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

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

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

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

Objective To investigate the value of a radiomics model based on non-contrast CT scans in predicting hematoma expansion in patients with spontaneous intracerebral hemorrhage (SICH). Methods We retrospectively enrolled 393 SICH patients from two medical centers between January 2016 and March 2025.A total of 281 patients from the Shandong University of Traditional Chinese Medicine Affiliated Hospital were randomly divided into a training set (n=197) and an internal validation set (n=84) at a 7:3 ratio,while 112 patients from Shandong Provincial Qianfoshan Hospital served as an independent external test set.Inclusion criteria were:diagnosis of SICH,initial non-contrast CT performed within 6 hours of symptom onset,and a follow-up CT within 72 hours.Exclusion criteria included secondary intracranial hemorrhage,surgical intervention or death prior to the follow-up CT,hemorrhagic transformation of an ischemic cerebral infarction,and images with significant artifacts.A total of 393 eligible patients were enrolled and randomly allocated into training,validation,and test sets in a 7:2:1 ratio.Manual segmentation of the regions of interest was performed using 3D Slicer software.Features were extracted and selected using both traditional radiomics and a ViT-3D deep learning model.After feature fusion and selection,multiple machine learning classifiers—including Logistic Regression,Support Vector Machine,K-Nearest Neighbors,Random Forest,Extra Trees,Extreme Gradient Boosting (XGBoost),LightGBM,and Multi-layer Perceptron were built on the PixelMedAI platform.The models were optimized via five-fold cross-validation on the training set to identify the best-performing one. Results The model incorporating combined features demonstrated superior predictive performance across the training,validation,and test sets compared to other models.Its performance metrics were as follows:Training set (AUC 0.929,Sensitivity 0.923,Specificity 0.788,Accuracy 0.832); Validation set (AUC 0.961,Sensitivity 0.929,Specificity 0.946,Accuracy 0.940); Test set (AUC 0.927,Sensitivity 0.810,Specificity 0.923,Accuracy 0.902).Delong's test confirmed that the AUC of this model was statistically significantly different from those of the other five models. Conclusion A radiomics model based on non-contrast CT shows favorable performance in predicting early hematoma expansion in SICH.The LightGBM model,which integrates clinical,conventional imaging,and radiomic features,demonstrated the best predictive efficacy.

Objective To address the inefficiency and subjectivity of manual segmentation of knee cartilage,this study aimed to develop an automated segmentation model based on deep learning to enhance the accuracy and efficiency of early osteoarthritis(OA)diagnosis. Methods A spatial attention-enhanced Res-nnUNet model was proposed,which integrates residual connections and gated attention mechanisms to optimize feature fusion.The model was trained and validated using the SK10 public database(800 cases)and a clinical single-center dataset(249 knee MRI cases with T₁WI/PD sequences).A gold standard was established through double-blind labeling. Results On the test set,Res-nnUNet achieved a mean Dice similarity coefficient(DSC)of 89.0%,significantly outperforming comparison models(U-Net:77.9%,nnUNet:85.3%,ResUNet:79.7%;all P<0.01).The DSCs for femoral and tibial cartilage were(77.0±0.86)% and(84.7±1.30)%,respectively,with lower standard deviations compared with other models.Visualization demonstrated accurate identification of the femoral trochlear groove(error<0.5 mm)and thin posterior tibial plateau cartilage(Hausdorff distance 1.58 mm). Conclusion By leveraging spatial attention mechanisms,Res-nnUNet improves the segmentation accuracy and stability of fine cartilage structures.It provides a reliable automated tool for early quantitative assessment of OA and shows significant clinical potential.

Objective To observe the diagnostic accuracy of CT-based radiomics models in predicting severe acute pancreatitis(SAP)through Meta-analysis and explore their clinical value. Methods Relevant studies on the diagnostic value of CT-based radiomics models for predicting SAP were retrieved from PubMed,Embase,Wanfang Medical Network,VIP Database,and China National Knowledge Infrastructure(CNKI)from the establishment of each database to March 2025.Literature screening,quality assessment,and data extraction were performed,and Meta-analysis was conducted using RevMan 5.3 and Stata 16.0 software. Results A total of 9 studies were included.For the analysis of CT-based radiomics models alone,the pooled sensitivity was 0.82(95%CI:0.76-0.87),pooled specificity was 0.75(95%CI:0.66-0.83),pooled diagnostic odds ratio(DOR)was 14.23(95%CI:9.53-21.24),and pooled area under the curve(AUC)was 0.85(95%CI:0.81-0.89).For the comprehensive models combining CT-based radiomics with clinical indicators,the pooled sensitivity was 0.85(95%CI:0.77-0.90),pooled specificity was 0.89(95%CI:0.80-0.95),pooled DOR was 47.21(95%CI:25.89-86.07),and pooled AUC was 0.91(95%CI:0.90-0.92).The comprehensive models exhibited higher diagnostic accuracy than CT-based radiomics models alone.Both Deek's funnel plot and Egger's test indicated no publication bias among the included studies.Conclusion CT-based radiomics models show good predictive value for severe acute pancreatitis.However,more prospective,multi-center studies with both internal and external validation are required in the future to verify their actual clinical significance.

