Objective This study aimed to investigate the neuroanatomical substrates underlying cognitive heterogeneity in adolescents with major depressive disorder(MDD) by comparing brain structural differences between two cognitive subtypes,analyzing between-group differences using voxel-based morphometry(VBM) and surface-based morphometry(SBM),and examining the relationships between these structural alterations and cognitive performance. Methods A total of 85 adolescents with MDD and 30 healthy controls were recruited.All participants underwent magnetic resonance imaging(MRI) and neuropsychological assessments.The MDD group was divided into a cognitively impaired subtype(MDD-Ⅰ) and a relatively cognitively preserved subtype(MDD-Ⅱ) using K-means clustering.VBM and SBM were used to analyze between-group brain structural differences,and correlation analyses were performed to examine associations between brain features and clinical indices. Results Brain structural analyses showed that the MDD-Ⅰ subtype exhibited more extensive reductions in gray matter volume and cortical thinning,mainly involving the prefrontal cortex,insula,temporal lobe,and postcentral gyrus,whereas the MDD-Ⅱ subtype showed relatively mild and limited structural alterations.Partial correlation analyses revealed that cortical thickness in the bilateral postcentral gyrus was negatively correlated with inhibitory control in MDD-Ⅰ,while gray matter volume in the left middle frontal gyrus was positively correlated with short-term memory in MDD-Ⅱ. Conclusion This study revealed distinct brain structural characteristics across cognitive subtypes of adolescent MDD,supporting the possibility that cognitive heterogeneity may have corresponding neuroanatomical correlates.Multimodal morphological findings provide complementary evidence from different dimensions and may offer preliminary clues for future subtype assessment and individualized intervention.

Objective To compare water-referenced relative levels of gamma-aminobutyric acid plus co-edited macromolecular signals(GABA+) and glutamate-glutamine complex(Glx) in the medial prefrontal cortex between participants with cerebral small vessel disease(CSVD)-related high white matter hyperintensity(WMH) burden and controls,and to examine their associations with WMH burden. Methods This prospective study enrolled 30 participants with CSVD-related high WMH burden and 25 demographically matched controls.GABA+ and Glx signals in the medial prefrontal cortex were acquired using Mescher-Garwood point-resolved spectroscopy(MEGA-PRESS).Voxel-based morphometry(VBM) was performed on three-dimensional T₁-weighted imaging(3D-T₁WI) structural images,and WMHs were automatically segmented on fluid-attenuated inversion recovery(FLAIR) images using the Lesion Segmentation Tool.Brain tissue volumes,WMH volume,and metabolite levels were compared between groups;analysis of covariance was used to adjust for potential confounders.In the high-WMH-burden group,partial correlation analyses were performed to examine the associations of WMH volume with GABA+ and Glx levels after controlling for age,sex,education,and the proportion of gray matter within the volume of interest relative to total gray and white matter volume[VOI-GM/(GM+WM)]. Results The water-referenced relative levels of GABA+ and Glx were lower in the high-WMH-burden group than in the control group[GABA+:（1.68±0.40）IU vs. （2.52±0.97）IU,P<0.001;Glx:（8.47±1.76）IU vs. （10.46±3.78）IU,P=0.013].WMH volume was greater in the high-WMH-burden group(P=0.007);however,after FDR correction,no significant between-group clusters were found in whole-brain VBM analysis.In the high-WMH-burden group,WMH volume was moderately negatively correlated with GABA+ levels(partial r=-0.476,95%CI：-0.729 to -0.108,P=0.014) but was not significantly correlated with Glx levels. Conclusion In the high-WMH-burden CSVD group,GABA+ levels in the medial prefrontal cortex were significantly lower than in the control group.WMH volume was negatively correlated with GABA+ but not significantly associated with Glx,suggesting that GABAergic metabolic abnormalities may occur with increasing WMH burden.

Objective To investigate the predictive value of hippocampal subfield volumes and diffusion tensor imaging(DTI) metrics for the progression from subjective cognitive decline(SCD) to mild cognitive impairment(MCI). Methods This longitudinal study included 207 patients with SCD enrolled from the Alzheimer's Disease Neuroimaging Initiative(ADNI) database between March 2011 and October 2015.Baseline information and T₁-weighted images were collected.Patients were divided into the SCD group(n=151) and the MCI group(n=56) according to whether they progressed to MCI.All patients underwent structural and diffusion MRI,and baseline hippocampal subfield volumes and white matter DTI metrics were extracted.Baseline characteristics,hippocampal subfield volumes,and white matter DTI metrics were compared between the two groups.Least absolute shrinkage and selection operator(LASSO) regression was used to screen feature variables.Logistic regression(LR),support vector machine(SVM),and Gaussian naive Bayes(GNB) were used to construct predictive models for MCI progression. Results At the end of follow-up,the MMSE score in the MCI group was significantly lower than that in the SCD group(t=11.827,P<0.001).At baseline,the MCI group showed significant volume atrophy in hippocampal subfields,mainly including the subiculum,CA1,and molecular layer,compared with the SCD group.In addition,DTI metrics were significantly altered in white matter regions such as the corpus callosum,internal capsule,corona radiata,external capsule,and superior longitudinal fasciculus(all P<0.05).Among the three predictive models,the SVM model had the largest area under the curve(AUC) and higher clinical net benefit. Conclusion Compared with SCD patients with preserved cognition,those with cognitive deterioration exhibit specific alterations in hippocampal subfield volumes and white matter fiber DTI metrics.The predictive model for the progression from SCD to MCI constructed based on these features has high potential for clinical application.

