Objective To investigate cortical structural alterations and associated functional circuit changes in patients with neuropsychiatric systemic lupus erythematosus（NPSLE）using surface-based morphometry（SBM）combined with resting-state functional magnetic resonance imaging（rs-fMRI）,and to explore their relationships with systemic inflammation and clinical features. Methods Thirty-one NPSLE patients and 33 age- and sex-matched healthy controls（HCs） were prospectively enrolled.High-resolution 3D-T₁ structural images and rs-fMRI data were acquired.Cortical thickness and volume were calculated using FreeSurfer to identify intergroup differences.Brain regions showing significant morphological differences were defined as seed regions of interest（ROIs）;time series were extracted and whole-brain functional connectivity（FC）analysis was performed.Imaging indices with significant group differences were correlated with hematological markers（erythrocyte sedimentation rate,immunoglobulin M）and neuropsychological scale scores. Results Compared with the HC group,NPSLE patients exhibited significant cortical thinning in the right pars opercularis of the inferior frontal gyrus,left insula,and left paracentral lobule,as well as reduced cortical volume in the left lateral orbitofrontal cortex,left rostral middle frontal gyrus,and left middle temporal gyrus（vertex-wise P<0.01,cluster-wise P<0.05）.Using the left insula as a seed,NPSLE patients showed decreased FC with the left middle cingulate cortex and right dorsolateral superior frontal gyrus.Using the left paracentral lobule as a seed,decreased FC was observed with bilateral insula（all P<0.05）.Correlation analyses revealed that cortical thickness of the right pars opercularis and left insula,as well as FC of the right dorsolateral superior frontal gyrus,were negatively correlated with erythrocyte sedimentation rate;left insular thickness was positively correlated with immunoglobulin M（all P<0.05）.No linear correlations were found between imaging indices and neuropsychological scores. Conclusion Widespread cortical atrophy exists in NPSLE patients,primarily involving core nodes of the salience network and executive control network.Systemic inflammation is closely associated with these structure-function co-alterations,suggesting its potential involvement in the underlying pathological process.The combined SBM and FC metrics may serve as objective multimodal imaging biomarkers for evaluating inflammation-related neurological damage in NPSLE.

Objective Depression is often comorbid with pain,and the underlying mechanisms remain incompletely understood.This study aimed to investigate the brain structural characteristics in adolescent patients with depression comorbid with pain,utilizing artificial intelligence-based automated brain segmentation technology. Methods This study enrolled 34 adolescents with comorbid depression and pain,26 patients with depression only,and 31 healthy controls.Based on 3D-T₁ structural sequences,whole-brain,hippocampal,and amygdala subregional structural metrics were analyzed using an automated brain segmentation technique.Group differences in structural metrics were compared,and the association between differential metrics and clinical variables was explored. Results Compared to healthy controls,the depression-only group showed differences in regions including the left frontal pole,right parahippocampal gyrus,and fimbria in measures such as volume fraction and cortical surface area.Relative to the depression-only group,the depression-pain comorbidity group exhibited differences in the left frontal pole,left medial orbitofrontal cortex,right precuneus,right hippocampal CA3 subregion,and right lateral nucleus of the amygdala in metrics including volume fraction,cortical surface area,and cortical thickness.In the depression-pain comorbidity group,the volume fraction of the right CA3 was positively correlated with numeric rating scale （NRS） scores. Conclusion Adolescent patients with depression exhibit structural abnormalities in the prefrontal-limbic system.In those comorbid with pain, the aberrant pattern extends to additional brain regions,including prefrontal cortex,precuneus,hippocampus,and amygdala,which are associated with pain perception and emotional regulation.These findings offer direct neuroimaging evidence for understanding brain structural alterations in adolescent depression and in depression comorbid with pain.

Objective To explore the predictive value of whole-brain CT perfusion （CTP） quantitative parameters combined with clinical factors for early postoperative cerebral infarction（Post-CI） in adult patients with moyamoya disease （MMD） undergoing revascularization. Methods A total of 121 adult MMD patients who underwent revascularization were retrospectively enrolled and divided into a Post-CI group（14 patients） and a non-Post-CI group（107 patients） according to whether cerebral infarction occurred within 2 weeks after surgery.All patients underwent one‑stop whole‑brain CTP within 1 week preoperatively.Volumes of time to maximum of residual function（V_{Tmax>4 s},V_{Tmax>6 s},V_{Tmax>8 s},V_{Tmax>10 s}）and cerebral blood flow（V_CBF<20%,V_CBF<30%,V_CBF<34%,V_CBF<38%,V_CBF<50%）at different thresholds were obtained by postprocessing with F-Stroke software.Clinical characteristics and CTP parameters were compared between the two groups.Binary logistic regression was used to identify independent predictors of Post-CI,and the efficacies for predicting Post-CI were analyzed with receiver operating characteristic curves. Results There were significant differences between the two groups in onset type,history of cerebral infarction within 2 months preoperatively,modified Rankin Scale score at admission,and CTP staging （all P<0.05）.V_{Tmax>4 s},V_{Tmax>6 s},V_{Tmax>8 s},V_{Tmax>10 s},V_CBF<30%,V_CBF<34%,V_CBF<38%,and V_CBF<50% were all higher in the Post-CI group than in the non-Post-CI group（all P<0.05）.Binary logistic regression demonstrated that V_{Tmax>6 s}（OR=1.010,95%CI:1.002-1.018,P=0.019）and history of cerebral infarction within 2 months preoperatively（OR=7.744,95%CI:1.919-31.244,P=0.004）were independent risk factors for Post-CI.The areas under the curve（AUC） of V_{Tmax>6 s} and preoperative cerebral infarction within 2 months for predicting Post‑CI were 0.856 and 0.770,respectively.The combined AUC of the two parameters was 0.893,which was significantly higher than that of preoperative infarction within 2 months（P=0.005）,whereas no significant difference was found compared with V_{Tmax>6 s} alone （P=0.281）. Conclusion V_{Tmax>6 s} combined with a history of cerebral infarction within 2 months preoperatively can effectively predict new cerebral infarction after revascularization in adult patients with MMD.