Objective To investigate the risk factors for hemoptysis recurrence in lung cancer patients with massive hemoptysis after interventional hemostatic surgery,and to construct and validate a prediction model,thereby providing a reference for clinical risk assessment and intervention strategies. Methods Clinical data of 362 lung cancer patients with massive hemoptysis who underwent interventional hemostatic surgery at the Second Affiliated Hospital of Navy Medical University from January 2023 to August 2025 were retrospectively collected.The patients were randomly allocated to a training set(n=254)and a validation set(n=108) at a 7∶3 ratio using a random number table.Both sets were further divided into recurrence and non-recurrence groups according to postoperative hemoptysis recurrence status.General data,tumor characteristics,surgical parameters,and laboratory indicators were collected and compared.LASSO regression was used to screen potential predictive factors,followed by multivariate Logistic regression to identify independent risk factors and construct a prediction model,which was visualized using a nomogram.The model's discrimination,calibration,and clinical utility were assessed using receiver operating characteristic(ROC)curves,calibration curves,and decision curve analysis(DCA). Results The hemoptysis recurrence rate in the training set was 25.59%(65/254).Multivariate analysis identified TNM stage,maximum tumor diameter,prothrombin time(PT),C-reactive protein(CRP),albumin(ALB),and creatinine(Cr)as independent risk factors for postoperative hemoptysis recurrence(all P<0.05).The prediction model based on these factors achieved an area under the curve(AUC)of 0.787(95%CI:0.723-0.851)in the training set and 0.738(95%CI:0.666-0.810)in the validation set.The Hosmer-Lemeshow goodness-of-fit test indicated good calibration in both the training set (χ²=2.851,P=0.094)and the validation set(χ²=4.126,P=0.127),with calibration curves fitting well against the ideal line(all P>0.05).DCA demonstrated a positive net clinical benefit within a threshold probability range of 0.05-0.80. Conclusion The prediction model constructed in this study demonstrates good predictive performance and clinical applicability for hemoptysis recurrence after interventional surgery in lung cancer patients.It can be used to identify high-risk patients and guide individualized interventions,thereby helping to reduce the risk of postoperative hemoptysis recurrence.

Objective To construct a predictive model combining MRI radiomics and clinical based on machine learning,and to explore its diagnostic value in clinically significant prostate cancer(csPCa). Method sA total of 180 patients who underwent prostate MRI were retrospectively included as study subjects,including 82 patients with csPCa and 98 patients without csPCa.They were randomly divided into a training group(n=125) and a test group(n=55) at a ratio of 7∶3.Diffusion-weighted imaging,apparent diffusion coefficient,and T₂-weighted imaging(T₂WI) images were segmented and feature extraction was performed.An imaging radiomics model was constructed through cascade feature selection;independent predictors of clinical data were screened,and clinical models and combined radiomics-clinical models were constructed,respectively.The performance of each model was evaluated using receiver operating characteristic curve(ROC)and area under the curve(AUC),and the net benefit of the model was assessed using decision curve analysis(DCA);a nomogram was drawn for the optimal model. Result sThirteen radiomics features most closely related to csPCa and two independent predictors were finally selected.ROC and AUC indicated that the combined model had the best diagnostic performance,with AUC values of 0.963 and 0.927 in the training and testing groups,respectively.DCA suggested that the combined model had a higher net benefit than other models;a nomogram was also drawn for the combined model. Conclusion The MRI radiomics combined with clinical model constructed based on machine learning has high diagnostic value for csPCa.

Objective To investigate the value of a multiphasic CT-based radiomics model combined with clinical risk factors for preoperative prediction of central cervical lymph node metastasis(CLNM)in patients with papillary thyroid carcinoma(PTC). Methods This retrospective study included 181 pathologically confirmed PTC patients, randomly divided into a training set(n=126)and a testing set(n=55)in a 7∶3 ratio.Univariate and multivariate Logistic regression analyses were performed to identify independent predictors of CLNM and establish a clinical model.Tumor volumes of interest(VOIs)were manually segmented using ITK-SNAP,and radiomics features were extracted from intratumoral and peritumoral(1 mm and 2 mm extensions)to construct corresponding radiomics models.Model performance was evaluated using the area under the receiver operating characteristic curve(AUC),sensitivity,specificity,and accuracy.The optimal radiomics model was integrated with the clinical model to develop a combined model.AUC differences were compared using the DeLong test,while calibration curves and decision curve analysis(DCA)were used to assess calibration and clinical utility. Results Multivariate analysis identified serum FT4 level（OR=1.207）and maximum tumor diameter（OR=1.103）as independent predictors of CLNM（P<0.05）.The intratumoral radiomics model demonstrated the highest AUCs in both the training and testing sets（0.804 and 0.817,respectively),outperforming the peritumoral 1mm model（AUCs：0.647 and 0.633）and the peritumoral 2 mm model（AUCs：0.681 and 0.669).The combined clinicalradiomics model achieved further improved AUCs of 0.871（training）and 0.851（testing),which were significantly higher than those of the standalone radiomics model（AUCs：0.804, 0.817）and the clinical model（AUCs：0.768, 0.763)（P<0.05）. Conclusion Intratumoral radiomics features derived from multiphasic CT can effectively predict CLNM in PTC.Integrating with serum FT4 level and maximum tumor diameter further enhances predictive performance,providing a noninvasive tool to aid preoperative risk stratification and support individualized surgical planning for PTC patients.

Objective To develop and validate a magnetic resonance imaging（MRI)-based scoring model for risk stratification of Ovarian-Adnexal Reporting and Data System（O-RADS）category 4 cystic-solid lesions. Methods A retrospective study was conducted on 137 patients with pathologically confirmed O-RADS 4 cystic-solid lesions who underwent pelvic mass resection.Preoperative MRI features were analyzed,including maximum lesion diameter,amount of ascites,morphology of solid tissue,signal intensity on high b-value diffusion-weighted imaging（DWI),and degree of contrast enhancement.Multivariable Logistic regression was used to identify independent predictors of malignancy and to construct a scoring model.The diagnostic performance of the model and individual predictors was assessed using receiver operating characteristic（ROC）curve analysis,with the DeLong test used for AUC comparisons. The Hosmer-Lemeshow test was employed to evaluate the model's calibration. Results The morphology of the solid component,high signal intensity on DWI,and marked contrast enhancement were identified as independent predictors of malignancy,with AUCs of 0.739,0.744,and 0.738,respectively.The scoring model incorporating these three features achieved a significantly higher AUC of 0.867（95%CI:0.796-0.937）for discriminating benign from malignant lesions.The Hosmer-Lemeshow test indicated good model fit（P=0.462). Conclusion The proposed MRI-based scoring model serves as an effective tool for distinguishing benign from malignant O-RADS category 4 cystic-solid ovarian-adnexal lesions,potentially aiding in more precise preoperative risk stratification.