Objective To investigate the evolution of white matter hyperintensities(WMHs) in maintenance hemodialysis(MHD) patients with different dialysis vintages and their relationships with cerebral perfusion and cognitive function. Methods Forty-three MHD patients were enrolled and divided by dialysis vintage into a short-term group(6-60 months,n=23) and a long-term group(≥60 months,n=20).Twenty-two demographically matched healthy volunteers were included as the control group.All participants underwent cranial MRI to measure global,gray matter,and white matter cerebral blood flow(CBF).WMH severity was assessed using the Fazekas scale,and cognitive function was evaluated using the Mini-Mental State Examination(MMSE) and Montreal Cognitive Assessment(MoCA).Ordered Logistic regression was used to analyze the independent effect of dialysis vintage on WMH grade,and correlation analyses were performed to examine the relationships among WMHs,CBF,and cognitive function. Results Fazekas scores were significantly higher in both the short-term and long-term groups than in the control group(all P<0.05).Ordered Logistic regression confirmed that dialysis vintage was an independent risk factor for WMH progression(OR=1.011,P=0.034).Global,gray matter,and white matter CBF were compensatorily elevated in the short-term group(all P<0.05),but returned to normal levels in the long-term group.Correlation analysis revealed a significant negative correlation between Fazekas score and white matter CBF(r=-0.326,P=0.033).MoCA scores in the long-term group were significantly lower than those in the control and short-term groups(all P<0.05).Although cognitive function did not show significant decline in the short-term group,WMH scores were already significantly higher than those in the control group(P<0.05). Conclusion Dialysis vintage is independently associated with higher WMH grades in MHD patients.The exhaustion of cerebral perfusion compensatory mechanisms and the cumulative effect of WMHs associated with long-term dialysis may jointly contribute to progressive cognitive deterioration.

Objective To investigate the longitudinal evolution of brain network topology and cognitive function during chemotherapy in patients with breast cancer with different δ-catenin expression levels. Methods This prospective longitudinal study included 73 patients with breast cancer scheduled to receive doxorubicin plus cyclophosphamide (AC) chemotherapy.According to immunohistochemical assessment of δ-catenin expression,the patients were classified into a δ-catenin high-expression group(DH,n=35) and a δ-catenin low-expression group(DL,n=38).All participants underwent neuropsychological assessments and resting-state functional magnetic resonance imaging (rs-fMRI) at four time points:before chemotherapy (Cycle 0,C0),after the second cycle (Cycle 1,C1),after the third cycle (Cycle 2,C2),and after the fourth cycle (Cycle 3,C3).The rs-fMRI data were processed using the DPARSFA and GRETNA toolboxes in MATLAB 2013b,and brain network topology metrics were calculated.Within each group,repeated-measures analysis of variance was performed with chemotherapy stage as the within-subject factor to assess longitudinal changes in each metric.Spearman rank correlation analysis was used to examine the associations between brain network metrics and neuropsychological scores. Results After four cycles of chemotherapy,both groups showed significant decreases in Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) scores compared with baseline.In the DH group,impairments in short-term and long-term memory emerged as early as C1,whereas short-term memory declined after C2 and delayed memory declined after C3 in the DL group.Brain network topology:Multiple brain regions in the DH group showed significant longitudinal changes in nodal efficiency,degree centrality,and other topological metrics during chemotherapy,involving multiple brain regions and exhibiting temporal evolution across chemotherapy cycles.The DL group also demonstrated longitudinal changes in several brain network topology metrics,with significant alterations in some metrics emerging during the middle-to-late stages of chemotherapy.Correlation analysis:The DH group showed widespread associations between brain network metrics and cognitive scores as early as the early stage of chemotherapy,with these associations evolving dynamically across chemotherapy cycles.The DL group also exhibited associations to some extent,mainly involving memory-related measures,including AVLT-short,AVLT-long,and AVLT-delay. Conclusion Breast cancer patients with different δ-catenin expression levels exhibit distinct dynamic patterns of cognitive function and brain network topology during chemotherapy.δ-Catenin expression may be associated with susceptibility to chemotherapy-related cognitive impairment (CRCI) and patterns of brain network response,dynamic changes in graph-theoretical metrics of resting-state brain networks can objective ly reflect this pathological process,providing potential imaging clues for CRCI risk stratification.

