Objective To construct a nomogram model for predicting the isocitrate dehydrogenase (IDH) genotype of glioma based on the histogram features of intravoxel incoherent motion (IVIM) imaging. Methods A total of 66 patients with glioma who underwent conventional MRI and IVIM scans and were confirmed by surgical pathology were enrolled. Histogram features of IVIM-derived parameters—true diffusion coefficient (D),pseudo-diffusion coefficient (D^*),and perfusion fraction (f)—were extracted. Differences in clinical data,conventional MRI features,and IVIM histogram features between the IDH wild-type group and the mutant group were compared. Parameters were selected via the least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression to construct IDH genotype prediction models. Receiver operating characteristic (ROC) curve and nomogram were drawn,and the clinical utility and predictive accuracy of the model were evaluated using decision curve and calibration curve. Results There were statistically significant differences between the two groups in age,lesion location,boundary,T₂-FLAIR mismatch sign,diffusion restriction,enhancement,D_10th percentile,D_mean,D_median,f_10th percentile,f_90th percentile,f_energy,f_entropy,f_maximum,f_mean absolute deviation,f_mean,f_median,f_range,f_root mean squared,f_total energy,f_uniformity,f_variance,D^*_entropy,D^*_mean,D^*_median,and D^*_uniformity.And,the differences in lesion boundary,f_90th percentile,f_entropy,f_maximum,f_mean,f_range,f_root mean squared,f_uniformity,and D^*_uniformity after Benjamini-Hochberg correction remained statistically significant between the two groups (P<0.05). The area under the curve (AUC) values for predicting IDH genotype using D,f,D^*,and IVIM histogram models were 0.734,0.755,0.688,and 0.812,respectively. The sensitivities were 75.0%,75.0%,70.0%,and 85.0%,and the specificities were 65.2%,80.4%,69.6%,and 69.6%. The AUC values predicted by the combined model and the corrected combined model were 0.867 and 0.838,respectively,with sensitivities of 65.0% and 80.0%,and specificities of 97.8% and 80.4%. The C-index of the corrected combined model's nomogram was 0.839. Calibration curve and decision curve analysis confirmed that the nomogram's predicted probability aligned well with actual probability and showed favorable clinical utility. Conclusion The nomogram based on the IVIM histogram demonstrates good predictive performance for glioma IDH genotype and shows potential for preoperative non-invasive visualization in predicting glioma IDH genotype.

Objective To investigate microstructural changes in the thalamus of patients with Cervical spondylotic myelopathy (CSM) using diffusion kurtosis imaging (DKI) and perform comparison with diffusion tensor imaging (DTI). Methods A total of 43 CSM patients and 43 age-and sex-matched healthy controls were enrolled. Subgroup analyses were performed based on sensory function status and the presence of hyperintensity on spinal cord T₂-weighted imaging (T₂WI). All participants underwent brain diffusion magnetic resonance imaging (dMRI). Diffusion parameters were measured in the bilateral thalamus on DKI and DTI parametric maps. Intergroup differences were compared.Spearman correlation analysis was used to evaluate associations between imaging parameters and the modified Japanese Orthopaedic Association score (mJOA) as well as its sensory subscore (s-mJOA). Diagnostic performance was compared using receiver operating characteristic (ROC) curve analysis. Results Compared with 43 healthy controls,the 43 CSM patients showed significantly lower thalamic DKI_MK and DKI_FA,and significantly higher DKI_AK and DKI_RD (P<0.05). DKI_MK showed a strong positive correlation with the mJOA score (r=0.611,P<0.001) and an even stronger correlation with the s-mJOA (r=0.743,P<0.001). The AUCs of DKI_MK and DKI_AK for diagnosing CSM were 0.766 and 0.782,respectively,higher than that of DTI_FA (0.634),but the differences were not statistically significant (DeLong test,P>0.05). Subgroup analysis showed that DKI_MK yielded an AUC of 0.857 for identifying sensory dysfunction,which was significantly superior to DTI_FA (DeLong test,P<0.001). Conclusion DKI detects microstructural changes in the thalamus of CSM patients more sensitively than DTI.