Objective To investigate the application value of the test-bolus technique(TBT)combined with Smart mA and organ dose modulation(ODM)in computed tomography angiography(CTA)and CT perfusion(CTP)examinations for acute ischemic stroke, aiming to reduce radiation dose and contrast medium volume while ensuring the reliability of perfusion parameter measurements in normal brain tissue. Methods A total of 160 patients were prospectively enrolled and randomly divided into four groups.Groups A, B, and C adopted the test-bolus technique(10-15 ml contrast agent test to determine scan timing), followed by injection of 50 ml contrast agent to simultaneously complete CTA+CTP scans, with a total contrast agent dosage of 60-65 ml.Group D underwent conventional split scanning, with a total contrast agent dosage of 90-95 ml.Group A used Smart mA combined with ODM technology, Group B used Smart mA only, and Group C used fixed tube current.Radiation doses(DLP, ED), contrast agent dosages, image quality, and core cerebral perfusion parameters were compared among the four groups. Result sThere was no statistically significant difference in radiation dose between Group A[DLP:(1500.1±123.7)mGy·cm;ED:(3.19±0.25)mSv]and Group B[DLP:(1507.1±50.0)mGy·cm;ED:(3.21±0.11)mSv](P>0.05).However, both groups were significantly lower than Group C[DLP:(1950.6±10.9)mGy·cm;ED:(4.16±0.25)mSv]and Group D[DLP:(3127.6±33.4)mGy·cm;ED:(6.63±0.07)mSv](P<0.01).The contrast agent dosage in Group A was approximately 31.6% lower than that in Group D(P<0.05).There were no statistically significant differences in image quality or core perfusion parameters(CBF, CBV, MTT)of normal brain tissue among the four groups, and subjective evaluation results were consistent(P>0.05). Conclusion The combined protocol allows for significant reduction in radiation dose and contrast medium volume without compromising image quality.Notably, the ODM technique employed in Group A provides targeted protection for radiosensitive organs such as the ocular lens and thyroid by specifically reducing anterior tube current, which aligns with the ALARA (as low as reasonably achievable) principle.

Objective To explore the effectiveness of deep learning-based super-resolution dynamic enhancement MRI radiomics in predicting deep myometrial invasion of endometrial cancer. Methods A total of 212 patients with endometrial cancer who underwent surgical treatment were retrospectively collected and divided into superficial and deep myometrial invasion group based on pathological results.The original resolution dynamic enhancement MR images were converted into super-resolution images through super-resolution reconstruction technology,and radiomics features were extracted separately.A support vector machine algorithm was used to build models for original and super-resolution images.The performance of the models was evaluated using receiver operating characteristic curves and decision curve analysis. Results The diagnostic performance of the super-resolution model（area under the curve in training and testing set：0.877 and 0.872）was better than that of the original resolution model（area under the curve:0.787 and 0.717）and showed greater clinical benefits. Conclusion Deep learning-based super-resolution reconstruction technology can enhance the ability of dynamic enhancement MRI radiomics models to predict deep myometrial invasion in endometrial cancer,and may provide more accurate information for clinical treatment.

Objective To investigate the clinical utility of unsupervised diffusion model for synthesizing CT images from CBCT in radiotherapy,with the aim of reducing patient radiation exposure and enhancing adaptive radiotherapy precision. Methods This paper introduces the Wavelet-domain High-frequency Guided Diffusion model（WHGD).Based on the Daubechies-4 wavelet frequency domain framework for four-level decoupling of CBCT images, precisely separating low-frequency structures and high-frequency details to ensure anatomical integrity in synthesized CT images;a novel Sobel-VAE-GAN guidance mechanism that leverages gradient feature mapping and shared latent space to accurately recover CBCT high-frequency anatomical details during reverse diffusion;and unsupervised cross-domain generalization to enhance robustness in clinical radiotherapy scenarios.The model was trained and validated using public and in-house datasets（a total of 231 head/pelvis/chest pairs),which were split into training,validation,and test sets at a 7∶1∶2 ratio.Image quality was assessed using peak signal-to-noise ratio（PSNR),structural similarity index（SSIM),and mean squared error（MSE),with comparisons against six state-of-the-art baseline models. Results On the head dataset,WHGD achieved a PSNR of 29.975,SSIM of 0.939,and MSE comparable to the top baseline.Relative to CycleGAN,PSNR improved by 6.1% and SSIM by 2.1%;compared to MUNIT,PSNR increased by 8.1% and SSIM by 2.4%;versus the diffusion-based TPDM,PSNR rose by 42.0% and SSIM by 12.9%.For the pelvis dataset,WHGD yielded a PSNR of 28.171,SSIM of 0.947,and MSE of 0.00136,surpassing the second-best FGDM method with a 5.7% PSNR gain and 24.0% MSE reduction.On the chest dataset,WHGD attained a PSNR of 25.175,SSIM of 0.932,and MSE of 0.00222,outperforming all comparators.Visual evaluations highlighted WHGD's excellence in suppressing metal implant artifacts and preserving soft tissue fidelity.Ablation studies validated the efficacy and superior performance of individual modules. Conclusion This approach establishes an efficient paradigm for unsupervised CBCT-to-CT synthesis,holding promise for substantially mitigating patient radiation risks and advancing precision radiotherapy in clinical practice.

Objective To investigate the value of intratumoral and peritumoral magnetic resonance imaging (MRI)-based radiomics models in the preoperative assessment of parametrial invasion (PMI) in early-stage cervical squamous cell carcinoma. Methods Clinical and MRI data of 162 patients with cervical squamous cell carcinoma were retrospectively collected. Radiomics features were extracted from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). The least absolute shrinkage and selection operator (LASSO) algorithm was employed for feature selection. Intratumoral, peritumoral, and combined intratumoral-peritumoral radiomics signatures were constructed, and radiomics scores (Radscores) were calculated. Clinical, radiomics, and combined models were developed by incorporating independent clinical predictors. Model performance was evaluated using receiver operating characteristic (ROC) curves, cross-validation, calibration curves, and decision curve analysis (DCA). Results Maximum tumor diameter (MTD) and squamous cell carcinoma antigen (SCC-Ag) were identified as independent clinical predictors of PMI. ROC curve analysis demonstrated that the combined model achieved an area under the curve (AUC) of 0.898, which was superior to that of the intratumoral model (0.819), the peritumoral model (0.838), the intratumoral-peritumoral radiomics model (0.851), and the clinical model (0.762). Cross-validation and calibration curves indicated good model stability and calibration. DCA revealed that the combined model provided the highest net benefit in clinical decision-making. A web-based nomogram was developed based on the combined model to facilitate clinical application. Conclusion The combined model integrating intratumoral and peritumoral MRI radiomics with clinical factors can effectively predict PMI in early-stage cervical squamous cell carcinoma, potentially providing reference for preoperative risk stratification and individualized treatment.