Objective To investigate the value of CT-based intratumoral habitat imaging and peritumoral radiomics models in predicting central lymph node metastasis（CLNM） in papillary thyroid carcinoma（PTC）. Methods Clinical and imaging data of 216 PTC patients confirmed by pathology were retrospectively collected.All patients were divided into CLNM‑positive（n=108） and CLNM‑negative（n=108） groups based on pathology,and randomly divided into a training set（n=151） and a test set（n=65） at the ratio of 7∶3.Volumes of interest（VOI） of the primary tumor were delineated on arterial-phase CT images,and peritumoral VOIs were obtained by outward expansion of 1 mm,3 mm,and 5 mm.K‑means clustering,with the optimal number of clusters determined by the silhouette coefficient,was used to partition the tumor into three habitat subregions.A total of 107 radiomic features were extracted from intratumoral,peritumoral,and habitat VOIs.In the training set, Spearman correlation analysis and least absolute shrinkage and selection operator（LASSO） algorithm were applied to select the features most strongly associated with CLNM.Clinical predictive factors were selected by univariate logistic regression,and a clinical model was subsequently constructed.Light Gradient Boosting Machine（LightGBM） was employed to separately construct intratumoral,peritumoral,and habitat models,and logistic regression was employed to develop a combined model that integrated the optimal peritumoral and habitat models with clinical predictors.A combined model integrating the optimal peritumoral and habitat features and the clinical predictors was then established. Model performance was evaluated using the area under the receiver operating characteristic curve（AUC）,calibration curve,and decision curve analysis. Results Capsular interruption sign was significantly associated with CLNM（OR=1.871,95%CI：1.298-2.697,P=0.005）.Among the peritumoral models,the peritumoral 1 mm model performed best,with AUCs of 0.846（95%CI：0.783-0.908） in the training set and 0.734（95%CI：0.612-0.855） in the test set.Habitat 2 achieved the highest performance among habitat models,with AUCs of 0.876（95%CI：0.821-0.932） and 0.851（95%CI：0.755-0.948） in the training and test sets,respectively.The combined model incorporating clinical,peritumoral 1 mm,and habitat 2 features yielded AUCs of 0.907（95%CI：0.862-0.952） in the training set and 0.869（95%CI：0.780-0.957） in the test set.Calibration curves and decision curve analysis demonstrated that the combined model had good predictive performance and clinical net benefit. Conclusion The CT-based combined model integrating intratumoral habitat and peritumoral radiomics can accurately predict CLNM in PTC,providing a noninvasive imaging basis for individualized surgical decision-making.

Objective To explore the prognostic factors of radical concurrent chemoradiotherapy for locally advanced nasopharyngeal carcinoma after neoadjuvant chemotherapy and analyze the clinical application value of nimotuzumab,and further construct the relevant clinical prediction model to guide the subsequent clinical diagnosis and treatment work. Methods A total of 193 patients with locally advanced nasopharyngeal carcinoma treated by neoadjuvant chemotherapy and radical concurrent chemoradiotherapy in Affiliated Hospital of Guangdong Medical University from January 2016 to December 2021 were retrospectively included and the baseline clinical characteristics and follow-up survival conditions were analyzed.Univariate analysis and multivariate analysis of prognostic factors for locally advanced nasopharyngeal carcinoma treated with neoadjuvant chemotherapy followed by radical concurrent chemoradiotherapy.Construction of a prognostic model and analysis of predictive efficacy for locally advanced nasopharyngeal carcinoma treated with radical synchronous radiochemotherapy after neoadjuvant chemotherapy. Results The median follow-up of 193 patients was 33.0（21.0-40.0）months.During the follow-up,35 cases had disease progression,7 cases had local recurrence,31 cases had distant metastasis,and 13 cases died.The 3-year distant metastasis-free survival rate,progression-free survival rate,and overall survival rate were 84.3%,81.6%,and 90.1%,respectively.Univariate analysis showed that age,clinical stage,histopathological type,EGFR expression level,neutrophil‑lymphocyte ratio（NLR）,and combination with nimotuzumab were influencing factors for progression‑free survival（PFS）in patients with locally advanced nasopharyngeal carcinoma（LA‑NPC）receiving radical concurrent chemoradiotherapy（CCRT）following neoadjuvant chemotherapy（NACT）（all P<0.05）.Multivariate Cox regression analysis revealed that clinical stage,histopathological type,NLR,and combination with nimotuzumab were independent prognostic factors for LA‑NPC patients treated with NACT followed by radical CCRT（P<0.05）.A nomogram model for long‑term prognosis was established based on multivariate results.The time-dependent ROC areas under the curve（AUC）of this model for predicting 1‑,2‑,and 3‑year PFS were 0.774,0.835,and 0.738,respectively. Conclusion The prognosis of radical concurrent chemoradiotherapy for locally advanced nasopharyngeal carcinoma after neoadjuvant chemotherapy is not only affected by clinical stage,histopathological type and the neutrophil/lymphocyte ratio,but also closely related to whether combined with nimotuzumab treatment or not.Using the above factors to predict the progression free survival of patients has good efficacy.External validation is needed before clinical application of this model.