Objective To investigate the predictive value of longitudinal changes in body composition parameters derived from quantitative computed tomography(QCT) for pathological complete response (pCR) after neoadjuvant chemotherapy(NAC) in breast cancer. Methods A retrospective analysis was conducted on the clinical and pathological data of 95 patients with histopathologically confirmed breast cancer who received NAC and subsequent surgery at our hospital.Patients were categorized into pCR and non-pCR groups according to the Miller-Payne system.All patients underwent QCT scans before and after NAC.Bone mineral density(BMD) was measured at the L₁ and L₂ vertebral levels,and visceral fat area(VFA) and skeletal muscle area(SMA) were assessed at the L₃ vertebral level.The rates of change for these parameters(ΔBMD,ΔVFA,and ΔSMA) were calculated.Independent predictors of pCR after NAC were identified using univariate and multivariate logistic regression analyses,and a combined prediction model and nomogram were constructed accordingly.Model performance and clinical net benefit were evaluated using receiver operating characteristic(ROC) curves,the DeLong test,Bootstrap internal validation,the Hosmer-Lemeshow test,calibration curves,and decision curve analysis (DCA) based on 10 repeated 5-fold cross-validation. Results Univariate analysis showed that ΔVFA,ΔSMA,progesterone receptor (PR) status,human epidermal growth factor receptor 2 (HER-2) status,and axillary lymph node metastasis status were associated with pCR (all P<0.05).Multivariate logistic regression analysis revealed that ΔVFA,ΔSMA,and axillary lymph node metastasis status were independent predictors of pCR (all P<0.05).The combined prediction model based on these three variables achieved an area under the curve (AUC) of 0.896 (95% CI：0.816~0.949),with a sensitivity of 81.58% and a specificity of 82.46%.After optimism correction,internal validation yielded an AUC of 0.881 (95% CI：0.825~0.945).The Hosmer-Lemeshow test yielded P=0.251,and the calibration curve demonstrated satisfactory calibration.DCA indicated that the combined model provided favorable clinical net benefit. Conclusion The combined model incorporating ΔVFA,ΔSMA,and axillary lymph node metastasis status demonstrates favorable predictive efficacy for pCR after NAC in breast cancer.QCT-based body composition analysis may provide valuable information for evaluating NAC response.

Objective To investigate the clinical value of deep learning-based artificial intelligence（AI） technology in assessing the degree of coronary artery stenosis and plaque calcification detected by dual-source CT-Flash scanning. Methods The imaging data of 153 patients with suspected coronary heart disease admitted between June 2023 and June 2025 were retrospectively analyzed.All patients underwent coronary angiography（CAG） and dual-source CT-Flash scanning.CAG and intravascular ultrasound（IVUS） findings were used as the gold standard,and the consistency of AI-processed and manually processed image results with the gold standard was compared. Results A total of 459 vessels were included,of which 282（61.44%） had moderate or greater stenosis;315 plaques were detected.AI processing time was significantly shorter than manual processing time[（9.24±2.61）min vs. （19.37±5.49）min,P<0.001].AI assessment of stenosis severity showed good agreement with CAG findings（Kappa=0.863）,superior to manual assessment（Kappa=0.702）.The accuracy of AI in diagnosing moderate and severe stenosis was 93.97% and 92.91%,respectively,both higher than that of manual assessment,while the accuracy of the two methods in diagnosing occlusion was comparable.In plaque classification,AI showed good agreement with IVUS findings（Kappa=0.762）,superior to manual assessment（Kappa=0.681）,and the accuracy of AI in diagnosing calcified plaques was 90.16%. Conclusion Deep learning-based dual-source CT-Flash scanning can significantly improve the diagnostic accuracy of coronary artery stenosis and plaque calcification and has high clinical value.

Objective To investigate the value of a predictive model combining CT radiomics and clinical features for identifying IASLC grade 3 high-risk pathology in ground-glass nodule(GGN)-type lung adenocarcinoma. Methods A total of 314 patients with surgically and pathologically confirmed GGN-type lung adenocarcinoma were retrospectively included and randomly divided into training and test cohorts at a ratio of 7∶3.Univariate and multivariate Logistic regression analyses were performed to screen independent clinical predictors and construct a clinical model.CT radiomic features were extracted using PyRadiomics.After dimensionality reduction using the maximum relevance and minimum redundancy(mRMR) algorithm and the least absolute shrinkage and selection operator(LASSO),12 machine learning algorithms were compared for performance,and the optimal algorithm was selected to construct the radiomic model.SHapley Additive exPlanations(SHAP) was used to visualize feature weights,and the radiomic score(Radscore) was calculated.A combined clinical-radiomic model was constructed by integrating the clinical model and Radscore using Logistic regression,and a nomogram was generated.Model performance was evaluated using receiver operating characteristic(ROC) curves,calibration curves,and decision curve analysis(DCA).The DeLong test was used to compare differences in the area under the curve(AUC) among models. Results Multivariate analysis showed that spiculation,lobulation,and vascular convergence signs were independent predictors of high-risk grading in GGN-type lung adenocarcinoma(P<0.05).The AUCs of the clinical model were 0.850 in the training cohort and 0.796 in the test cohort.The radiomic model constructed using XGBoost showed the best performance,with AUCs of 0.910 and 0.814 in the training and test cohorts,respectively.The AUCs of the combined clinical-radiomic model were 0.915 in the training cohort and 0.864 in the test cohort.The DeLong test showed a statistically significant difference between the combined clinical-radiomic model and the clinical model in the test cohort(P<0.05),while no significant difference was found between the combined model and the radiomic model.Calibration curves and DCA demonstrated good calibration ability and clinical net benefit. Conclusion The combined model based on clinical features and CT radiomics can be used to predict high-risk pathological features of GGN-type lung adenocarcinoma and may help guide clinical decision-making.