Objective This study aimed to explore the value of habitat imaging and peritumoral-containing radiomics models based on dynamic contrast-enhanced MRI in predicting postoperative early recurrence（ER）of hepatocellular carcinoma（HCC). Methods Clinical and imaging data of 282 patients with HCC confirmed by postoperative pathology in the Henan Cancer Hospital were analyzed retrospectively.All patients were randomly divided into a training set（n=199）and a validation set（n=83）at the ratio of 7∶3.The region of interest（ROI）was drawn based on the tumor boundary and extended outwardly by 3 mm,5 mm,and 10 mm to obtain peritumoral information with different ranges;the tumor region was divided into three subregions using habitat imaging technology,radiomics features were extracted and screened. Logistic regression（LR）classifier and 10-fold cross-validation method were used for modeling and verification.Intratumoral,peritumoral-containing,and habitat radiomics models were established.and after univariate and multivariate Logistic regression analysis was used to screen the independent predictors and construct the clinical model.A combined model was developed by incorporating habitat model,optimal peritumoral-containing model and clinical independent predictors,and presented as a nomogram.ROC curves,decision curves,and calibration curves were used to evaluate model performance. Results Maximum tumor diameter（MTD)（P=0.009）and microvascular invasion（MVI)（P<0.001）were identified as independent predictors of early postoperative recurrence in HCC.The clinical model and intratumoral model demonstrated moderate predictive performance,achieving AUCs of 0.723 and 0.717,0.794 and 0.74 in the training and validation set,respectively.The peritumoral-containing radiomic model demonstrated superior performance within an extended range of 5 mm,with AUCs of 0.866 and 0.847 in the training and validation set,respectively.The Habitat model exhibited outstanding predictive performance,achieving AUCs of 0.902 and 0.903 in the training and validation set,respectively.The combined model achieved the highest AUCs（0.924 and 0.915 in the training and validation set,respectively).The combined model showed good predictive performance and clinical utility on both decision curve and calibration curve. Conclusion sHabitat imaging and peritumoral-containing radiomics models based on dynamic contrast-enhanced MRI can effectively predict the early recurrence after HCC resection,the combined model integrating the clinical factor can further enhances predictive performance,demonstrating optimal clinical benefit.

Objective To evaluate the ability of large language models（LLMs）combined with the Bone Tumor Risk Stratification and Management System（Bone-RADS）based on radiographs to distinguish benign and malignant bone tumors and provide histological diagnoses. Methods Bone tumor cases from two public medical imaging databases（Eurorad and Radiopaedia）and one radiology journal（Skeletal Radiology）were identified,with a deadline of September 30,2025.We collected the age,gender,clinical history,radiological descriptions,and pathological diagnoses of the cases.The DeepSeek-R1-0528 was used for testing,and was required to output Bone-RADS classification and one definitive diagnosis and three differential diagnoses,based on patients’ data and radiological descriptions.The performance of the LLM in distinguishing benign and malignant tumors,and making histological diagnoses was analyzed. Results A total of 511 cases were included in the study, among which 310（60.67%）were male.The median age（interquartile range）was 30.00（15.00,55.00）years.The primary lesion locations included the lower extremities（246 cases,48.14%),upper extremities（115 cases,22.50%),and multiple sites（49 cases,9.59%).Histological diagnoses were categorized as benign（277 cases,54.21%),intermediate（54 cases,10.57%),and malignant（180 cases,35.23%).For distinguishing benign and malignant tumors based on Bone-RADS classification generated by the LLM,the sensitivity was 0.876（95%CI：0.829-0.924),specificity was 0.699（95%CI：0.644-0.754),and the area under the receiver operating characteristic curve was 0.788（95%CI：0.751-0.824).Regarding histological diagnosis,the LLM directly achieved correct diagnoses in 306 cases（59.88%),achieved correct diagnoses among differential options in 84 cases（16.44%),and failed to make correct diagnoses in 121 cases（23.68%).Notably,among the cases that directly achieved correct diagnoses,14.91%（41/275）were misclassified in terms of benign/malignant nature excluding Bone-RADS 0 and Bone-RADS 5 cases. Conclusion LLMs have value in assisting with risk assessment of bone tumors on radiographs and suggesting potential histological diagnoses.However,there are certain differences between their diagnostic generation process and the diagnostic thinking of radiologists.Their output should be critically interpreted by radiologists.

Objective To evaluate the degree of pulmonary fibrosis and predict pulmonary function parameters in patients with connective tissue disease-related interstitial lung disease(CTD-ILD)using quantitative computed tomography(QCT)indicators. Methods Retrospectively enroll CTD-ILD patients diagnosed at the Affiliated Hospital of Yan'an University from January 2019 to April 2025.Patients were divided into mild group(Warrick score<8)and moderate-to-severe group(Warrick score≥8).Quantitative analysis was performed using 3D-Slicer software,and statistical analysis was conducted with R language. Results Quantitative CT indicators were significantly correlated with pulmonary function parameters (all P<0.05 except FEV1/FVC).Among them,Kurtosis had the strongest correlation with FVC%(r=0.36),and normal lung percentage(NL%)had the strongest correlations with DLCO%(r=0.39)and DLCO/VA(r=0.42).NL% showed the highest diagnostic value for moderate-to-severe patients(AUC=0.820).Kurtosis was the optimal predictor for FVC%(β=5.04,P=0.042),Skewness was the optimal predictor for DLCO%(β=19.67,P=0.013),and ground-glass opacity percentage(GGO%)was the optimal predictor for DLCO/VA(β=1.56,P=0.023).Scatter plots and restricted cubic spline(RCS)plots showed linear relationships between FVC%,DLCO%,DLCO/VA and their respective optimal quantitative CT predictors.Conclusion Quantitative CT can objectively and accurately predict the changing trends of pulmonary function,and exhibits high clinical application value especially when patients have poor cooperation or are unable to cooperate with pulmonary function tests.