Objective This study aims to establish a new kind of scorecard model to diagnose ground-glass-like invasiveness lung adenocarcinoma（LAC） and verify the application value of it. Methods Patients with ground-glass nodules（GGN） diagnosed ,dividing into a non-ⅠA and ⅠA group.The data was divided into a training set and a validation set in a ratio of 7∶3.Clinical,imaging and radiomics features were extracted.Synthetic minority oversampling technique for nominal and continuous was used for addressing the issue of imbalanced sample distribution.Population stability index was adopted for feature stability analysis.A Logistic regression analysis was employed to construct the prediction model,and a nomogram and scorecard model were established based on it.Receiver operating characteristic and Hosmer-Lemeshow test were used to validate the discrimination and calibration. Results Multivariable analysis showed the sex,age,lobulation,air bronchogram,pleural indentation,vacuole sign and original_firstorder_10Percentil to be independent factors for prediction invasive adenocarcinoma（IA）.Internal verification showed the model had a sensitivity of 78.4% and specificity of 81.6% with the Youden index（YI） of 0.600 and an area under the curve（AUC） of 0.894（0.871-0.917）.External verification showed the model had a sensitivity of 84.4% and specificity of 75.3% with the YI at 0.597 and AUC of 0.864（0.805-0.923）.Calibration curves indicated good internal and external stability（P=0.544,P=0.934）.The AUC of the scorecard model is significantly superior to that of the nomogram model（P=0.015）. Conclusion This scorecard model has certain application potential on diagnosing invasive GGN,but further optimization of feature engineering and multi-center and prospective validation are still needed.

Objective To develop and validate an interpretable radiomics machine learning model based on high-resolution CT（HRCT）for predicting the risk of rheumatoid arthritis-associated interstitial lung disease（RA-ILD）. Methods A total of 328 RA patients without baseline RA-ILD were retrospectively enrolled and divided into a training set （n=229）and a test set （n=99） at a 7∶3 ratio.Whole-lung radiomics features were derived from HRCT scans.For feature selection,we ran Spearman correlation,Mann‑Whitney U,and LASSO regression in sequence.Nine machine learning classifiers were built and compared to identify the best one.The chosen radiomics model was then interpreted using SHAP.Next,we added independent clinical risk factors to the radiomics signature,forming a combined model.Performance was evaluated with AUC,calibration curves,and decision curve analysis（DCA）. Results Among the nine classifiers,logistic regression radiomics model demonstrated good generalizability and stability.It gave AUCs of 0.804 in the training cohort and 0.781 in the testing cohort.SHAP analysis showed that first‑order and texture features contributed a lot to the predictions.Adding disease duration and anti‑CCP improved discrimination further,raising AUCs to 0.866（training） and 0.858（test）.DCA also confirmed that the combined model offered good clinical net benefit across various threshold probabilities. Conclusion This interpretable whole-lung HRCT radiomics machine learning model showed good performance in predicting RA-ILD risk in patients with RA.Incorporating clinical variables further improved model performance and may help support individualized risk evaluation.

Objective This study aimed to investigate the diagnostic performance of PET/CT combined with spectral CT for detecting lymph node metastasis（LNM）in patients with non-small cell lung cancer（NSCLC）. Methods The clinical data,¹⁸F-fluorodeoxyglucose（¹⁸F-FDG）PET/CT and spectral CT examination data of 80 patients with NSCLC admitted to our hospital from April 2024 to March 2026 were retrospectively analyzed.Patients were classified into LNM group（n=51）and non-LNM group（n=29）based on pathological confirmation.Clinical features,PET/CT parameters,and spectral CT indices were compared between the two groups.Multivariate Logistic regression was performed to identify independent predictors of LNM,and receiver operating characteristic（ROC）analysis was used to evaluate the diagnostic efficiency of the combined model. Results Tumor diameter,maximum standardized uptake value（SUV_max）of lymph nodes,primary lesion SUV_max and short diameter of lymph nodes in the metastasis group were higher than those in the non-metastasis group;while the proportion of T1 stage tumors,normalized iodine concentration（NIC）and spectral curve slope（λ_HU）of lymph nodes were lower than those in the non-metastasis group（P<0.05）.Multivariate Logistic regression analysis showed that tumor diameter,T stage,lymph node SUV_max,primary lesion SUV_max,lymph node short diameter,NIC and lymph node λ_HU were independent influencing factors for lymph node metastasis in NSCLC patients.ROC curve analysis indicated that both PET/CT parameters and spectral CT parameters had certain predictive value for lymph node metastasis,with AUC values of 0.910 and 0.882 respectively.The combined application achieved the highest predictive efficiency with an AUC of 0.940,a sensitivity of 84.29% and a specificity of 93.13%. Conclusion PET/CT combined with spectral CT possesses favorable diagnostic value for lymph node metastasis in NSCLC patients.Changes in tumor diameter,tumor T stage,PET/CT and spectral CT parameters can affect the occurrence of lymph node metastasis.