Objective To evaluate the performance of photon-counting detector CT (PCD-CT) compared with energy-integrating detector CT (EID-CT) for visualizing the posterior chorda tympani nerve canal in the temporal bone. Methods Thirty-five patients who had undergone non-contrast EID-CT of the temporal bone (120 kV) at our institution within the previous 12 months and were scheduled for repeat non-contrast temporal bone CT for lesion evaluation or postoperative follow-up were prospectively enrolled.All patients underwent additional scanning on a PCD-CT system (120 kV).Temporal bone sides with destruction or defects of the posterior chorda tympani nerve canal or adjacent mastoid structures due to lesions or surgery were excluded.PCD-CT images were reconstructed with 0.2 mm and 0.6 mm slice thickness (PCD-0.2,PCD-0.6),and EID-CT images with 0.6 mm slice thickness (EID-0.6).Oblique sagittal reconstructions were generated for all image sets.The three image groups were compared in terms of CT attenuation,standard deviation (SD),signal-to-noise ratio (SNR),contrast-to-noise ratio (CNR) measured at the mastoid compact bone and external auditory canal air region on the axial plane at the level of the canal's pre-tympanic segment,as well as radiation dose.Two radiologists independently assessed the visibility of the proximal (from the facial nerve origin),middle,and distal (pre-tympanic) segments of the posterior chorda tympani nerve canal on oblique sagittal images using a 5-point scale. Results The mean age of the 35 patients was (42.51±14.74) years.Thirty temporal bone sides were excluded due to inflammation,mass,or postoperative changes,leaving 40 sides with intact bony structures (5 patients contributed bilateral sides) for analysis.The radiation dose of PCD-CT was 37% lower than that of EID-CT. Both PCD-0.2 and PCD-0.6 images showed higher absolute CT attenuation of air and bone than EID-0.6.In the air region,PCD-0.6 exhibited significantly lower image noise than EID-0.6 [(62.20±9.05）HU vs. (82.88±10.17)HU,P<0.001],and yielded higher SNR and CNR of the air region and compact bone than EID-0.6.By contrast,PCD-0.2 exhibited higher noise in the air region than EID-0.6 [(112.42±15.09）HU vs. （82.88±10.17）HU,P<0.001],and consequently lower SNR and CNR than EID-0.6 (all P<0.001).Subjectively,both PCD-CT image sets demonstrated superior visualization of all segments of the posterior chorda tympani nerve canal compared with EID-CT (all P<0.001),with PCD-0.2 showing better visualization of the distal (pre-tympanic) segment than PCD-0.6 (P<0.05). Conclusion Compared with EID-CT,PCD-CT achieves a 37% reduction in radiation dose while providing superior image quality and clearer visualization of the posterior chorda tympani nerve canal.The 0.2 mm slice thickness image offers particular advantages for demonstrating the distal (pre-tympanic) segment.

Objective Based on the deep learning model,using high-resolution vessel wall imaging (HR-VWI) precontrast T₁WI images to generate high-quality virtual contrast-enhanced T₁WI (vce-T₁WI) images,and evaluate their value in assessing the stability of carotid atherosclerotic plaques. Methods Pix2Pix generative adversarial network was implemented for the vce-T₁WI synthesis task. The similarity between vce-T₁WI and the real contrast-enhanced T₁WI images (CE-T₁WI) was evaluated using objective quality assessment and subjective visual quality scoring. Based on the Plaque-RADS classification score,the efficacy of vce-T₁WI in judging the stability of carotid atherosclerotic plaques was evaluated. Results This study enrolled 533 patients (comprising 1,066 carotid arteries) with paired non-contrast and contrast-enhanced images. The dataset was devided into a training/validation set (n=960) and a testing set (n=106). The real CE-T₁WI images and the vce-T₁WI images show a high concordance.Structural Similarity Index of 0.59,average Peak Signal-to-Noise Ratio of 22.04,and Histogram Mutual Information of 1.30. Four error metrics (NRMSE,SMAPE,Log-Accuracy Ratio,and Median Symmetric Accuracy),yielding mean scores of 0.07,25.17%,0.26,and 0.25. Furthermore,the Plaque-RADS scoring results on the vce-T₁WI images are highly consistent with those obtained from the real images. Conclusion The Pix2Pix-based deep learning facilitates the generation of high-quality vce-T₁WI from non-contrast carotid HR-VWI,demonstrating good diagnostic efficacy in the assessment of plaque stability.