Objective To investigate the imaging characteristics of malignant transformation in adult sacrococcygeal teratomas and to improve clinical diagnostic accuracy. Methods A retrospective analysis was performed on CT and MRI data from 18 adults with pathologically confirmed malignant transformation of sacrococcygeal teratoma.The morphological features and functional imaging characteristics were systematically evaluated and summarized. Results Morphological analysis revealed mixed cystic-solid lesions in 15 cases(83.3%,15/18),solid lesions in 2 cases(11.1%,2/18),and purely cystic lesions in 1 case(5.6%,1/18).Ill-defined margins were observed in 14 cases(77.8%,14/18).Regarding composition,typical fat components or calcifications were present in 9 cases(50.0%,9/18).Indicators suggestive of malignancy included ill-defined margins in 14 cases(77.8%,14/18),hyperintensity on diffusion-weighted imaging(DWI)in all 14 cases with ill-defined margins(100%),heterogeneous marked enhancement of solid/mixed cystic-solid components on contrast-enhanced scans in 13 cases(92.9%,13/14),and enhancement of septations and cyst walls in 1 cystic case(7.1%,1/14).Associated malignant manifestations included destruction of the sacrum and coccyx(27.8%,5/18)and pelvic/inguinal lymph node metastasis or peritoneal thickening(38.9%,7/18). Conclusion Malignant transformation in adult sacrococcygeal teratomas demonstrates characteristic imaging features in lesion composition,morphological characteristics,and functional imaging.Accurate recognition of these features provides an important basis for clinical formulation of individualized treatment plans.

Objective To investigate the imaging characteristics of ovarian Brenner tumors and to improve the differential diagnostic accuracy between benign and borderline/malignant forms. Method sA retrospective analysis was conducted on the clinical and imaging data of patients with pathologically confirmed Brenner tumors from five medical centers.Univariate and multivariate Logistic regression analyses were employed to identify independent risk factors associated with borderline/malignant Brenner tumors. Result sClinically, the incidence of elevated CA125 levels was significantly higher in the borderline/malignant group compared to the benign group（P<0.05）.Imaging analysis demonstrated that variables including maximum tumor diameter,tumor composition,tumor margin definition,presence of papillary projections,calcification type,degree of contrast enhancement,and CT attenuation values during arterial and venous phases were significantly different between benign and malignant tumors（P<0.05）.Multivariate Logistic regression identified maximum tumor diameter,presence of papillary projections,and CT attenuation during the venous phase as independent predictors of borderline/malignant Brenner tumors. Conclusion Larger tumor size,presence of papillary projections,and increased CT attenuation in the venous phase are indicative of a higher likelihood of borderline or malignant ovarian Brenner tumors and should be carefully considered in clinical evaluation.

Objective To explore the value of magnetic resonance imaging(MRI)in differentiating Duane retraction syndrome(DRS)from congenital fibrosis of the extraocular muscles(CFEOM). Methods A retrospective analysis was conducted on 64 pediatric patients with congenital strabismus who underwent MRI examinations between June 2015 and October 2024, including 35 patients with DRS and 29 patients with CFEOM.All patients underwent cranial nerve and orbital MRI examinations to observe abnormalities in the oculomotor nerve(CN3)and abducens nerve(CN6).The transverse diameters, longitudinal diameters, and cross-sectional areas of the extraocular muscles[medial rectus（MR）, lateral rectus （LR）, superior rectus（SR）, and inferior rectus（IR）]were measured and compared between the two groups. Result sAmong the 35 patients with DRS,34 had CN6 deficiency or dysplasia, and no obvious abnormalities were observed in CN3.Based on the presence of CN6 abnormalities,70 eyes of 35 patients were divided into the affected-side group(44 eyes)and the unaffected-side group(26 eyes).The transverse diameter of LR in the affected-side group was smaller than that in the unaffected-side group(P<0.05).All CFEOM patients had CN3 deficiency or dysplasia,12 of whom had concurrent CN6 deficiency or dysplasia.The transverse diameters, longitudinal diameters, and cross-sectional areas of MR, SR, and IR in the CFEOM group were significantly smaller than those in the DRS group(P<0.01).The longitudinal diameters and cross-sectional areas of LR in the CFEOM group were also significantly smaller than those in the DRS group(P<0.01).However, there was no significant difference in the transverse diameter of LR between the two groups(P>0.05). Conclusion The MRI features of DRS and CFEOM differ significantly.DRS is characterized by CN6 absence or dysplasia with mild atrophy of LR.CFEOM presents more complex manifestations, mainly characterized by CN3 absence or dysplasia and significant atrophy of the extraocular muscles.

Objective To develop and validate an interpretable, individualized prediction model integrating multi-regional radiomics features and body composition parameters for predicting microvascular invasion (MVI) in patients with hepatocellular carcinoma (HCC). Methods This retrospective study enrolled 248 patients with pathologically confirmed HCC from two independent medical centers. Patients from Center 1 were randomly allocated to a training set (n = 93) and an internal validation set (n = 41) at a 7∶3 ratio, while patients from Center 2 served as an independent external test set (n = 114). The endpoint was MVI status (positive vs. negative). Body composition parameters were measured on non-contrast CT images, and parameters associated with HCC-MVI were selected using univariate and multivariate logistic regression analyses. Subsequently, on portal venous phase CT images, the tumor region was segmented, and multiple concentric regions were generated by expanding outward from the tumor boundary at 2-mm intervals (4-14 mm), resulting in a total of 13 regions of interest (ROIs) grouped into three categories: tumor region, peritumoral region, and combined tumor-peritumoral region. Radiomics features were extracted from these ROIs. Optimal feature subsets were constructed using multiple feature selection methods, and corresponding radiomics models were built using a support vector machine (SVM) classifier. The SHAP (SHapley Additive exPlanations) method was employed to visualize the contribution of features in the better-performing models. Finally, an individualized prediction model was constructed by integrating selected body composition parameters and the radiomics models. Results Among the multi-regional radiomics models constructed, the ROI12 model (combined tumor and 12 mm peritumoral region) demonstrated optimal performance across all datasets, with AUC values of 0.918 (95% CI: 0.862-0.974), 0.841 (95% CI: 0.720-0.961), and 0.790 (95% CI: 0.707-0.872) in the training, internal validation, and external test sets, respectively. SHAP analysis revealed the importance of features included in this model. The body composition parameter model, comprising subcutaneous fat area and visceral fat area, achieved AUC values of 0.775 (95% CI: 0.682-0.868), 0.686 (95% CI: 0.524-0.849), and 0.764 (95% CI: 0.678-0.850), respectively. The integrated model significantly outperformed the single models in the training set, internal validation set, and external test set (all P < 0.05 by DeLong's test), with AUC values of 0.949 (95% CI: 0.906-0.993), 0.922 (95% CI: 0.838-1.000), and 0.868 (95% CI: 0.803-0.932), respectively. A nomogram was developed for visualization. Calibration curves and decision curve analysis demonstrated good predictive consistency and clinical utility across all datasets. Conclusion Across multiple centers, the individualized prediction model integrating tumor, peritumoral, and body composition parameters effectively predicts the presence of MVI in HCC, supporting precision medicine in clinical practice.