Objective To investigate the correlation between preoperative magnetic resonance imaging（MRI）-derived apparent diffusion coefficient（ADC）values and pathological indicators of aggressiveness in rectal cancer. Methods A total of 92 patients with pathologically confirmed rectal adenocarcinoma who underwent preoperative 3.0 T MRI（including a DWI sequence）were retrospectively enrolled.Regions of interest（ROIs）were delineated along the tumor margins on the three largest tumor slices on ADC maps,and the minimum（Min ADC）,mean（Mean ADC）,and maximum ADC（Max ADC）values were recorded.Postoperative pathology served as the gold standard for determining tumor T stage,N stage,differentiation grade,perineural invasion（PNI）,and lymphatic invasion（LI）.Normally distributed data were analyzed using one-way ANOVA or independent samples t-test,while non-normally distributed data were analyzed using Kruskal-Wallis test or Mann-Whitney U test.Spearman correlation analysis was performed to assess the relationship between ADC values and pathological features. Results Mean ADC values showed positive correlations with pathological T stage（r=0.298,P=0.004）and N stage（r=0.446,P<0.001）.The mean ADC was significantly higher in PNI-positive group compared to PNI-negative group（0.911×10^-3mm²/s vs. 0.809×10^-3mm²/s,P=0.006）,and in the LI-positive group compared to LI-negative group（0.891×10^-3mm²/s vs. 0.850×10^-3mm²/s,P=0.046）.Maximum ADC values were positively correlated with N stage（r=0.279,P=0.007）.No significant differences in ADC values were observed among groups with different tumor differentiation grades.Across all comparisons,the minimum ADC value showed no significant difference. Conclusion An elevated preoperative mean ADC is positively associated with advanced T/N stages,perineural invasion,and lymphatic invasion in rectal cancer.These findings suggest the Mean ADC may serve as a potential noninvasive imaging biomarker for assessing tumor aggressiveness in rectal cancer.

Objective To evaluate the overall diagnostic performance of radiomics and its extended techniques for preoperative non-invasive prediction of lymph node metastasis（LNM）in gastric cancer,and to explore methodological factors affecting performance. Methods Mainstream Chinese and English databases were searched.Study quality was assessed using the QUADAS-2 tool.A bivariate random-effects model was used to pool sensitivity,specificity,and diagnostic odds ratio（DOR）,and to generate summary receiver operating characteristic（SROC）curves with area under the curve（AUC）.Sources of heterogeneity were explored by meta-regression and subgroup analyses.Leave-one-out sensitivity analysis was performed.Publication bias was evaluated with Deeks' funnel plot.Clinical utility was visualized using a Fagan nomogram and a likelihood ratio scatterplot matrix. Results Seventeen retrospective studies involving 5037 patients were included.Pooled sensitivity was 0.84（95%CI：0.78-0.89）,specificity 0.79（95%CI：0.71-0.85）,DOR 19（95%CI：9-39）,and AUC 0.88（95%CI：0.85-0.91）.Meta-regression identified no statistically significant covariates.In subgroup analysis,DOR was 32.42 for studies with sample size >300 and 18.16 for those incorporating clinical features,both higher than their respective reference groups,with the former showing more prominent improvement.Sensitivity analysis confirmed robustness.Deeks' test showed no publication bias（P=0.669）.Visual analysis illustrated post-test probability changes and subgroup differences. Conclusion Radiomics and its extended techniques demonstrate good diagnostic performance for preoperative prediction of gastric cancer LNM,though accuracy varies with multiple interacting factors.Subgroup analysis indicates that larger sample size and incorporation of clinical features may improve diagnostic performance.Deep learning and functional imaging also show potential,yet are respectively limited by the small number of studies and the diversity of modalities within this group,requiring further validation.

Objective Poorly cohesive gastric carcinoma（PCC）is frequently underestimated at preoperative endoscopy and associated with poor prognosis.This study aimed to evaluate the value of dual-energy CT（DECT）for preoperative prediction of PCC components in gastric cancer（GC）. Methods We retrospectively analyzed clinicopathological and DECT data of 328 consecutive GC patients who underwent radical resection at our hospital between September 2022 and September 2024.Delayed-phase iodine concentration（IC）-related quantitative parameters and CT morphological features（tumor location,size,serosal invasion,layered enhancement,enhancement pattern）were evaluated.Significant DECT features were screened by univariate analysis and included in multivariate logistic regression to construct a DECT-based combined prediction model.The area under the receiver operating characteristic curve（AUC）,sensitivity,and specificity were used to compare the diagnostic efficacy of the combined model,key DECT quantitative parameters,and preoperative endoscopic biopsy for predicting PCC components. Results Multivariate analysis identified three independent risk factors for PCC：age（OR 0.96,95%CI：0.93-0.99,P=0.011）,normalized iodine concentration（NIC,%）（OR 1.08,95%CI：1.03-1.13,P=0.002）,and enhancement pattern（OR 2.51,95%CI：1.43-4.39,P=0.001）.The combined model based on these three factors yielded an AUC of 0.848（95%CI：0.805-0.890）for diagnosing PCC-containing GC,which was similar to that of delayed-phase NIC alone（AUC 0.834,95%CI：0.789-0.880）,but significantly superior to preoperative endoscopic biopsy（AUC 0.634,95%CI：0.574-0.695）. Conclusion The delayed phase of DECT is effective for identifying PCC components within GC.Its NIC is a valid imaging biomarker for PCC discrimination,serving as an important adjunct to endoscopic biopsy and offering imaging support for personalized management and biopsy optimization.