Objective To explore the diagnostic value of the Kaiser score and quantitative/semi-quantitative hemodynamic parameters from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for breast BI-RADS category 4 lesions. Methods A retrospective analysis was conducted on 108 patients with a total of 129 lesions,all rated as BI-RADS category 4 based on breast DCE-MRI scans.Without knowledge of pathological results,the Kaiser score was assessed for each lesion,and both quantitative and semi-quantitative parameters were obtained.Quantitative parameters included volume transfer constant (K^trans),rate constant (K_ep),and extravascular extracellular space volume fraction (V_e).Semi-quantitative parameters included early enhancement rate (EER),maximum slope of increase (MSI),initial area under the gadolinium curve (iAUC),positive enhancement integral (PEI),and time to peak (TTP).Two-sample t-test or Mann-Whitney U test was used to analyze differences in each parameter between benign and malignant breast BI-RADS category 4 lesions.ROC curve analysis and Kappa consistency test were used to evaluate the consistency between each model and pathological findings to determine diagnostic performance. Results Among the 129 breast lesions categorized as BI-RADS 4,38 were malignant and 91 were benign.The malignant group showed significantly higher Kaiser scores,K^trans,K_ep,EER,and MSI,lower V_e,and shorter TTP than the benign group (all P<0.05).The DCE-MRI model was established by combining K^trans,K_ep,V_e,EER,MSI,and TTP.The Kaiser score had an AUC of 0.956,sensitivity of 100%,specificity of 75.8%,and an optimal cutoff value of 4.5.The DCE-MRI model had an AUC of 0.850,sensitivity of 89.5%,and specificity of 74.7%.The combined model of Kaiser score and DCE-MRI had an AUC of 0.966,sensitivity of 92.1%,and specificity of 90.1%.The AUC of the Kaiser score was significantly higher than that of the DCE-MRI model (Z=3.127,P=0.002),while no significant difference was found between the Kaiser score and the combined model (Z=-1.159,P=0.247).However,the combined model exhibited significantly higher specificity than the Kaiser score (P<0.001) and significantly higher AUC than the DCE-MRI model (Z=-3.969,P<0.05).Kappa consistency analysis showed that the agreement between the Kaiser score and pathological findings was 0.78 (95%CI:0.68-0.88),between the DCE-MRI model and pathology was 0.64 (95%CI:0.53-0.75),and between the combined model and pathology was 0.83 (95%CI:0.75-0.91). Conclusion The combined use of the Kaiser score and hemodynamic quantitative/semi-quantitative parameters (K^trans,K_ep,V_e,EER,MSI,TTP) can improve the diagnostic performance for breast BI-RADS category 4 lesions,offering high sensitivity and specificity in predicting malignancy.This approach has the potential to reduce unnecessary biopsies and holds clinical application value.

Objective To develop and validate interpretable multiple machine learning models using radiomic habitat features for preoperative risk stratification of thymic epithelial tumors (TETs). Methods This retrospective multicenter study included 145 patients with histologically confirmed TETs.Preoperative contrast-enhanced CT images and clinical data were collected.Conventional semantic features were also collected.Tumors were partitioned into three subregions (Habitat1-3) using K-means clustering.Radiomic features were extracted from the entire tumor (Intra) and from each subregion.After feature selection,selected features from each subregion were integrated into a combined habitat signature (Habitat-All).Six machine learning classifiers were trained on the six feature sets.Model performance was assessed using ROC curves,calibration curves,and decision curve analysis (DCA).Feature contributions were interpreted using Shapley additive explanations (SHAP). Results Models based on Habitat-All achieved AUCs (95%CI) ranging from 0.807 to 0.948 (0.722-0.985) in the training cohort,overall higher than those based on other feature sets.The extreme gradient boosting (XGBoost) model demonstrated the best performance,with an AUC (95%CI) of 0.948 (0.911-0.985) on the Habitat-All feature set.SHAP analysis identified original_firstorder_90Percentile_h2 as a major contributor to the output of the optimal model. Conclusion The XGBoost model with the Habitat-All feature set demonstrates good predictive performance for preoperative risk stratification of TETs,providing a noninvasive method to quantify intratumoral heterogeneity and an effective auxiliary tool for TET risk assessment.

Objective To investigate CT imaging characteristics and clinical outcomes of elderly patients with acute pancreatitis (AP) under different comorbidity burdens. Methods Clinical data,laboratory and CT findings,and Cumulative Illness Rating Scale (CIRS) scores were retrospectively collected from 265 AP patients.Based on CIRS scores,patients were categorized into three comorbidity groups:mild (1≤CIRS≤14),moderate (15≤CIRS≤18),and severe (CIRS≥19).CT features,disease severity,and clinical outcomes were compared among the three groups.Correlations of comorbidity status and CT findings with clinical outcomes were analyzed. Results The severe comorbidity group had significantly higher BISAP,EPIC,and MCTSI scores compared with the mild group (all P<0.05).The mild comorbidity group had lower age,shorter hospital stay,and lower APACHE Ⅱ scores than the other two groups (P<0.05).The severe comorbidity group had higher incidences of severe AP (SAP),ICU admission,and mortality (P<0.05). Conclusion AP patients with severe comorbidities show more pronounced peripancreatic inflammation,necrosis,and encapsulated necrosis on CT,and both clinical and imaging scores indicate more severe disease.Comorbidity status is closely correlated with pancreatitis severity,and clinical practice should emphasize the interaction between comorbidities and the pathophysiology of pancreatitis as well as their overall impact on the disease.