Objective To develop and validate a deep learning model integrating Vision Transformer (ViT) architecture with radiomics features for preoperative noninvasive prediction of perineural invasion (PNI) in esophageal squamous cell carcinoma (ESCC). Methods This retrospective study enrolled 197 patients with histopathologically confirmed ESCC from two medical centers. Patients from center 1 (n = 133) constituted the training set, and those from center 2 (n = 64) constituted the test set. Radiomics features, including topological and fractal features, were extracted from preoperative contrast-enhanced CT images. Concurrently, deep learning features were extracted using the ViT architecture. Five models were constructed: clinical model, radiomics model, ViT model, deep learning-radiomics (DLR) fusion model, and comprehensive model. Results In the training set, the area under the curve (AUC) values for the clinical, radiomics, ViT, DLR, and comprehensive models were 0.541, 0.700, 0.785, 0.858, and 0.855, respectively. The DLR model demonstrated optimal performance, significantly outperforming the single-modality models (all P < 0.01). In the test set, the AUC values for the clinical, radiomics, ViT, DLR, and comprehensive models were 0.669, 0.585, 0.657, 0.640, and 0.662, respectively. Conclusion The DLR fusion model integrating ViT and radiomics features enables effective preoperative noninvasive prediction of PNI status in ESCC patients, providing a valuable decision-support tool for individualized clinical management.

Objective To explore the predictive value of coronary CT angiography（CCTA）for major adverse cardiovascular events（MACE）in patients with type 2 diabetes mellitus（T2DM）. Methods A retrospective analysis was conducted on 150 patients with T2DM admitted to our hospital from June 2021 to June 2024.Patients were followed up for 12 months, and the occurrence of MACE was recorded.Based on follow-up results, patients were divided into two groups：69 patients who experienced MACE were included in the MACE group, and 81 patients who did not experience MACE were included in the non-MACE group.All patients underwent coronary CT angiography.Baseline characteristics and CCTA parameters were compared between the two groups.Receiver operating characteristic（ROC）curve analysis was used to evaluate the predictive value of CCTA parameters for MACE in T2DM patients. Result sThere was no statistically significant difference in baseline characteristics between the two groups（P>0.05）.The MACE group had a significantly higher proportion of patients with severe obstructive stenosis and high-risk spotty calcification.In addition, total plaque volume, plaque length, plaque burden, and plaque stenosis rate were all significantly greater in the MACE group compared to the non-MACE group（P<0.05）.However, the minimum lumen area and CT-derived fractional flow reserve（CT-FFR）were significantly smaller in the MACE group than in the non-MACE group（P<0.05）.Logistic regression analysis revealed that stenosis severity, total plaque volume, plaque length, plaque burden, plaque stenosis rate, and high-risk spotty calcification were all risk factors for MACE in T2DM patients, whereas minimum lumen area and CT-FFR were protective factors.ROC curve analysis showed that the AUC values for stenosis severity, minimum lumen area, total plaque volume, plaque length, plaque burden, plaque stenosis rate, high-risk spotty calcification, and CT-FFR were 0.678,0.692,0.726,0.741,0.758,0.792,0.659, and 0.778, respectively.DeLong test results demonstrated that the combined prediction model had a higher AUC value than any individual parameter（P<0.05）. Conclusion Patients with MACE had a significantly higher proportion of severe obstructive stenosis and high-risk spotty calcification, along with significantly larger total plaque volume, plaque length, plaque burden, and plaque stenosis rate, but significantly smaller minimum lumen area and CT-FFR.The combination of coronary CT angiography parameters demonstrates high predictive value for MACE in patients with type 2 diabetes mellitus.

Objective To explore the value of artificial intelligence（AI）-based quantitative computed tomography（CT）features in predicting the invasiveness of nodular and mass-type lung adenocarcinoma. Methods Preoperative high-resolution chest CT data of 1218 patients with pathologically confirmed nodular and mass-type lung adenocarcinoma were retrospectively analyzed, including 1052 patients in the training cohort（491 in the non-invasive group and 561 in the invasive group）and 166 patients in the validation cohort（98 in the non-invasive group and 68 in the invasive group）.Sixteen quantitative parameters, including histogram features, entropy, and lesion size, were automatically extracted using AI-based pulmonary nodule analysis software.The Mann-Whitney U test was used to compare differences in parameters between the non-invasive and invasive groups.Variables with area under the curve（AUC）>0.750 were selected by receiver operating characteristic（ROC）curve analysis and further entered into binary Logistic regression analysis to identify independent predictors.The predictive performance of single-factor and multi-factor models was also evaluated. Result sIn the training cohort, AI-extracted CT quantitative features showed significant differences between the non-invasive and invasive groups（P<0.05）.ROC curve and Logistic regression analyses identified mean CT attenuation, entropy, and long diameter as independent predictors of invasiveness.When the thresholds were set at mean CT attenuation>-400.8 HU, entropy>8.0, and long diameter>10 mm, the sensitivity and specificity were 63.99% and 79.23%,75.40% and 82.28%, and 81.82% and 74.75%, respectively.The combined model achieved an AUC of 0.918, with sensitivity of 82.35% and specificity of 88.19%, significantly higher than those of single indicators.In the validation cohort, the predicted probability calculated by the Logistic regression model（AUC=0.939）also demonstrated superior performance compared with mean CT attenuation（AUC=0.807）, entropy（AUC=0.874）, and long diameter（AUC=0.882）. Conclusion AI-based quantitative CT feature analysis can effectively assess the invasiveness of nodular and mass-type lung adenocarcinoma.Mean CT attenuation, entropy, and long diameter are major independent predictive indicators.Combined analysis of these three parameters can significantly improve predictive accuracy, providing an important reference for preoperative invasiveness assessment and individualized treatment strategy formulation.