Objective To investigate the value of pretreatment clinical features,conventional CT findings,and dual-energy CT（DECT）quantitative parameters for predicting pathologic response to neoadjuvant immunochemotherapy（NICT）in locally advanced gastric cancer（LAGC）. Methods A total of 111 patients with LAGC who underwent NICT followed by radical surgery between February 2023 and August 2025 were retrospectively enrolled.Based on postoperative tumor regression grade（TRG）,patients were classified as responders（TRG 0-1）and nonresponders（TRG 2-3）.Pretreatment clinical characteristics,conventional imaging features,and DECT quantitative parameters were compared between the two groups.Univariate and multivariate Logistic regression analyses were performed to identify independent predictors and construct a combined predictive model.Model performance was evaluated using receiver operating characteristic（ROC）curves,calibration curves,and decision curve analysis（DCA）. Results Elevated CA19-9,CT-based Borrmann type Ⅳ,and higher venous-phase effective atomic number（V-Zeff）were independently associated with a lower probability of treatment response.The combined model achieved an area under the curve（AUC）of 0.782（95%CI：0.694-0.855）,which was significantly higher than that of each individual predictor（all P<0.05）.Calibration analysis demonstrated good model fit（P=0.110）,and DCA indicated that the model offered relatively high clinical net benefit across a threshold probability range of 0.03-0.74. Conclusion Pretreatment V-Zeff combined with CA19-9 and CT-based Borrmann classification was associated with pathologic response to NICT in patients with LAGC and may provide complementary information for pretreatment response assessment and risk stratification.

Objective To explore the automatic segmentation of colorectal cancer liver metastases（CRLM）on contrast-enhanced CT using nnU-Net and to develop a predictive model for early disease progression after thermal ablation of CRLM. Methods Patients with CRLM from two medical centers were retrospectively enrolled.Primary lesions were pathologically confirmed in all patients,and CRLM were confirmed by biopsy or clinical multidisciplinary diagnosis.All validated target lesions were manually delineated using ITK-SNAP software,and nnU-Net was trained for automatic segmentation.CRLM were divided into two groups based on target lesion size（<10 mm vs. 10-30 mm）.The performance of nnU-Net for automatic segmentation of CRLM was statistically compared between the two size groups.Furthermore,a Deep Radiomics（D-R）model was constructed using the automatically segmented regions to differentiate early recurrence from non-early recurrence after thermal ablation.The sensitivity,specificity,and performance of the D-R model for predicting early disease progression after CRLM thermal ablation were statistically analysed. Results A total of 186 patients（534 lesions）from the two centres were included,with an average age of（58.03±10.69）years,including 110 male patients.The accuracy（Dice coefficient）of nnU-Net for automatic segmentation of target lesions was （0.83±0.07）,and there was a significant difference in segmentation accuracy between lesions <10 mm and those 10-30 mm（Dice coefficient：0.71±0.04 vs. 0.87±0.03,P<0.01）.Among them,78 CRLM lesions underwent thermal ablation,with 21 lesions showing early disease progression（DP group）and 57 showing no early progression（NDP group）.The pre-ablation clinical model yielded an AUC of 0.877（95%CI：0.784-0.970）,sensitivity of 90.5%,and specificity of 22.8% for differentiating DP from NDP.In the test set,the D-R model achieved a predictive performance（AUC）of 0.796（95%CI：0.579-0.952）,sensitivity of 100%,and specificity of 66.7% for early disease progression after CRLM thermal ablation. Conclusion nnU-Net is suitable for automatic segmentation of CRLM target lesions larger than 10 mm,and its combination with a deep radiomics model enables end-to-end prediction of early disease progression after thermal ablation of CRLM.

Objective To explore the clinical application value of an intratumoral heterogeneity（ITH）habitat analysis model constructed based on multiparametric magnetic resonance imaging（mpMRI）for non-invasively predicting the full range of International Society of Urological Pathology（ISUP）grades 1-5 of prostate cancer（PCa）before surgery. Methods A retrospective study was conducted on 554 patients with prostate cancer（PCa）.Tumor habitat regions were delineated using unsupervised clustering to derive intratumoral heterogeneity（ITH）scores.These scores were then integrated with clinical parameters to develop a multi-class support vector machine（SVM）fusion model.To evaluate the incremental diagnostic value,clinical-only and whole-tumor radiomics models were established as baselines for comparative analysis. Results ITH scores demonstrated a significant positive correlation with ISUP grade groups（P<0.05）,with significantly higher scores observed in the ISUP 4-5 group compared to theI SUP 1-2 group（P<0.001）.The proposed fusion model exhibited superior diagnostic performance in predicting the full spectrum of ISUP grade groups.In the training cohort,the micro-average and macro-average AUCs were 0.812 and 0.803,respectively.In the internal validation cohort,the corresponding AUCs were 0.729 and 0.715.In the external test cohort,the fusion model（micro-average AUC：0.659;macro-average AUC：0.721）outperformed both the whole-tumor radiomics model（0.618;0.660）and the clinical model（0.602;0.704）.Furthermore,the model demonstrated excellent discriminative capability for ISUP grade group 5,with favorable consistency with histopathological findings. Conclusion ITH habitat analysis effectively quantifies the spatial heterogeneity of PCa.Compared to traditional approaches,the multi-class framework presented in this study provides superior discriminative power and serves as a promising non-invasive tool for preoperative precise risk stratification.