Objective To investigate the value of diffusion kurtosis imaging (DKI) and three-compartment restriction spectrum imaging(RSI) in diagnosing lymph node metastasis(LNM) in lung adenocarcinoma,and to compare with conventional diffusion-weighted imaging (DWI)and18F-FDG PET. Methods A total of 76 patients with surgically and pathologically confirmed lung adenocarcinoma were retrospectively analyzed,all of whom underwent preoperative 18F-FDG PET/MR.According to postoperative pathology,patients were divided into LNM-positive(n=28) and LNM-negative(n=48)groups.Clinicopathological characteristics,DKI,RSI,DWI and PET parameters were compared between groups to evaluate the value of each single parameter and their combination in predicting LNM status.Intraclass correlation coefficient was used to assess inter-observer agreement.Inter-group comparisons were performed using independent-sample t-test or Mann-Whitney U test.Logistic regression was applied to identify independent predictors and construct a combined model.Receiver operating characteristic curve was used to evaluate the diagnostic performance. Results The maximum diameter,f₃,SUV_max and MK were significantly higher in the LNM-positive group than in the negative group,while ADC and f₁ were lower (all P<0.05);MD and f₂ showed no significant differences between groups.Among single parameters,the areas under the curve (AUC) of MK and SUV_max were 0.804 and 0.824,respectively.Multivariate Logistic regression revealed that maximum diameter,f₃,MK and SUV_max were independent predictors(all P<0.05).The combined model (maximum diameter +f₃+MK+SUV_max) achieved an AUC of0.961,superior to each single parameter. Conclusion DKI parameter MK,RSI parameter f₃ ,PET parameter SUV_max and tumor maximum diameter are independent predictors of LNM in lung adenocarcinoma;the combined model can improve preoperative prediction and provide a reliable imaging basis for clinical lymph node staging.

Objective To investigate the value of left atrial strain combined with epicardial adipose tissue (EAT) in identifying myocardial fibrosis in obese patients using cardiac magnetic resonance (CMR). Methods A prospective study included 100 obese patients who underwent CMR examination at the First Affiliated Hospital of Wannan Medical University between June 2023 and December 2025.Patients were divided based on late gadolinium enhancement (LGE) positivity into LGE-positive group(n=43) and LGE-negative group(n=57).Additionally,100 healthy volunteers with normal body mass index (BMI)during the same period were recruited as controls.All subjects received standard CMR scans.Left atrial strain parameters and epicardial adipose tissue volume (EATV) were measured using CVI42 post-processing software.Spearman correlation analysis assessed the relationship between EATV and left atrial strain parameters.Multivariate Logistic regression identified independent factors associated with myocardial fibrosis.Receiver operating characteristic (ROC) curve analysis and DeLong test evaluated the discriminative ability of each marker.Intraclass correlation coefficient (ICC) was used to assess inter- and intra-observer agreement. Results Compared to the control group,the LGE-positive group showed significantly decreased left atrial strain parameters and markedly increased EATV(all P<0.05).Compared to the LGE-negative group,the LGE-positive group exhibited significant reductions in Es,Ee,Ea,SRs,and SRe,with EATV significantly increased (all P<0.05);however,SRa showed no significant difference(P>0.05).Spearman analysis revealed that EATV was negatively correlated with Es (ρ=-0.240) and Ee (ρ=-0.226),and positively correlated with SRe (ρ=0.246),with all P<0.05.Multivariate Logistic regression indicated that EATV(odds ratio (OR)=1.05,95% confidence interval (CI):1.03-1.07),Ee (OR=0.90,95% CI:0.84-0.97),and left ventricular mass (LV mass)(OR=1.02, 95% CI:1.00-1.03) were independent predictors of myocardial fibrosis in obese patients(all P<0.05).The ROC curve combining EATV and Ee yielded an area under the curve(AUC) of 0.907,which was significantly better than EATV alone(Z=2.004,P=0.045) and Ee alone (Z=11.583,P<0.001) by DeLong test.The intra-observer ICC ranged from 0.873 to 0.960,and the inter-observer ICC ranged from 0.840 to 0.934, both greater than 0.75. Conclusion The combination of left atrial strain parameters and EATV demonstrates excellent performance in identifying myocardial fibrosis in obese patients,providing a non-invasive imaging basis for risk stratification.

