequences included the conventional T₂-weighted imaging (T₂WI) SPACE sequence combined with Generalized Autocalibrating Partially Parallel Acquisitions (GRAPPA) parallel acquisition technology (conventional group) and the T₂WI SPACE sequence combined with CS technology (CS group). Using a double-blind method, two senior radiologists evaluated the subjective scores of portal vein visualization completeness, vessel edge sharpness, image artifacts, and overall image quality. One radiologist measured the signal intensity (SI) and standard deviation (SD) of the main portal vein and liver at the portal vein level of the scanned images, and objectively evaluated the image quality by calculating the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the portal vein. Results Under the condition that the image quality fully met the requirements of clinical diagnosis, the two radiologists had good consistency in scoring the images of the two groups [intraclass correlation coefficient (ICC) > 0.75], and the overall scores of the two groups were both > 3 points. In terms of subjective scores, the portal vein visualization completeness [(4.74 ± 0.63) points vs. (3.56 ± 0.84) points], vessel edge sharpness [(4.83 ± 0.67) points vs. (3.78 ± 0.74) points], and overall image quality score [(4.55 ± 0.63_ points vs. (3.52 ± 0.76) points] of the conventional group were all superior to those of the CS group (all P < 0.05). There was no statistical difference in the evaluation of image artifacts between the two groups (P > 0.05) (see Table 2 for details). In terms of objective evaluation, the SNR (365.50 vs. 174.00) and CNR (311.50 vs. 125.50) of the conventional group were both superior to those of the CS group. However, the scanning time of the CS group was 130 seconds, which was 52% shorter than the 270 seconds of the conventional group. Conclusion On the premise of ensuring image quality and meeting clinical diagnosis requirements, the scanning time of the T₂WI SPACE sequence combined with CS technology is 52% shorter than that of the conventional T₂WI SPACE sequence combined with GRAPPA parallel acquisition technology, providing a more efficient examination protocol for clinical examinations.

Objective To investigate the value of diffusion tensor imaging (DTI) combined with arterial spin labeling (ASL) sequence in evaluating renal function impairment in early chronic kidney disease (CKD). Methods A total of 59 CKD patients were selected from Zhangjiagang Traditional Chinese Medicine Hospital and the First Affiliated Hospital of Soochow University between January 2024 and January 2025, and 22 healthy volunteers were recruited as the control group. All patients underwent renal biopsy and were diagnosed with CKD. Serum creatinine (SCr) and 24-hour urinary protein (24 h-UPRO) were measured, and the estimated glomerular filtration rate (eGFR) was calculated using the CKD-EPI formula. Patients were divided into the normal eGFR group (eGFR ≥ 90 ml/min/1.73m², n=36) and the abnormal eGFR group (eGFR < 90 ml/min/1.73m², n=23). All subjects underwent abdominal routine MRI, DTI, and ASL examinations. Bilateral renal cortical renal blood flow (RBF), corticomedullary apparent diffusion coefficient (ADC), and corticomedullary fractional anisotropy (FA) values were measured respectively. One-way analysis of variance (ANOVA), nonparametric tests, and Chi-Square tests were used to compare differences in general clinical data among the healthy control group, normal eGFR group, and abnormal eGFR group. Intraclass correlation coefficient (ICC) was used to compare the consistency of MRI parameters measured by two diagnostic physicians. Paired sample t-test was used to compare differences in MRI parameters between the two kidneys. One-way ANOVA was used to compare differences in MRI parameters among the three groups. Pearson correlation analysis was used to study the correlation between each MRI parameter and clinical markers (eGFR, SCr, 24 h-UPRO). Finally, binary Logistic regression and ROC curve analysis were used to evaluate ASL and DTI parameters individually and synergistically, and the most sensitive imaging indicators were obtained. Results Cortical RBF values, renal medullary FA and ADC values, and renal cortical FA and ADC values showed significant differences among the three groups (all P < 0.05). Renal cortical RBF, renal medullary FA, and renal cortical ADC were positively correlated with eGFR; renal cortical FA value was negatively correlated with eGFR. The AUCs of cortical FA, RBF, and combined DTI and ASL imaging indicators for distinguishing the control group from the normal eGFR group were 0.774, 0.836, and 0.943, respectively. Conclusion The combined application of DTI and ASL parameters has high diagnostic performance for early CKD, with higher sensitivity and specificity than combined clinical biochemical indicators.