Objective To explore the diagnostic value of virtual magnetic resonance elastography (vMRE) and intravoxel incoherent motion (IVIM) imaging for distinguishing benign and malignant prostate lesions. Methods Imaging data of vMRE,IVIM and diffusion-weighted imaging (DWI) were collected from 80 patients with prostate lesions,including 37 benign cases and 43 malignant cases.The virtual shear modulus (μ),true diffusion coefficient (D),pseudo-diffusion coefficient(D^*),perfusion fraction(f)and apparent diffusion coefficient(ADC) of primary lesions were measured respectively. Clinical data such as patient age,prostate volume,total prostate-specific antigen (TPSA) and PI-RADS score were also recorded.The area under the receiver operating characteristic curve (AUC)and Delong test were adopted to evaluate the diagnostic efficacy of each parameter.Logistic regression analysis was performed to identify independent predictive factors and establish combined diagnostic models. Results Compared with the benign group,the malignant group presented significantly higher μ and TPSA values,as well as remarkably lower ADC,D and f values.Among the single parameters,the μ value had the highest diagnostic efficacy,with an AUC of 0.860.Logistic regression analysis demonstrated that μ,D and ADC were independent predictors for differentiating benign and malignant prostate lesions.The combined diagnosis of the three parameters yielded an AUC of 0.945,with a sensitivity of 79.07% and a specificity of 97.30%.Delong test revealed significant differences in AUC between the combined model and single parameters including μ,D(AUC = 0.833),f (AUC=0.756),ADC(AUC=0.775),TPSA(AUC=0.637),and PI-RADS score(AUC=0.812),(Z=2.780,2.916,3.499,3.490,3.685,all P<0.05). Conclusion vMRE,IVIM and DWI all contribute to the differentiation of benign and malignant prostate lesions.The multiparameter combined diagnosis based on independent predictors μ,D and ADC achieves the optimal diagnostic performance.

Objective To characterize the timing of in-situ recurrence after high-intensity focused ultrasound (HIFU) ablation of uterine fibroids and to use a mixture cure model to separate cured (long-term recurrence-ree) from uncured patients and identify their determinants. Methods Retrospective cohort study.258 patients treated with ultrasound-guided HIFU (September 2020-December 2022) were analyzed,with recurrence-free survival (RFS) as the outcome and administrative censoring at 30 April 2026.A proportional-hazards mixture cure model (Logistic incidence,Weibull proportional-hazards latency) assessed the non-perfused volume ratio (NPVR),intratumoral flow voids,fat-suppressed T₂WI (FS-T₂WI) signal and age;a Cox model served as comparator,with time-dependent AUC,calibration,decision curve analysis(DCA) and bootstrap internal validation. Results Over a median follow-up of 55 months,in-situ recurrence occurred in 116/258(45.0%;142 censored),median 17 months,concentrated within the first 2 years.The estimated cure fraction was 54.5%,driven mainly by NPVR(per +1 SD,OR for being uncured 0.07)and flow voids (OR：25.1);predicted cure fractions across NPVR tertiles were 15%,59% and 90%.In the latency component only NPVR remained significant(HR：0.77).The Cox C-index was 0.835 (bootstrap-corrected 0.832);24- and 36-month time-dependent AUC were both 0.91;NPVR violated the proportional-hazards assumption(P=0.004),supporting the cure model. Conclusion Post-HIFU in-situ recurrence can be framed as a continuum of ablation completeness,with about half of patients reaching statistical cure;NPVR and tumor vascularity determine chiefly whether-not how fast-recurrence occurs.The mixture cure model fits this structure better than conventional Cox regression and can inform patient selection and risk-stratified follow-up.

Objective This study aimed to quantitatively evaluate lumbosacral transitional vertebra (LSTV)-associated lumbosacral radiculopathy using diffusion tensor imaging(DTI).Additionally,based on multidimensional data analysis,primarily using DTI-derived parameters of compressed nerve roots,we investigated the relationship between bony structural variations of different types of LSTV and the corresponding lumbosacral nerve root compression syndrome. Methods This retrospective study included 365 patients with lumbosacral transitional vertebra. All patients underwent both routine MR and DTI-MR imaging of the lumbosacral nerve plexus.We measured the cross-sectional dimensions of the nerve root proper and nerve root tunnel,as well as the fractional anisotropy(FA) and apparent diffusion coefficient (ADC) values of the lumbosacral nerve roots.Statistical analysis was performed using the paired-sample t-test,one-way ANOVA,and binary Logistic regression. Results (1) For middle-segment nerve roots,the nerve root diameter,bony tunnel diameter,and FA values on the diseased side were significantly lower than those on the healthy side(P<0.05),while ADC values showed no significant differences;(2)On the affected side,the differences in diameter and FA values among the proximal,middle,and distal segments of the nerve root were statistically significant(P<0.05).Specifically,the diameter of the nerve root at the middle segment was significantly lower than that at the proximal and distal segments,and the FA value at the middle segment was significantly lower than that at the distal segment;(3)Particularly,significant correlations were observed between the minimum bony tunnel diameter,nerve root FA values,and nerve root injury in patients with type Ⅱ LSTV. Conclusion DTI can quantitatively evaluate microstructural changes in LSTV‑affected nerve roots.Quantitative assessment of nerve root tunnel cross-sectional dimensions can provide a reliable reference for evaluating LSTV-associated lumbosacral radiculopathy.

Objective To explore the diagnostic value of combining the dynamic Graph-Mamba Brain Network(GMBNet)model with resting-state functional magnetic resonance imaging(rs-fMRI)for pediatric autism spectrum disorder (ASD). Methods A retrospective analysis was performed on rs-fMRI images and clinical data of 607 children diagnosed with ASD and 568 healthy controls across 15 independent imaging sites.Transfer entropy combined with the sliding window technique was used to construct dynamic brain functional networks.The GMBNet was built by integrating graph neural networks and the Mamba architecture.Leave-one-site cross-validation was adopted to evaluate the overall performance of the model.Integrated Gradients interpretability analysis was conducted to locate key brain regions and abnormal connectivity patterns related to ASD. Results The GMBNet model demonstrated stable diagnostic efficacy across 15 independent imaging sites.The average accuracy of the entire dataset was 87.32%,and the average AUC was 0.863.Even at small sample sites,the model performance did not show significant degradation,suggesting that the model has good cross-center generalization ability.To verify the contribution of each module of the model,through ablation experiments,after removing the spatial Graph-Mamba block,the average AUC was 0.817;after removing the temporal Mamba block,the average AUC was 0.829;and after removing the graph convolution enhancement,the average AUC was 0.841.Interpretability analysis showed that the default mode network(DMN) presented hypoconnectivity,which was associated with social impairments in children with ASD;the salience network (SN) exhibited abnormal hyperconnectivity,corresponding to clinical symptoms including sensory hypersensitivity and emotional dysregulation. Conclusion The GMBNet established based on rs-fMRI has satisfactory diagnostic performance and generalization ability.The detected abnormal brain network changes are highly consistent with the neuropathological mechanisms of ASD.This model may serve as a potential auxiliary diagnostic tool for childhood ASD.