Objective To investigate the value of a combined model based on MR arterial phase intratumoral habitat imaging and peritumoral radiomics for preoperative prediction of P53-mutated hepatocellular carcinoma(HCC). Methods Clinical and imaging data of HCC patients with tumor diameter ≤5 cm were retrospectively enrolled.Based on preoperative MR arterial phase images,habitat imaging technology was employed to segment the tumor region into four subregions for intratumoral habitat analysis,while peritumoral radiomics analysis was performed with 5 mm and 7 mm expansions from the tumor boundary.Radiomics features were extracted and selected,following which multiple machine learning classifiers combined with 10-fold cross-validation were utilized to construct and validate the models.The intratumoral habitat radiomics model,peritumoral radiomics model,clinical model(age,sex,alpha-fetoprotein,maximum tumor diameter),and combined model were established respectively,with nomogram for visualization.Model performance was evaluated using receiver operating characteristic (ROC) curve,calibration curve,and decision curve analysis. Results A total of 185 patients were enrolled and randomly divided into the training set(n=129) and validation set(n=56) at a ratio of 7∶3.The peritumoral 5 mm radiomics model achieved area under the curve (AUC) values of 0.838(95% CI:0.772-0.904)and 0.667(95% CI:0.524-0.809)in the training and validation sets,respectively,with superior performance in the training set compared with the peritumoral 7 mm model.The intratumoral habitat radiomics model yielded AUC values of 0.851(95% CI: 0.790-0.911) and 0.710(95% CI:0.581-0.839) in the training and validation sets,respectively,demonstrating superior predictive efficacy compared with the peritumoral radiomics model.The combined model integrated the clinical model,peritumoral 5 mm radiomics model,and intratumoral habitat radiomics model,achieving AUC values of 0.904(95% CI:0.855-0.953) and 0.788(95% CI: 0.666-0.909) in the training and validation sets,respectively,with specificities of 0.939 and 0.857.Decision curve and calibration curve analyses demonstrated favorable predictive efficacy and clinical utility of the combined model. Conclusion The combined model based on MR arterial phase intratumoral habitat imaging and peritumoral radiomics can effectively predict P53-mutated HCC preoperatively,with the combined model exhibiting better predictive performance.This model provides a reliable imaging tool for noninvasive preoperative identification of P53 mutation status and guidance of individualized treatment decisions.

Objective To compare the performance of low-dimensional interpretable quantitative CT imaging markers and high-dimensional radiomics features in predicting recurrence within 6 months after surgery(DFS6) in patients with pancreatic ductal adenocarcinoma(PDAC),and to evaluate the performance of the interpretable quantitative CT imaging markers for risk stratification. Methods This retrospective study included patients with pathologically confirmed PDAC after surgery at Ruijin Hospital between 2018 and 2024 who underwent contrast-enhanced CT within 50 days before surgery and had an ascertainable DFS6 outcome.Twenty-seven interpretable quantitative CT imaging markers were extracted from manually delineated tumor regions on arterial-phase contrast-enhanced CT images,with 1409 conventional radiomics features serving as the comparator.After patient-level data matching,178 patients were included,of whom 41 were DFS6-positive.The patients were assigned using a fixed split to a training cohort set of 140 patients and an internal test cohort set of 38 patients.The two feature representations were compared using identical data partitions,cross-validation procedures,evaluation metrics,and 504 candidate modeling pipelines. Results The prediction model based on low-dimensional interpretable quantitative CT imaging markers achieved a mean out-of-fold area under the receiver operating characteristic curve (AUC) of 0.739 in the training cohort set using 100-times repeated five-fold cross-validation and an AUC of 0.705 in the independent internal test cohort set.In comparison,the corresponding high-dimensional radiomics model achieved an out-of-fold AUC of 0.724 in the training cohort set but an AUC of only 0.510 in the internal test cohort set.The model comprised six interpretable CT phenotypes,including three-dimensional tumor surface area,axial tumor extent,density range,density complexity,mean core density,and the proportion of low-density voxels within the tumor core.These features characterized tumor spatial burden,intratumoral density heterogeneity,and low-density characteristics of the tumor core.Using an out-of-fold prediction-derived Youden cutoff of 0.485,the entire cohort was stratified into high- and low-risk groups,with a statistically significant difference in disease-free survival between the groups(log-rank P=0.004). Conclusion Low-dimensional interpretable quantitative CT imaging markers provided a compact and reproducible alternative to complex high-dimensional radiomic feature combinations for predicting early postoperative recurrence in PDAC.Compared with high-dimensional radiomic features,the proposed marker system showed better stability in internal validation and more coherent risk stratification performance.These findings support the potential value of interpretable CT-based quantitative phenotypes as a concise,transparent,and clinically translatable imaging approach for assessing early recurrence risk after PDAC surgery.

Objective To evaluate the incremental prognostic value of preoperative portal venous phase contrast-enhanced CT radiomics beyond the Heidelberg Prognostic Pancreatic Cancer (HELPP)score. Methods This retrospective study included 194 patients with pancreatic ductal adenocarcinoma who underwent radical resection.Patients were divided chronologically into a training cohort(n=133) and a validation cohort(n=61).HELPP,Radiomics,and Combined Cox models were constructed and evaluated using the C-index,patient-level paired bootstrap,and fixed-cutoff risk stratification. Results The Combined model achieved a C-index of 0.728 in both cohorts.In the validation cohort,the paired C-index difference(ΔC) versus HELPP was 0.066 (95% CI 0.001-0.133).Using the fixed cutoff,median OS was 21.52 vs. 9.46 months for the low- vs. high-risk groups(P<0.001). Conclusion Portal venous phase CT radiomics provided complementary prognostic information to the HELPP score,and the Combined model improved postoperative survival-risk discrimination.