Objective To explore the value of magnetic resonance (MR) continuous-time random walk (CTRW) diffusion model and diffusion-weighted imaging (DWI) in assessing Ki-67 expression level in rectal cancer. Methods CTRW and DWI data of 23 rectal cancer patients in the Ki-67 low-expression group (≤ 50.00%) and 37 patients in the Ki-67 high-expression group (> 50.00%) were collected. The temporal heterogeneity index (α), spatial heterogeneity index (β), diffusion index (Dm), and apparent diffusion coefficient (ADC) values were measured and compared between the two groups. The area under the receiver operating characteristic curve (AUC) and Delong test were used to evaluate diagnostic performance. Logistic regression analysis and Pearson correlation coefficient were applied for multi-parameter combined diagnosis and evaluation of the correlation between each parameter and Ki-67 expression level, respectively. Results For parameter comparison, the α, β, Dm, and ADC values in the Ki-67 high-expression group were all lower than those in the Ki-67 low-expression group. In terms of diagnostic performance, among single parameters, β had the highest diagnostic efficacy with an AUC of 0.874, sensitivity of 70.27%, and specificity of 91.30%. Among different imaging methods, CTRW (combining α, β, and Dm) had the highest diagnostic efficacy with an AUC of 0.954, sensitivity of 86.49%, and specificity of 95.65%. The differences in AUC between CTRW and DWI (ADC), α, and Dm were statistically significant (AUC = 0.801, 0.840, 0.838 respectively; Z = 2.367, 2.334, 2.496 respectively; all P < 0.05). For correlation analysis, Ki-67 expression level was mildly negatively correlated with α, β, Dm, and ADC values (r=-0.408,-0.460,-0.472, -0.527 respectively; all P < 0.05). Conclusion Both CTRW and DWI can be used to assess Ki-67 expression level in rectal cancer, and the former has higher diagnostic performance.

Objective Assessment of abnormal cortical development in preterm infants with developmental language disorder (DLD) using structural magnetic resonance imaging. Methods The study followed up on 39 premature infants who were admitted to the neonatal ward /NICU in our hospital from January 2021 to March 2023,and followed up in the high-risk pediatric clinic until 12 months of corrected gestational age.They were stratified into a DLD (Developmental Language Disorder) group（20 cases） and a normal language development group（19 cases）.Whole brain 3D-T₁WI MRI was performed using a 3.0 T magnetic resonance examination equipment at corrected gestational age of 40-42 weeks.Image processing and cortical surface reconstruction were conducted using United Imaging's segmentation model and Infant Freesurfer image analysis software based on brain surface morphology.Cortical thickness(CT) and cortical surface area (SA) of all regions in the brain were calculated (68 regions in total).SPSS 25.0 statistical software was used to input and analyze the data. Results (1) Comparison of CT values between the two cerebral hemispheres showed that the brain regions with significant asymmetry in DLD group were isthmus of anterior cingulate gyrus,isthmus gyrus,inferior temporal lobe,lateral occipital lobe,lateral orbitofrontal lobe,lingual gyrus,collateral hippocampus,central limbic gyrus,posterior central gyrus,posterior cingulate gyrus,anterior cuneiform gyrus,anterior cingulate gyrus,and superior marginal gyrus (P<0.05).The brain regions with significant asymmetry in normal group were isthmus of anterior cingulate gyrus,entolfactory,inferior parietal,inferior temporal,isthmus of cingulate gyrus,lateral occipital lobe,lateral orbitofrontal lobe,lingual gyrus,accessory hippocampus,central limbic gyrus,inferior frontal gyrus tegmental,prefrontal gyrus,posterior central gyrus,anterior cingulate gyrus,superior limbic gyrus and temporal pole (P<0.05).(2) Comparison of SA values in both cerebral hemispheres showed that In DLD group,the brain regions with significant asymmetry were isthmus anterior cingulate gyrus,caudal central frontal gyrus,inferior parietal lobe,isthmus anterior cingulate gyrus,lateral orbitofrontal lobe,lingual gyrus,medial orbitofrontal lobe,middle temporal lobe,collateral hippocampus,central limbic gyrus,caput anterior gyrus,ascending branch of inferior frontal gyrus,triangular part of inferior frontal gyrus,posterior cingulate gyrus,superior parietal gyrus,frontal pole,transverse temporal gyrus,insula (P<0.05); In the normal group,the brain regions with significant asymmetry were isthmus of anterior cingulate gyrus,caudal gyrus of central frontal gyrus,cuneiform gyrus,inferior parietal lobe,isthmus of cingulate gyrus,lateral orbitofrontal lobe,medial orbitofrontal lobe,middle temporal lobe,collateral hippocampus,central limbic gyrus,tegmental branch of inferior frontal gyrus,triangular part of inferior frontal gyrus,posterior central gyrus,posterior cingulate gyrus,anterior cingulate gyrus,superior temporal gyrus,frontal pole,transverse temporal gyrus,and insula (P<0.05). Conclusion sDLD preterm infants exhibited a reduced number of hemispherical asymmetrical areas on CT and SA compared to the normal group,indicating diminished left-sided development.These findings suggest that decreased left-sided dysplasia holds predictive value for the language development of preterm infants with DLD.Therefore,it is crucial to pay attention to preterm infants displaying reduced left-sided dysplasia and promptly initiate early intervention measures in order to enhance their quality of life.