Objective To evaluate the performance of a CT-based habitat radiomics model combined with clinical factors for predicting short-term response to transarterial chemoembolization (TACE) in patients with intermediate-to-advanced hepatocellular carcinoma(HCC). Method sThis retrospective study included 147 patients with intermediate-to-advanced HCC who underwent preoperative contrast-enhanced CT and were randomly divided into a training set(n=102) and a testing set(n=45). Tumor habitats were identified using K-means clustering, with optimal cluster number determined by the silhouette coefficient and sum of squared errors, followed by whole-cohort segmentation using a Gaussian mixture model. Radiomics features were extracted from each habitat subregion and selected using intraclass correlation coefficient analysis,Spearman correlation,maximum relevance minimum redundancy,and least absolute shrinkage and selection operator regression. Habitat-based radiomics models and a fusion model incorporating independent clinical predictors were constructed. Model performance was evaluated using receiver operating characteristic curves, calibration curves, and decision curve analysis. Result sAlpha-fetoprotein, alanine aminotransferase, and platelet count were independent clinical predictors of TACE response. The habitat-combined radiomics model achieved AUCs of 0.865 and 0.802 in the training and testing sets, respectively. After integration with clinical factors, the fusion model showed improved performance, with AUCs of 0.882 and 0.814. Decision curve and calibration analyses demonstrated good clinical utility and agreement. Conclusion A CT-based habitat radiomics model combined with clinical factors enables effective prediction of short-term response to TACE in patients with intermediate-to-advanced HCC and may assist individualized treatment decision-making.

Objective s To explore the feasibility of predicting the efficacy of induction chemotherapy (IC) in patients with nasopharyngeal carcinoma (NPC) based on pre-treatment tumor T-stage and magnetic resonance imaging (MRI) features. Methods A total of 238 patients with biopsy-confirmed NPC were included. All patients received IC and were categorized as responders or poor responders according to the Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST 1.1). The following clinical characteristics and MRI features were assessed: age,sex,T-stage,N-stage,clinical stage,IC regimen; involvement of retropharyngeal lymph nodes (RLNs) and cervical lymph nodes (CLNs),their locations,nodal grouping (NG),lymph node necrosis,extranodal extension (ENE),and carotid space involvement (CSI). The area under the receiver operating characteristic curve (AUC),sensitivity,specificity,and accuracy were used to evaluate the model's performance in predicting IC efficacy. A 10-fold cross-validation approach was employed to assess the diagnostic performance and generalizability of the predictive model. The DeLong test was used to compare differences in diagnostic efficacy between the model and each independent risk predictor. Results Of the 238 patients,124 were classified as responders and 114 as poor responders. Multivariate Logistic regression analysis revealed that T-stage (OR = 0.096,95% CI: 0.026-0.349,P < 0.001),CSI (OR = 0.125,95% CI: 0.053-0.298,P < 0.001),and ENE of cervical lymph nodes (OR = 0.039,95% CI: 0.018-0.085,P < 0.001) were independent predictors of IC response. The model constructed based on these independent predictors achieved an AUC of 0.867 (95% CI: 0.817-0.907),which was significantly higher than that of each individual predictor (all P < 0.001). The model demonstrated a sensitivity of 0.855 (95% CI: 0.780-0.912),a specificity of 0.754 (95% CI: 0.665-0.830),and an accuracy of 0.807 (95% CI: 0.751-0.855). The mean AUC obtained through 10-fold cross-validation was 0.872 (95% CI: 0.836-0.909). Conclusion A model incorporating tumor T-stage and MRI features shows potential for predicting the efficacy of IC in NPC before treatment,which may provide valuable guidance for developing personalized therapeutic strategies.

Objective To explore the clinical value of deep learning reconstruction (DLR) combined with black-blood technique in evaluating morphological characteristics of carotid atherosclerotic plaques,using contrast-enhanced high-resolution magnetic resonance imaging of the carotid artery vessel wall (HRMRI) as the reference standard. Methods We retrospectively collected data from 20 patients with carotid atherosclerosis who underwent both carotid computed tomography angiography (CTA) and HRMRI within a two-week interval between December 2024 and May 2025.A total of 40 plaques were included.CTA and black-blood images were reconstructed using DLR and hybrid iterative reconstruction (HIR) algorithms,respectively,forming four groups: D-black blood,D-CTA,H-black blood,and H-CTA.Using HRMRI as the reference standard,morphological characteristics including the degree of lumen stenosis,maximum wall thickness (Max WT),normalized wall index (NWI),and fibrous cap status were evaluated.Diagnostic performance for NWI > 0.56,severe lumen stenosis,and thin and/or ruptured fibrous cap was assessed.The Wilcoxon signed-rank test and Friedman test were used to compare measurements among groups. Results The degree of carotid lumen stenosis in the D-black blood group was significantly higher than that in the D-CTA group (Z = 2.634,P = 0.008).The D-black blood group demonstrated a sensitivity of 1.000 and specificity of 0.971 for detecting severe lumen stenosis,outperforming the H-black blood group (sensitivity 1.000,specificity 0.912).Both Max WT and NWI values were significantly higher in black-blood groups than in CTA groups (P < 0.001).For diagnosing NWI > 0.56,both D-black blood and H-black blood groups achieved a sensitivity of 1.000 and specificity of 1.000,superior to the D-CTA group (sensitivity 0.895,specificity 1.000) and H-CTA group (sensitivity 0.868,specificity 1.000).The D-black blood group showed a sensitivity of 1.000 and specificity of 0.950 for detecting thin and/or ruptured fibrous caps,exceeding the H-black blood group (sensitivity 1.000,specificity 0.750). Conclusion CTA black-blood images provide high diagnostic value for evaluating lumen stenosis,maximum wall thickness,plaque burden,and fibrous cap characteristics.DLR combined with black-blood technique enables better assessment of plaque and lumen morphological features and is suitable for clinical application.