Objective To investigate CT and MRI characteristics of eosinophilic solid and cystic renal cell carcinoma (ESC-RCC). Methods Clinical,pathological,and imaging data of ten patients with ESC-RCC who underwent surgical resection and received pathological confirmation were retrospectively collected and analyzed.All ten cases underwent CT examinations,with six additionally undergoing MRI. Results The mean age of patients was (52.7±10.0) years,comprising six females and four males.All ten patients presented with unilateral solitary lesions.Tumor sizes ranged from 2.6 to 13.0 cm,with a mean diameter of (5.5±3.4) cm.On non-contrast CT,three cases showed slightly high density,four cases showed isodensity and low density,two cases showed isodensity and slightly high density,and one case showed isodensity.On non-contrast MRI,among the six cases,three showed signal intensity higher than the renal cortex on T₁WI,and three showed signal intensity lower than or equal to the renal cortex.On T₂WI,three cases showed signal intensity lower than the renal cortex,and three showed signal intensity equal to or higher than the renal cortex.On contrast-enhanced scans,six cases predominantly showed homogeneous enhancement,though the degree of enhancement remained consistently below that of the normal renal cortex.The remaining four cases showed heterogeneous enhancement.Two cases with cystic areas of varying sizes and two cases with large cystic areas showed progressive enhancement,but the cystic areas themselves did not enhance.Furthermore,tumors rarely showed hemorrhage or calcification. Conclusion ESC-RCC shows diverse imaging manifestations that require specific analysis based on the cystic-solid ratio.However,hemorrhage and calcification are uncommon in this tumor,and the enhancement pattern consistently shows lower intensity than the normal renal cortex at all phases.

Objective To explore the clinical value of the apparent diffusion coefficient (ADC) on magnetic resonance imaging combined with the ratio of free prostate-specific antigen (fPSA) to total prostate-specific antigen (tPSA) in evaluating the therapeutic effect of endocrine therapy for prostate cancer. Methods Clinical and imaging data of 150 prostate cancer patients diagnosed and treated with endocrine therapy at our hospital from January 2022 to December 2024 were retrospectively analyzed.Based on the therapeutic effect evaluation after treatment,patients were divided into a good response group and a poor response group.Pretreatment ADC values and fPSA/tPSA ratios were compared between the two groups.Receiver operating characteristic (ROC) curves were used to analyze the predictive value of each indicator alone and in combination for therapeutic efficacy.Multivariate Logistic stepwise regression was performed to identify independent factors influencing the therapeutic effect of endocrine therapy for prostate cancer. Results The poor response group had significantly higher testosterone and alkaline phosphatase levels,a higher proportion of patients with Gleason score≥8,TNM stage Ⅳ,and lymph node metastasis compared with the good response group (all P<0.05).Compared with the good response group,the poor response group showed significantly lower ADC values and fPSA/tPSA ratios (all P<0.05).ROC analysis showed that the AUCs of ADC value and fPSA/tPSA for predicting therapeutic efficacy were 0.804 (95%CI：0.759-0.854) and 0.831 (95%CI：0.781-0.876),respectively,and the AUC of their combination was 0.902 (95%CI：0.857-0.952).Multivariate analysis showed that TNM stage Ⅳ (OR=2.115,95%CI:1.246-3.590),Gleason score≥8 (OR=2.282,95%CI:1.303-3.997),ADC value<1.08×10^-3 mm²/s (OR=2.418,95%CI:1.559-3.751),and fPSA/tPSA<0.14 (OR=2.465,95%CI:1.617-3.756) were independent risk factors for poor therapeutic effect of endocrine therapy for prostate cancer (all P<0.05). Conclusion Pretreatment ADC value and fPSA/tPSA ratio are both effective indicators for predicting the therapeutic effect of endocrine therapy for prostate cancer.The combination of the two has the potential to further improve predictive efficacy.