Objective To evaluate the safety,feasibility,and clinical efficacy of an innovative minimally invasive interventional technique—the through-and-through guidewire technique combined with the Perclose ProGlide™ vascular suture system—for the treatment of iatrogenic femoral artery pseudoaneurysm (iFAP)following failed ultrasound-guided compression therapy. Methods A retrospective analysis was conducted on patients diagnosed with iFAP at our hospital between January 2020 and February 2026.Patients who experienced failed ultrasound-guided compression and subsequently underwent interventional treatment were included.They were divided into two groups:the new technique group (treated with the through-and-through guidewire technique combined with the Perclose ProGlide™ system)and the conventional technique group.Clinical data,surgical approach,operative time,complications,intraoperative radiation dose,and procedure-related costs were compared between the two groups. Results The mean time from initial intervention to pseudoaneurysm development was 6.63 days in the new technique group and 4.00 days in the conventional technique group.According to CTA measurements,the mean pseudoaneurysm neck and sac diameters were 3.54 mm and 32.44 mm,respectively,in the new technique group,and 2.74 mm and 42.1 mm in the conventional technique group.Among the four patients in the conventional technique group,three underwent stent implantation and one received balloon occlusion combined with thrombin injection.In the new technique group,all eight patients were successfully treated,achieving a 100% technical success rate.The mean procedure-related cost was significantly lower in the new technique group(¥3,854.96 vs. ¥12,780.72,P<0.05).The mean radiation dose was also lower in the new technique group(216.06 mGy vs. 367.93 mGy,P<0.05).Postoperative ultrasound follow-up confirmed effective closure of the pseudoaneurysm neck and disappearance of vortex and blood flow signals within the sac.No major perioperative complications occurred.During a mean follow-up of 10.8 months,no pseudoaneurysm recurrence was observed. Conclusion The combination of the through‑and‑through guidewire technique and the Perclose ProGlide™ system demonstrates preliminary technical feasibility and safety in the management of iFAP.This technique offers a novel therapeutic alternative for elderly patients with complex iFAP.

Objective To investigate the safety and efficacy of direct angioplasty for acute intracranial atherosclerotic-related large-vessel occlusion(ICAS-LVO) and a positive microcatheter first-pass effect. Methods Clinical data of 26 consecutive ICAS-LVO patients who underwent direct balloon angioplasty/stenting between October 2017 and September 2024 in the Interventional Therapy Department of the First Affiliated Hospital of Dalian Medical University were retrospectively analyzed.Neurological functional outcome was assessed at 3 months using the modified Rankin Scale(mRS; mRS 0-2 defined as favorable outcome).Safety endpoints included symptomatic intracerebral hemorrhage,procedure-related complications,and 3-month mortality.Efficacy endpoints were successful recanalization (TICI 2b-3)rate and proportion of favorable outcome at 3 months. Results Successful recanalization was achieved in all 26 patients,with TICI 3 flow in 22 (85%).One patient (3.8%) experienced intraprocedural embolization.No other major procedure-related complications occurred.One patient died during hospitalization(mortality 3.8%).Among 25 survivors with 3-month follow-up,18(72%) had a favorable outcome. Conclusions For ICAS-LVO lesions with small thrombus burden selected via the "first-pass effect" of microcatheter,direct balloon angioplasty/stenting appears safe and feasible.Its shorter procedure time may potentially improve 3-month clinical outcomes in such patients.

Objective To assess the application value of combining deep learning reconstruction(DLR) with variable helical pitch(VHP) for low-dose “one-stop” coronary-carotid-cerebral CT angiography(CTA),compared to a conventional scanning protocol. Methods This prospective study enrolled 70 patients undergoing coronary-carotid-cerebral CTA in our institution,who were allocated to the routine-dose (RD)group(n=35)and low-dose (LD) group (n=35) according to block randomization.The RD group was performed separate coronary-carotid-cerebral CTA performed at 120 kVp with automatic tube current modulation;CT images of RD group were reconstructed with a hybrid iterative reconstruction (HIR)algorithms.The LD group was conducted as a one-stop VHP acquisition from the head to the heart at 100 kVp;the cardiac phase was ECG-gated and the head/neck phases were non-ECG gated;LD images were reconstructed with both HIR and DLR.Different noise indices were set for the two scanning protocols to further adjust the radiation dose,with the remaining scanning parameters being identical.The CT attenuation and standard deviation(SD)were measured at the mid-segment of the right coronary artery,the left Anterior Descending,the left Circumflex Artery,the aortic arch,the bilateral common carotid arteries,and the M1 segment of the left middle cerebral artery;the signal-to-noise ratio(SNR) and contrast-to-noise ratio (CNR)were also calculated.Two experienced radiologists independently assessed subjective image quality.One-way analysis of variance (ANOVA) was used for comparisons among the three groups;Mann-Whitney U test was applied for comparisons between the RD group and LD group-HIR/DLR;Wilcoxon signed-rank test was used for intragroup comparisons between the two subgroups of the LD group. Results Except for the mid-segment of the right coronary artery,which showed no significant intergroup difference,CT values of other arteries in the LD-DLR and LD-HIR images were significantly higher than those in the RD-HIR images (P<0.05).The LD-DLR demonstrated the lowest image noise at all measured vessels,with significantly higher SNR and CNR than the RD-HIR and LD-HIR groups,respectively(P<0.05 for all).Interobserver agreement for subjective image quality assessment was good,and no significant difference in subjective scores was observed between the RD-HIR and LD-DLR groups(P>0.05).Compared to the RD protocol,the LD protocol reduced the effective dose by 62.90% and iodine dose by 27.78%(both P<0.05). Conclusion The combination of DLR and VHP in one-stop head-neck-cardiac CTA can markedly reduce radiation exposure and contrast medium administration while maintaining image quality and diagnostic confidence,thereby showing promising potential for clinical implementation.