Objective To compare and analyze the clinical efficacy of CT-guided percutaneous microwave ablation (MWA) and laparoscopic liver resection (LLR) in the treatment of primary small hepatocellular carcinoma. Methods A retrospective analysis was conducted on the clinical data of 189 patients with primary small hepatocellular carcinoma admitted to Xinxiang Central Hospital from January 2017 to December 2021.According to the inclusion and exclusion criteria,160 patients were ultimately included.Propensity score matching at a 1∶1 ratio was performed using age,sex,tumor size,number of tumors,Child-Pugh grade,and the China Liver Cancer (CNLC) staging system as covariates.After matching,77 patients received CT-guided MWA (MWA group) and 77 patients received LLR (LLR group).Operative time,intraoperative blood loss,length of hospital stay,treatment cost,liver function indicators,tumor radical cure rate,and incidence of complications were compared between the two groups. Results There were no significant differences in baseline data between the two groups,including sex,age,tumor location,tumor size,number of tumors,and Child-Pugh classification of liver function (P>0.05).Comparison of surgical indicators showed that operative time and hospital stay were shorter in the MWA group than in the LLR group,and intraoperative blood loss and treatment costs were lower in the MWA group than in the LLR group (P<0.05).At 7 days and 1 month after surgery,liver function indicators showed that alanine aminotransferase and aspartate aminotransferase levels were lower in the MWA group than in the LLR group (P<0.05).There were no statistically significant differences between the two groups in the imaging-based tumor radical cure rate at 1 month postoperatively or in the incidences of grade I,grade II,and overall postoperative complications within 30 days after surgery (P>0.05).However,the incidence of grade III complications differed significantly between the two groups (P<0.05). Conclusion For primary small hepatocellular carcinoma,CT-guided MWA and LLR demonstrate comparable tumor eradication efficacy.However,MWA offers advantages including less trauma,faster recovery,less impact on liver function,and lower economic burden,making it worthy of clinical promotion and application.

Objective To evaluate the value of combining time-dependent diffusion imaging(TDD-MRI) with vesical imaging-reporting and data system (VI-RADS)based on biparametric MRI for assessing muscle-invasive bladder cancer. Methods Retrospective collection of 204 bladder cancer patients who had undergone routine bladder MRI and TDD-MRI sequence scans,with surgical pathology confirmation.Patients were divided into a training set(143 cases) and a test set(61 cases) in a 7∶3 ratio.Microstructural parameters were derived from the imaging microstructural parameters using limited spectrally edited diffusion model,including intracellular volume fraction(V_in),intracellular diffusion coefficient (D_in),extracellular diffusion coefficient(D_ex),cell diameter (d),cellularity_index,cellularity,3D cellularity (rho),and diffusion dispersion rate(DDR).Apparent diffusion coefficient (ADC)-related parameters were also calculated,including ADC_PGSE,ADC_25Hz,ADC_45Hz,ADC_25Hz-PGSE,ADC_45Hz-PGSE,and ADC_45Hz-25Hz.Univariate logistic regression analysis was used to identify statistically significant parameters,which were then used to build a TDD model.This model was further combined with VI-RADS scoring to establish a joint model.The diagnostic performance of each model in predicting muscle-invasive bladder cancer was assessed using the area under the receiver operating characteristic curve(AUC),and comparisons of AUC values were performed using the DeLong test. Results In the training set,there were statistically significant differences(P<0.05)between non-muscle-invasive bladder cancer and muscle-invasive bladder cancer groups in V_in,D_ex,cellularity_index,cellularity,rho,ADC_PGSE,ADC_25Hz,ADC_45Hz and ADC_25Hz-PGSE.After univariate logistic regression analysis,seven parameters were ultimately selected to construct the TDD model.The combined model demonstrated the best diagnostic performance,achieving an AUC of 0.915(95% CI:0.831-0.998)in the test set,significantly higher than the VI-RADS score's AUC of 0.874(95% CI:0.769-0.978)(P=0.038). Conclusion Combining TDD-MRI with the VI-RADS scoring system based on biparametric MRI may improve the diagnostic performance for bladder cancer muscle invasion.