Objective To investigate the application value of diffusion-weighted imaging (DWI) and perfusion-weighted imaging (PWI) in differentiating benign from malignant focal lesions in the peripheral zone of the prostate and in assessing disease severity. Methods A total of 104 patients with pathologically confirmed focal lesions in the prostatic peripheral zone who underwent examination at our hospital from October 2021 to September 2024 were retrospectively enrolled.Among them,51 had benign lesions (benign group) and 53 had prostate cancer (malignant group).All patients underwent DWI and PWI examinations,and the corresponding imaging parameters were measured.Disease severity in the malignant group was assessed based on Gleason grading, and patients were further divided into a low-risk subgroup (Gleason<7,n=29) and an intermediate-to-high-risk subgroup (Gleason≥7,n=24).Pearson correlation analysis was used to evaluate the relationships between DWI/PWI parameters and prostate cancer severity.Receiver operating characteristic (ROC) curves were used to assess the diagnostic value of DWI combined with PWI in differentiating benign from malignant lesions. Results Compared with the benign group,the malignant group showed significantly higher prostate-specific antigen (PSA) levels (21.49±3.26 ng/ml vs. 16.08±2.74 ng/ml),lower apparent diffusion coefficient (ADC) values [(0.81±0.15）×10^-3 mm²/s vs. （1.19±0.17）×10^-3 mm²/s],higher K^trans [(0.15±0.04） min^-1 vs. （0.10±0.03） min^-1], and higher K_ep [(0.52±0.10） min^-1 vs. （0.34±0.08） min^-1)] (t=9.144,12.070,7.190,10.112,respectively;all P<0.05).The area under the ROC curve (AUC) of combined DWI and PWI parameters for differentiating benign and malignant lesions was 0.905,significantly higher than that of individual parameters (Z=9.263,11.740,10.191;all P<0.05).In the intermediate-to-high-risk subgroup,ADC values were significantly lower [(0.71±0.14）×10^-3 mm²/s vs. （0.89±0.16）×10^-3 mm²/s],K^trans [(0.17±0.05 ）min^-1 vs. （0.13±0.03） min^-1] and K_ep [(0.65±0.12） min^-1 vs. （0.41±0.09） min^-1] were significantly higher than those in the low-risk subgroup (t=4.311,3.600,8.315,respectively;all P<0.05).ADC values were negatively correlated with Gleason scores,whereas K^trans and K_ep values were positively correlated with Gleason scores (all P<0.05). Conclusion The combination of ADC from DWI and K^trans/K_ep from PWI provides high diagnostic value for differentiating benign and malignant focal lesions in the prostatic peripheral zone.ADC,K^trans,and K_ep are closely associated with prostate cancer risk classification and Gleason scores,indicating significant potential for clinical application in assessing disease severity.

Objective To analyze chest and abdominal CT findings of three common pediatric lymphohematopoietic proliferative disorders presenting with acute pancytopenia,in order to provide insights for initial imaging differential diagnosis. Methods A total of 133 children treated at our hospital between May 2020 and May 2025 were enrolled,including 23 with malignancy-associated hemophagocytic lymphohistiocytosis (M-HLH) (Group 1),58 with acute leukemia (AL) (Group 2),and 52 with Epstein-Barr virus-associated HLH (EBV-HLH) (Group 3).The initial contrast-enhanced chest and abdominal CT findings at admission were blindly evaluated, and data were statistically analyzed using SPSS 26.0. Results Chest CT：The incidence of pulmonary lesions was higher in Group 3 than in the other two groups (P<0.001 and P= 0.003).Central interstitial thickening was more frequent in Group 1 than in the other two groups (P=0.015 and P<0.001). Pleural effusion was less common in Group 2 than in the other two groups (all P<0.0001).The short-axis diameter of the largest mediastinal lymph node was larger in Group 1 than in Groups 2 and 3 (all P<0.001);ROC curve analysis showed optimal cutoffs of 11.45 mm (Group 1 vs. Group 2) and 11.60 mm (Group 1 vs. Group 3).The distribution of the largest mediastinal lymph node in Group 1 differed from that in the other two groups (P=0.011 and P=0.032),with retrocaval nodes being more common.The short-axis diameter of the largest axillary lymph node was significantly different between Group 1 and Group 3 (P=0.027),with an optimal cutoff of 5.95 mm.Abdominal CT:The incidence of hepatomegaly did not differ significantly among the three groups (P=0.175).The incidence of periportal tracking sign was higher in the two HLH groups than in Group 2.Splenomegaly was more frequent in Group 3 than in the other two groups (P=0.006 and P<0.001).Ascites was less common in Group 2 than in the other two groups (all P<0.001).The incidence of focal renal lesions was lower in Group 3 than in the other two groups (P=0.076 and P=0.005).The short-axis diameter of the largest abdominal lymph node was larger in Group 1 than in the other two groups (P=0.001 and P=0.004);ROC curve analysis showed optimal cutoffs of 9.3 mm (Group 1 vs. Group 2) and 9.60 mm (Group 1 vs. Group 3).The incidence of iliac fossa lymphadenopathy in Group 3 was 0,which was significantly different from the other two groups (P=0.027 and P=0.013).The short-axis diameter of the largest inguinal lymph node did not differ significantly among the three groups. Conclusion Lymphohematopoietic proliferative disorders presenting with acute pancytopenia frequently involve thoracic and abdominal abnormalities,with distinct CT patterns that aid differential diagnosis.M-HLH is characterized by prominent central pulmonary interstitial thickening,significantly enlarged lymph nodes, and iliac fossa lymphadenopathy.EBV-HLH shows a high incidence of pulmonary parenchymal lesions,periportal tracking sign,and ascites,with rare occurrence of iliac fossa lymphadenopathy.AL shows relatively lower incidences of pulmonary lesions,pleural effusion/ascites,and periportal tracking sign,with typically milder lymphadenopathy.Recognition of these distinctive imaging patterns can provide crucial radiological evidence to support early clinical differential diagnosis.