Objective To systematically sort out the research topics and evolutionary trends of artificial intelligence (AI) and deep learning in the field of lung cancer imaging diagnosis. Methods Relevant literature from January 1995 to April 2026 was retrieved from the Web of Science Core Collection,and bibliometric analysis was carried out using the BERTopic topic modeling approach. Results Research in this field can be summarized into six major directions:radiomics and AI-assisted medical image analysis,targeted/immunotherapy and molecular biomarker screening,low-dose CT screening and pulmonary nodule management,non-invasive detection and spectral biomarkers,rare thoracic tumors and metastasis detection,as well as tumor immune microenvironment and pathological evaluation.The evolution of research topics presents dual characteristics of "continuous deepening" and "phased outbreak".Among them,AI-driven multimodal modeling and imaging evaluation of immunotherapy response are the current core frontiers. Conclusions The research focus of AI and deep learning in lung cancer imaging diagnosis has evolved from "algorithmic recognition" to "remodeling the whole-process diagnosis and treatment decision-making of lung cancer with AI".Future research should expand data sources,integrate multilingual literature and full-text information,and focus on the prospective validation of multimodal and multi-omics integration,interpretable AI and clinical trust establishment,and the closed-loop system for the whole-process intelligent diagnosis and treatment,so as to promote the transformation of AI imaging diagnosis from technology-driven to clinically mature,and contribute intelligent solutions to achieve the cancer prevention and control goals of the"Healthy China 2030" initiative and global precision medicine for lung cancer.

Objective To develop an efficient and stable automated tilt correction process for head CT images around the Z-axis,in order to solve the problem of image standardization caused by head misalignment and improve image quality and diagnostic applicability. Methods An automatic tilt correction pipeline for head CT was proposed in this study.First,affine registration was applied for rapid rough alignment of the head.On this basis,fine adjustment of the transverse plane rotation around the Z-axis (cranio-caudal direction) was performed according to the falx cerebri,a stable midline anatomical structure.A total of 216 head CT images retrospectively collected from Nanjing Drum Tower Hospital were enrolled for efficacy evaluation.Taking manual measurements by radiographers as the gold standard,the angular error between the corrected angle and the actual angle was calculated.Meanwhile,two radiologists independently scored the quality of original and corrected images in a blinded fashion. Results The tilt angle obtained from the reference standard was(8.48°±6.09°),and the tilt angle obtained from the automated process was(8.52°±6.51°);the mean absolute error between the two was(0.71°±1.23°).Subjective image quality scores were significantly improved after correction(P<0.001).The proportion of excellent images(4 points)increased from 60.2% to 90.3%,and no cases of degraded image quality were observed. Conclusion This study established an automatic tilt correction pipeline integrating affine registration and falx cerebri features.The pipeline can rapidly and accurately correct the Z-axis tilt of head CT images and optimize image quality,presenting promising prospects for clinical application.

Objective To use quantitative computed tomography(QCT)technology to measure visceral fat area (VFA),liver fat content (LFC),and bone mineral density(BMD),and to investigate their correlation with the cardiometabolic index(CMI)in a population undergoing health examinations. Methods A total of 3303 individuals who underwent combined chest CT and QCT examinations at the Department of Health Management,Henan Provincial People's Hospital,between 2018 and 2024 were selected as study subjects.Participants were divided into low CMI,medium CMI,and high CMI groups based on CMI tertiles,with 1101 subjects in each group.Demographic data,anthropometric measurements,biochemical indicators,and QCT-measured VFA,LFC,and BMD data were collected,and CMI was calculated using the formula.Univariate analysis of variance(ANOVA)or the Kruskal-Wallis rank-sum test,Pearson and Spearman correlation analyses,and univariate and multivariate linear regression analyses were used to examine the associations between CMI and VFA,LFC,and BMD. Results Compared with the low- and medium-CMI groups,both VFA and LFC were significantly higher in the high-CMI group;BMD in the medium- and high-CMI groups was lower than that in the low-CMI group(P<0.05).Correlation analysis showed that CMI was significantly positively correlated with VFA and LFC(ρ=0.586,0.372;both P<0.001)and weakly negatively correlated with BMD(ρ=-0.086;P<0.001).In multivariate linear regression,the independent effect of CMI on VFA and LFC remained significant after comprehensive adjustment for confounding factors,whereas the association between CMI and BMD lost statistical significance after comprehensive adjustment. Conclusion CMI is independently and positively correlated with VFA and LFC;the association between CMI and BMD is mediated by factors such as age and gender.CMI has potential clinical utility in the early screening of individuals at high risk of metabolic disorders.