Objective To investigate the hysterosalpingography (HSG) signs associated with positive magnetic resonance imaging (MRI) findings in pelvic endometriosis (EMS),and to evaluate the diagnostic performance of HSG using MRI as the reference standard. Methods A total of 127 patients with suspected EMS-related primary infertility were retrospectively enrolled from Beijing Obstetrics and Gynecology Hospital,Capital Medical University,Baoding Maternal and Child Health Hospital,and Taiyuan Maternal and Child Health Hospital between January 2024 and June 2026.All patients underwent both HSG and MRI during the same clinical visit.Using MRI diagnoses as the reference,the sensitivity,specificity,positive predictive value,negative predictive value,and accuracy of three HSG signs-uterine retroflexion,tubal convergence toward the uterus,and localized pelvic contrast pooling-were calculated individually and in combination.Receiver operating characteristic (ROC) curves were constructed,and the area under the curve (AUC) was derived to assess the overall diagnostic efficacy of HSG. Results Among the 127 patients,MRI confirmed EMS in 98（77.16%） and excluded it in 29（22.84%）.Of the individual signs,tubal convergence toward the uterus yielded the best diagnostic performance（AUC=0.712;sensitivity,71.43%;specificity,68.97%）.Uterine retroflexion showed no independent diagnostic value（AUC=0.611;P=0.069）.Localized pelvic contrast pooling demonstrated relatively high specificity（75.86%） but low sensitivity （59.18%）.When the three signs were combined （positive if any one was present）,the diagnostic metrics were as follows：sensitivity,93.88%（92/98）;specificity,75.86% （22/29）;positive predictive value,92.93%（92/99）;negative predictive value,78.57%（22/28）;and accuracy,89.76%（114/127）,with an AUC of 0.849（P<0.001）. Conclusion The combined use of the three HSG signs demonstrates substantial diagnostic value for EMS-related infertility,particularly with high sensitivity,supporting its utility as an effective screening tool.The presence of tubal convergence toward the uterus,or localized pelvic contrast pooling on HSG should raise clinical suspicion for underlying EMS.

Objective To compare the predictive value of different MRI features and the EPE-Grade scoring system for extraprostatic extension (EPE),and to construct an EPE prediction model combining multiparametric MRI (mpMRI) and ¹⁸F-PSMA-PET/CT and evaluate its potential value in improving the accuracy of EPE prediction. Methods This retrospective study collected clinical and imaging data from patients with pathologically confirmed prostate cancer (PCa) who underwent radical prostatectomy (RP) at our hospital between January 2019 and June 2024 (n=330).For all enrolled patients,mpMRI findings were interpreted and graded.Deep learning (DL) models were constructed using a CNN-ResNet-50 approach.Using pathological results as the reference,the sensitivity,specificity,positive predictive value,and negative predictive value of different MRI features,the EPE-Grade score,and the DL models for predicting EPE were calculated.ROC and DCA curves were used to evaluate model performance and clinical net benefit. Results Preoperative t-PSA,f-PSA,f/t-PSA,PI-RADS score,ISUP grade group,seminal vesicle invasion,and positive surgical margin rate differed significantly between the EPE-negative and EPE-positive groups (P<0.05); age,BMI,and prostate volume did not differ significantly between the two groups.The specificity of different MRI features for assessing EPE was high (66%~100%),whereas sensitivity varied widely (2%~54%).The area under the curve (AUC) of the mpMRI feature model and the EPE-Grade score model for predicting EPE were both 0.64,and their clinical net benefits were similar.The AUCs of the mpMRI-DL,PET/CT-DL,and Multi-DL models for predicting EPE were 0.76,0.67,and 0.82,respectively.The Multi-DL model showed better predictive performance and greater clinical net benefit than the single-modality DL models (P<0.05). Conclusion The MRI feature model and the EPE-Grade score model had similar performance in predicting EPE,but the latter had reduced subjective dependence and was easier to learn and apply.The Multi-DL model showed good predictive performance and clinical net benefit for EPE and has potential value for improving preoperative prediction of EPE.

Objective To construct a predictive model for postoperative recurrence risk stratification in patients with non-muscle-invasive bladder cancer (NMIBC) based on biparametric MRI (bp-MRI) radiomics combined with clinicopathological factors. Methods A total of 248 patients with NMIBC who underwent transurethral resection of bladder tumor (TURBT) in our hospital between January 2018 and December 2023 were retrospectively and consecutively enrolled.Patients were divided into a training set (n=173) and a validation set (n=75) by stratified random sampling at a 7∶3 ratio.FAE software was used to extract radiomic features from preoperative bp-MRI images.The least absolute shrinkage and selection operator (LASSO) Cox regression was used for feature screening to establish the Rad-score.Cox regression analyses were performed to identify independent prognostic factors,and clinical,radiomic,and combined predictive models were constructed.Model performance was evaluated using the C-index and 1- and 3-year time-dependent receiver operating characteristic (TD-ROC) curves and area under the curve (AUC). Results Nine radiomic features were ultimately selected to construct the Rad-score.Multivariate Cox regression analysis identified the Rad-score and pathological grade as independent risk factors (P<0.05),while preoperative gross hematuria was an independent protective factor (P<0.05).The combined model showed the best predictive performance,with a C-index of 0.734 in the training set and 0.767 in the validation set; the 1- and 3-year AUCs were 0.791 and 0.769 in the training set and 0.862 and 0.763 in the validation set,respectively.The nomogram derived from the combined model visually quantified the risk weight of each prognostic indicator and accurately predicted individual postoperative recurrence probability.Based on model scores,all patients were divided into low-,intermediate-,and high-risk recurrence subgroups,with recurrence rates of 13.3%,30.5%,and 59.0%,respectively,and the overall difference among groups was statistically significant (P<0.01). Conclusion The prediction model and nomogram constructed by integrating bp-MRI radiomics with clinicopathological factors can accurately predict postoperative recurrence risk in patients with NMIBC after TURBT.With stable predictive performance and good discrimination,the model enables preoperative individualized risk stratification and may help guide clinical follow-up planning and precise therapeutic decision-making.