Objective To assess the impact of ultra-high-resolution photon-counting detector CT (UHR PCD-CT) combined with ultra-low-dose protocols on detection rate, image quality, and volume measurement accuracy of pulmonary nodules across varying sizes and types. Methods A thoracic phantom containing simulated pulmonary nodules was scanned on a UHR PCD-CT system at four dose levels (CTDIvol: 9.05, 1.00, 0.07, and 0.03 mGy). All images were reconstructed with 0.2 mm and 1.0 mm slice thicknesses, a 1024×1024 matrix, a BI64 convolution kernel, and Quantum Iterative Reconstruction (QIR) at Level 3. Quantitative analysis included nodule detection rate, image noise (standard deviation, SD), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). Nodule volumes were measured using dedicated AI-based software, with absolute percentage error (APE_volume) calculated. Subjective image quality was scored on a 5-point scale using images acquired at 9.05 mGy as the reference standard. Results With decreasing dose, nodule detection rates decreased while APE_volume increased. Solid nodules (SNs) exhibited higher detection rates than ground-glass nodules (GGNs), and nodules ≥8 mm demonstrated superior detection rates and volume measurement accuracy compared with smaller nodules. For both nodule types, SD increased with decreasing dose, whereas SNR and CNR decreased; the reduction in CNR was more pronounced in nodules ≥8 mm (all P < 0.05). Subjective analysis indicated that image quality at 1.00 mGy was comparable to the reference dose of 9.05 mGy, and images acquired at 0.07 mGy remained diagnostically acceptable (all P < 0.05). Conclusion Even at ultra-low doses (0.03-0.07 mGy), UHR PCD-CT combined with AI-assisted diagnostic software maintains clinically acceptable detection rates for pulmonary nodules (particularly SNs and those ≥5 mm in diameter) and diagnostic image quality, demonstrating promising clinical potential.

Objective This study aimed to evaluate the effectiveness of a medical domain-specific large language model (LLM) in detecting errors in radiology report texts. Methods A total of 200 standardized radiology reports were retrospectively collected from Harbin Medical University Cancer Hospital between December 2024 and January 2025. Among them, 100 reports were randomly selected, and 150 textual errors were manually inserted by a radiology resident according to five predefined error categories, with clinical risk stratification. A medical domain-specific LLM (MedGemma), a general-purpose LLM (Qwen3), and three radiologists with different seniority levels simulated real-world report auditing processes and independently reviewed all reports within a 2-hour time limit. Error detection rate and F1 score were the primary evaluation metrics. Wald χ² test was used to compare the performance across LLMs and radiologists, and the Wilson method was applied to calculate the 95% confidence intervals. Reviewing efficiency was assessed by average reading time, with paired t-tests and Cohen’s d used to analyze statistical significance and effect size. Results The MedGemma model achieved an error detection rate of 91.33% (95%CI:85.74%-94.87%), comparable to that of the associate chief radiologist (P=0.83), and significantly outperformed the attending radiologist (80.67%,95%CI:73.61%-86.19%), the resident (76.67%,95%CI:69.28%-82.72%), and Qwen3 (84.67%,95%CI:78.04%-89.56%). For high-risk errors, MedGemma also demonstrated superior performance (92.7%,95%CI:80.6%-97.5%).Moreover,the average reading time for MedGemma was (17.54 ± 3.47) seconds, significantly faster than all other groups (P<0.001), with a large effect size (Cohen's d: |1.33-2.40|). Conclusion The medical-specific LLM MedGemma exhibits high accuracy in error detection, strong capability in identifying high-risk errors, and remarkable efficiency in report review. These results suggest its potential as a valuable clinical tool for assisting radiology report quality control.