Objective To validate the diagnostic performance of magnetic resonance elastography(MRE)in Chinese patients with chronic liver disease(CLD),quantify the interference effect of hepatic inflammation on MRE stiffness values,and establish stage-specific diagnostic thresholds tailored to the characteristics of this population. Methods A total of 242 CLD patients undergoing MRE and liver biopsy were retrospectively enrolled.A stratified merging strategy was applied to optimize statistical power in small-sample staging.Spearman correlation and multivariate linear regression were used to analyze the independent effects of fibrosis stage(F0-F4),inflammation grade(G0-G4),and demographic variables on MRE values,with stage-specific diagnostic thresholds constructed accordingly. Results MRE stiffness values showed a strong positive correlation with fibrosis stage(ρ=0.916,P<0.001).The AUCs for diagnosing≥F2,≥F3,and F4 fibrosis were 0.992,0.967,and 0.994,respectively,with optimal cutoff values of 2.77 kPa,3.46 kPa,and 4.19 kPa.Multivariate regression revealed that G3-4 inflammation significantly increased MRE values by 0.325 kPa(P=0.003),while age and sex showed no significant effects(P>0.05).Fibrosis stage independently explained 80.01% of the variation in MRE values.Conclusion This study establishes a stage-specific MRE diagnostic threshold system for the Chinese population,confirming that liver stiffness primarily reflects fibrosis severity.However,moderate-to-severe inflammation may lead to approximately 9.1% overestimation of MRE values,underscoring the need to integrate inflammation activity into clinical interpretation for accurate staging correction.

Objective To investigate the utility of habitat imaging (HI) based on multiparametric magnetic resonance imaging (mpMRI) for the noninvasive prediction of cervical cancer (CC) risk. Methods This retrospective study included 247 patients with pathologically confirmed CC treated. According to the NCCN guidelines and Sedlis criteria, patients were divided into high-risk and low-risk groups. Using simple random sampling, patients were assigned to a training set (n = 173) and a test set (n = 74) at a 7∶3 ratio. Tumor lesions were segmented into k habitat subregions via K-means clustering on integrated mpMRI data, including DCE-MRI and IVIM sequences. Habitat maps were constructed, and the proportion of each subregion was calculated. Logistic regression was employed to identify independent risk factors and to develop clinical (tumor marker), HI-based, and combined clinical-HI predictive models. 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 model performance. Results Three optimal habitat subregions were identified. DCE-MRI-derived habitats (H1) were defined as follows: H1.1 (high wash-in, high wash-out), H1.2 (high wash-in, low wash-out), and H1.3 (low wash-in, low wash-out). IVIM-derived habitats (H2) included: H2.1 (low f, D, and D^* values), H2.2 (low f, high D and D^* values), and H2.3 (high f and D, low D^* values). In both the training and test sets, the low-risk group exhibited a significantly lower proportion of H1.1 and higher proportions of H1.2 and H2.3 compared with the high-risk group (P < 0.05). Additionally, in the training set, the low-risk group showed a significantly higher proportion of H1.3 and a lower proportion of H2.1 than the high-risk group (P < 0.05). Logistic regression identified SCC-Ag, CA125, and H1.1 as independent risk factors for high-risk CC (OR = 1.074, 1.030, and 3.625; P = 0.031, 0.018, and 0.003, respectively), whereas H2.3 was a protective factor (OR = 0.009, P = 0.024). The combined clinical-HI model achieved the highest predictive performance, with AUCs of 0.814 and 0.818 in the training and test sets, respectively. In the training set, the AUC differed significantly between the clinical model and the clinical-HI model (P = 0.012), and between the clinical model and the clinical-H2.3 model (P = 0.042); however, no significant difference was observed between the clinical-HI model and the HI model. Conclusion The clinical-HI model provides a robust and noninvasive tool for preoperative risk stratification in cervical cancer. By integrating imaging-derived habitat features with clinical biomarkers, this approach enables accurate prediction of tumor risk, offering potential to optimize surgical planning, guide adjuvant therapy, reduce overtreatment, and ultimately improve patient outcomes.

Objective To explore the predictive efficacy of virtual magnetic resonance elastography (vMRE) based on diffusion-weighted imaging (DWI) for lymphovascular space invasion (LVSI) status in cervical cancer patients undergoing primary surgery. Methods Clinical and imaging data of patients who underwent radical hysterectomy at Henan Provincial People's Hospital from November 2021 to October 2022 were prospectively collected. All patients underwent preoperative pelvic MRI examinations,including conventional sequences and multi-b-value DWI. vMRE images were generated based on DWI data,and μ_diff parameters were extracted. Patients were divided into LVSI-positive group and LVSI-negative group according to postoperative pathological results. The t-test or Mann-Whitney U test was used to compare parameter differences between the two groups. Logistic regression analysis was performed to screen variables associated with LVSI status,and a predictive model was constructed. Receiver operating characteristic (ROC) curves were plotted,and the area under the curve (AUC) was used to evaluate the predictive efficacy of each model for LVSI. Results A total of 69 cervical cancer patients were finally included. Among clinical variables,FIGO stage was significantly correlated with LVSI. The mean,maximum,median,and minimum values of μ_diff in the LVSI-positive group were significantly higher than those in the LVSI-negative group (P < 0.05). The model constructed by combining the minimum μ_diff value with FIGO stage showed the best diagnostic efficacy,with an AUC of 0.804,sensitivity of 0.756,specificity of 0.750,and accuracy of 0.754,which was superior to the single-index models.Conclusion The μ_diff values generated based on vMRE exhibit good diagnostic efficacy,indicating that this technique has great potential for predicting the postoperative LVSI status of cervical cancer patients and can serve as a valuable clinical auxiliary assessment tool.