Objective To develop and validate a risk prediction model for failure of the initial fetal magnetic resonance imaging (MRI) examination and to identify independent risk factors. Methods Clinical data from 442 pregnant women who underwent their first fetal MRI examination between March 2023 and March 2025 were retrospectively analyzed.Failure of the initial examination was defined as the primary outcome.A total of 14 variables across three domains—fetal,maternal,and technical factors—were included.Group comparisons were conducted using the Chi-Square test and Mann-Whitney U test.Least absolute shrinkage and selection operator (LASSO) regression was applied for variable selection,followed by multivariable Logistic regression to identify independent predictors and construct the prediction model. Model discrimination was evaluated using the receiver operating characteristic (ROC) curve,and internal validation was performed using the bootstrap method. Results The failure rate of the initial fetal MRI examination was 31.22% (n=138).Multivariable Logistic regression analysis identified polyhydramnios (OR=14.11,95%CI:4.79-41.52,P<0.001),nuchal cord (OR=5.09,95%CI:2.89-8.97,P<0.001),maternal education level of high school or technical secondary school (OR=3.48,95%CI:1.43-8.48,P=0.006),claustrophobia (OR=74.51,95%CI:8.83-628.86,P<0.001),underlying diseases (OR=8.43,95%CI:2.75-25.82,P<0.001),and technician experience of 100-500 cases (OR=4.72,95%CI:2.06-10.83,P<0.001) as significant predictors.The area under the curve (AUC) of the model was 0.896,indicating good discriminative ability.Internal validation yielded a bootstrap-corrected C-index of 0.866,demonstrating robust predictive performance. Conclusion The prediction model developed in this study demonstrated good discrimination and calibration,and can effectively identify pregnant women at high risk of failure of the initial fetal MRI examination,thereby providing an evidence-based basis for early preventive interventions.

Objective To investigate the optimization of coronary artery virtual reality (VR) imaging using a dual low-dose CT scanning protocol combined with different iterative weight reconstruction algorithms and to evaluate its clinical application value in digital subtraction angiography (DSA) surgery. Methods A retrospective study was conducted on 60 patients with coronary artery disease,who were randomly divided into a low-dose coronary VR group (Group A,70 kVp,300 mgI/ml,n=30) and a conventional-dose coronary VR group (Group B,120 kVp,370 mgI/ml,n=30).Independent sample t-test was used to analyze differences in continuous variables between the two groups.Mann-Whitney U test and Cohen's Kappa were used to evaluate inter-observer agreement for qualitative data.Receiver operating characteristic (ROC) curves were used to obtain the diagnostic performance of coronary VR and to compare relevant perioperative outcomes. Results Compared with Group B,Group A effectively reduced radiation dose by approximately 80% and contrast agent dose by approximately 45%,with no significant difference in coronary VR quality between the two groups.Coronary VR effectively reduced total surgical time,total radiation time,total radiation exposure,and contrast agent dose in assisting DSA surgery. Conclusion sThe dual low-dose CT technique combined with the optimal iterative weight reconstruction algorithm (SAFIRE 5) achieves optimization of radiation dose and contrast agent dose for coronary VR without compromising image quality.Coronary VR can effectively improve perioperative outcomes in DSA.