Objective Establish a model based on radiomics combined with machine learning to predict the short-term prognosis of patients with spontaneous cerebral hemorrhage and validate the model. Methods This study retrospectively collected imaging and clinical data from 573 patients with spontaneous cerebral hemorrhage at the Affiliated Hospital of Jiujiang University,forming a training set; And the data of 100 patients with spontaneous cerebral hemorrhage from the First People's Hospital of Nanning were used as the external validation set.Based on LASSO algorithm for variable dimensionality reduction,10 fold cross validation is used to optimize its regularization parameters.Build a prediction model using the selected optimal parameters.Subsequently,a multi factor Logistic regression model was fitted onto the training set to establish a joint prediction model,and the Area Under the Receiver Operating Characteristic Curve (AUC) was calculated to evaluate the discriminative performance of the model. Results By comparing the performance of clinical models,radiomics models,and joint models,the sensitivity,specificity,and AUC of the clinical model in the training set and external validation set were 0.710,0.792,0.828,and 0.526,0.857,0.706; The sensitivity,specificity,and AUC of the radiomics model in the training set and external validation set were 0.713,0.785,0.832,and 0.772,0.612,0.717; The sensitivity,specificity,and AUC of the joint model in the training set and external validation set were 0.710,0.846,0.860,and 0.579,0.857,and 0.725.Overall,the joint model performs the best.The model evaluation results show that the calibration curve verifies the good predictive accuracy of the joint model; Meanwhile,clinical decision curve analysis confirms that the model has ideal clinical net benefits. Conclusion This study provides an effective personalized prediction tool for the short-term prognosis risk assessment of patients with spontaneous cerebral hemorrhage by integrating radiomics features and clinical baseline data,and constructing a column chart using machine learning methods.

Objective To evaluate the value of non-contrast 3D-FLAIR sequence in 5.0 T MRI for the diagnosis of endolymphatic hydrops. Methods 25 patients（12 males and 13 females;mean age 57.42±12.66 years）with clinically confirmed Ménière’s disease were prospectively included,and all underwent 5.0 T gadolinium magnetic resonance imaging of the inner ear.The scanning sequence included:conventional inner ear MRI scan,internal auditory canal water imaging and non-contrast 3D-FLAIR scan;Four hours after double dose contrast injection,enhanced 3D-FLAIR and thin T₁WI fat suppression sequences were performed.Two radiologists with 10 years of experience read the non-enhanced 3D-FLAIR and enhanced 3D-FLAIR sequences independently,the presence or absence of endolymphatic hydrops in the cochlea and vestibule was independently evaluated,if there is any difference,they should reach an agreement through negotiation. Results Using the diagnosis of the enhanced 3D-FLAIR sequence as the reference standard,endolymphatic hydrops of varying locations and degrees were detected in 25 patients,4 cases of endolymphatic hydrops combined with vestibular neuritis,and 1 of these 4 cases also had labyrinthitis.Among the 50 ears of the 25 patients,19 left ears had cochlear endolymphatic hydrops,and 18 right ears had cochlear endolymphatic hydrops.Endolymphatic hydrops was present in 9 left vestibules and 6 right vestibules.Vestibular endolymphatic hydrops in 15 ears could be observed with the same imaging findings on non-contrast 3D-FLAIR,and 34 of the 37 cochlear endolymphatic hydrops cases could be observed on the non-contrast sequence.The diagnostic consistency of non-contrast 3D-FLAIR for vestibular hydrops was 100%,for cochlear hydrops was 91.9%（34/37).The McNemar exact test P value of the two methods for the diagnosis of endolymphatic hydrops was greater than 0.05,and the difference was not statistically significant. Conclusion 5.0 T non-enhanced 3D-FLAIR has high accuracy in the diagnosis of endolymphatic hydrops,and can be used in Ménière patients who cannot tolerate enhanced examination.