Objective To compare the detection performance of CT lymphangiography (CTL) and non-contrast MR lymphangiography (MRL) for osseous lesions in Gorham-Stout disease (GSD). Methods The CTL and non-contrast MRL data of 15 patients diagnosed with GSD were retrospectively analyzed.Lesions in the cervical spine,thoracic spine,lumbar spine,pelvic bones,and spleen were evaluated.The Kappa test was used to assess agreement between CTL and MRL in lesion detection,and the McNemar test was used to compare differences in detection rates. Results Agreement between CTL and MRL varied across different skeletal sites,with moderate to good agreement at most sites.Agreement was relatively good for the cervical spine,lumbar spine,sacrum,iliac bone,and spleen,whereas agreement for the thoracic spine was relatively low.No significant differences in lesion detection rates were found between the two modalities at any site (all P>0.05).Abnormal intralesional lipiodol distribution was observed on CTL in 9 patients.Splenic lesions were detected only on MRL in 2 patients. Conclusion CTL and non-contrast MRL show variable agreement in the detection of lesions at different skeletal sites in GSD,with moderate to good agreement at most sites.CTL can demonstrate abnormal intralesional lipiodol distribution,and MRL may provide supplementary value for the evaluation of involvement of certain parenchymal organs.

Objective To investigate the differences in total maturation score (TMS) and white matter abnormality scores on cranial MRI among neonates of different gestational ages,and to analyze the correlation between cranial MRI scores and neurodevelopmental outcomes at 6 months corrected age in preterm infants. Methods According to gestational age (GA),81 neonates were divided into a preterm group (GA<37 weeks,n=46) and a full-term group (GA≥37 weeks,n = 35).All subjects underwent routine cranial MRI at a corrected gestational age of 37~45 weeks.Two neuroradiologists independently reviewed the MRI images to calculate TMS and evaluate white matter abnormalities.Valid follow-up data were available for 22 preterm infants at 6 months corrected age.TMS and white matter abnormality scores were compared between the two groups,and the correlations between neurodevelopmental outcomes and MRI scores in preterm infants were analyzed. Results Compared with the full-term group,the preterm group had lower total TMS scores and higher total white matter abnormality scores (P<0.05).Significant between-group differences were found in the TMS subscores (myelination and germinal matrix) and white matter abnormality subscores (cystic degeneration,delayed myelination,thinning of the corpus callosum,and dilated lateral ventricles) (all P<0.05).Focal signal abnormality scores were negatively correlated with adaptive behavior,gross motor function,fine motor skills,language,and social interaction in preterm infants (r_s=-0.638,-0.528,-0.531,-0.561,and -0.676,respectively; P<0.05). Conclusion Total TMS scores at term-equivalent age and white matter abnormality scores can objective ly quantify brain injury in preterm infants,and focal signal abnormality scores are negatively correlated with neurodevelopmental outcomes in preterm infants.

Objective To evaluate the value of radiomics models based on body composition features on initial abdominal CT for predicting early hemoglobin decline in patients with acute paraquat (PQ) poisoning. Methods This single-center retrospective study included 178 patients with confirmed PQ poisoning between July 2024 and August 2025.All patients had an initial hemoglobin (Hb) level of ≥115 g/L and underwent repeat Hb measurement on days 3~5 after admission.Patients were categorized according to whether Hb declined to <115 g/L and were randomly divided into training and test sets at a ratio of 7∶3.At the L₃ level on abdominal CT,regions of subcutaneous adipose tissue (SAT),muscle,and visceral adipose tissue (VAT) were segmented for radiomics feature extraction.After data preprocessing and feature selection,10 classifier models were constructed.Receiver operating characteristic (ROC) curves were used to evaluate model performance,and radiomics prediction scores (R-Pred scores) were obtained from model-predicted probabilities.Univariate and multivariable logistic regression analyses were performed for clinical variables,and a combined model and nomogram were developed by integrating relevant clinical factors with the optimal R-Pred score.Calibration and decision curve analyses were used to assess model performance. Results In the test set,the areas under the curve (AUCs) of the SAT,muscle,and VAT radiomics models were 0.793,0.759,and 0.764,respectively,with the SAT model showing the highest predictive performance.Under a uniform Hb threshold,female sex and urinary paraquat concentration (UPQC) were independently associated with early hemoglobin decline.The combined model incorporating sex,UPQC,and the SAT R-Pred score achieved AUCs of 0.948 and 0.975 in the training and test sets,respectively.Calibration and decision curve analyses suggested potential clinical utility of the combined model. Conclusion Radiomics models based on SAT,muscle,and VAT at the L₃ level on initial abdominal CT have predictive value for early hemoglobin decline in acute PQ poisoning,with the SAT model showing the best performance.The combined nomogram incorporating the SAT R-Pred score,UPQC,and sex may assist early risk stratification; however,its performance requires further validation in larger samples and external cohorts.