Objective To construct a multimodal prediction model integrating large language model (LLM)-derived radiology report features with deep learning-based imaging features for evaluating neoadjuvant chemotherapy (NAC) prognosis and metastasis risk in breast cancer patients. Methods This study retrospectively included magnetic resonance imaging data and corresponding case reports from 379 breast cancer patients.Feature extraction from radiology reports was performed based on fine-tuning of three mainstream LLMs (BERT,ChatGPT,and Qwen).These features were then fused with deep learning-based imaging features to construct multimodal prediction models for breast cancer prognosis and metastasis risk. Results In the NAC prognosis prediction task,the baseline ResNet50 model achieved an accuracy of 0.792,sensitivity of 0.754,F1-score of 0.772,and AUC of 0.760.After incorporating LLM features,ResNet50+Bert achieved an F1-score of 0.803,ResNet50+GPT 0.828,and ResNet50+Qwen showed the best performance with an accuracy of 0.884,sensitivity of 0.828,F1-score of 0.855,and AUC of 0.868.In the metastasis prediction task,the baseline ResNet50 model achieved an accuracy of 0.805,sensitivity of 0.801,F1-score of 0.803,and AUC of 0.815.After integrating LLM features,ResNet50+BERT achieved an F1-score of 0.824,ResNet50+GPT 0.841,and ResNet50+Qwen again achieved the best performance with an accuracy of 0.892,sensitivity of 0.876,F1-score of 0.883,and AUC of 0.862. Conclusion Systematic comparison of multimodal models revealed that incorporating LLM-derived radiology report features significantly improved predictive performance.This approach provides a robust tool for precision treatment planning and individualized clinical decision-making in breast cancer.

Objective To investigate the relationship between bone mineral density (BMD),different types of renal osteodystrophy (ROD),and bone metabolism markers in patients undergoing maintenance hemodialysis (MHD). Methods A total of 312 MHD patients were enrolled.Clinical data,biochemical parameters,and bone metabolism markers were collected.Lumbar spine BMD was measured using quantitative computed tomography (QCT).Based on intact parathyroid hormone (iPTH) levels and QCT BMD diagnostic criteria,patients were divided into low-turnover ROD group (n=141) and high-turnover ROD group (n=171).Additionally,bone mass status was classified as normal (n=197),osteopenia (n=79),or osteoporosis (n=36).Statistical analysis was performed accordingly. Results Patients in the low-turnover ROD group were significantly older and had significantly shorter dialysis duration,as well as lower BMD,alkaline phosphatase (ALP),calcium,and inorganic phosphorus levels compared with the high-turnover ROD group (all P<0.05).Multivariate Logistic regression analysis revealed that age (OR=1.077,95%CI：1.032-1.125),duration of dialysis (OR=0.992,95%CI：0.985-0.998),BMD (OR=1.013,95%CI：1.002-1.024),aspartate aminotransferase (AST) (OR=1.118,95%CI：1.023-1.223),ALP (OR=0.973,95%CI：0.963-0.983),and inorganic phosphorus (OR=0.234,95%CI：0.100-0.548) were independent influencing factors for low-turnover ROD (all P<0.05).Spearman correlation analysis showed that osteocalcin was negatively correlated with BMD,iPTH,and ALP (r=-0.343、-0.353、-0.481,P<0.001),while C-terminal telopeptide of type Ⅰ collagen (CTX) was positively correlated with these indicators (r=0.287、0.357、0.350,P<0.001). Conclusion In MHD patients,advanced age,shorter dialysis duration,higher BMD,elevated AST,low ALP,and low inorganic phosphorus levels are independent risk factors for low-turnover ROD.Early identification and dynamic monitoring of these patients may help prevent or delay the occurrence and progression of low-turnover ROD.

Objective To investigate the feasibility of using the deep learning-based Precise Image (PI) algorithm to reduce radiation dose and improve image quality in follow-up chest CT examinations for leukemia patients. Methods A total of 62 leukemia patients undergoing follow-up chest CT after chemotherapy were prospectively enrolled for image quality assessment.A low-dose CT scanning protocol was performed with 100 kV tube voltage,0.8 s gantry rotation time,automatic tube current modulation,and a Dose Right Index (DRI) of 4.Post-scan images were reconstructed using three algorithms:filtered back projection (FBP,Group A),hybrid iterative reconstruction iDose⁴ (level 3,Group B),and deep learning PI (DLIR,Group C).Lung kernel and soft tissue kernel were applied for each reconstruction algorithm,yielding six image sets per patient.Meanwhile,50 subjects who underwent routine-dose chest CT with iDose⁴ reconstruction were selected as the dose control group (Group D,DRI=19).Image quality among different low-dose reconstruction techniques was compared,along with radiation dose differences between the low-dose and conventional-dose groups. Results There were no significant differences in CT values among Groups A,B,and C in the low-dose cohort (P>0.05),whereas statistically significant differences were observed in standard deviation (SD),artifact index (AI),signal-to-noise ratio (SNR),and contrast-to-noise ratio (CNR) (P<0.01).Under identical convolution kernels,SNR and CNR decreased in the order Group C>Group B>Group A,while SD and AI increased in the order Group C<Group B<Group A,indicating that Group C had the best image quality.Significant differences in subjective image scores were found among the three groups (P<0.05),with substantial inter-observer agreement (Kappa=0.79,P<0.05).Group C achieved the highest score [3.7(3.3,4.1)],meeting diagnostic requirements;Group B ranked second [3.2(3.0,3.4)];Group A had the lowest score [2.3(1.9,2.7)] and failed to reach the diagnostic threshold.The mean volume CT dose index (CTDIvol),dose-length product (DLP),and size-specific dose estimate (SSDE) in the low-dose group were （1.35±0.22） mGy,（55.64±9.70） mGy·cm,and （1.78±0.16） mGy,respectively.Compared with the conventional-dose group,the average estimated dose saving rate was （80.61±2.61）%. Conclusion Low-dose chest CT combined with the PI deep learning reconstruction algorithm significantly reduces image noise and improves image quality while effectively reducing radiation exposure in leukemia patients.It can be recommended as the optimal protocol for follow-up chest CT examinations in this population.

Objective To analyze the application value of artificial intelligence (AI)-assisted low-dose CT thin-layer reconstruction in the differential diagnosis of solitary pulmonary nodules (SPN). Methods From December 2022 to November 2024,100 patients with SPN were prospectively and randomly enrolled.Pathological results confirmed 42 benign nodules and 58 malignant nodules,which were assigned to the benign group and malignant group,respectively.The mean independent film reading time of radiologists was recorded,and CT imaging features were compared.Binary Logistic regression analysis was performed to screen for independent risk factors for malignant SPN.Using pathological results as the gold standard,the diagnostic value for malignant SPN was analyzed. Results The mean film reading time and mean detection time under AI were 130 （124,141） s and 81 （78,89） s,respectively,both significantly shorter than the radiologists' film reading time (P<0.05).The proportions of nodules located in the upper lobe,lobulation sign,and vacuole sign were higher in the malignant group than in the benign group,while the proportion of calcification was lower in the malignant group (P<0.05).Binary Logistic regression analysis identified lobulation sign,spiculation sign,vessel convergence sign,and pleural indentation sign as independent risk factors for malignant SPN (P<0.05).Using pathological results as the gold standard,AI-assisted low-dose CT thin-layer reconstruction for identifying malignant SPN achieved an area under the curve (AUC) of 0.877,sensitivity of 89.66%,specificity of 85.71%,and accuracy of 88.00%.The AUC,sensitivity,and accuracy were significantly higher than those of low-dose CT thin-layer reconstruction alone (P<0.05). Conclusion AI-assisted low-dose CT thin-layer reconstruction demonstrates high efficacy for identifying the nature of SPN and has good auxiliary diagnostic value.

Objective Develop a TPZ delivery system based on nanobubbles(NBs) combined with ultrasound-targeted microbubble destruction(UTMD) and transcatheter arterial chemoembolization(TACE) for the treatment of hepatocellular carcinoma(HCC), to investigate its efficacy and safety. Methods TPZ-NBs were prepared using a high-shear dispersion method and characterized for their physicochemical properties, encapsulation efficiency(EE), and drug release profile. In vitro studies evaluated the effects of TPZ-NBs on HCC cell proliferation and migration under hypoxic conditions. A rabbit VX2 liver cancer model was established to compare the control,TAE alone,TPZ-NBs + TAE,and TPZ-NBs + UTMD + TAE groups regarding tumor necrosis, drug delivery efficiency, and safety. Results TPZ-NBs exhibited favorable EE (35.81%) and sustained release characteristics (cumulative release of 58.00% at 48 hours). In vitro experiments showed that 2 μg/ml TPZ-NBs had the best inhibitory effect under hypoxia and significantly inhibited the proliferation and migration of hepatoma cells. In animal studies, the TPZ-NBs + UTMD + TAE group significantly enhanced intratumoral drug delivery efficiency, induced tumor cell apoptosis, reduced microvessel density (MVD), and decreased tumor recurrence.Compared to other groups,this combination therapy achieved higher tumor necrosis rates and deeper drug penetration while maintaining a good safety profile. Conclusion The combination of TPZ-NBs,UTMD,and TAE represents an effective and safe therapeutic strategy that overcomes the limitations of conventional TACE,offering a novel potential approach for HCC treatment.