BMC Medical Imaging[JOURNAL]
Sun
200 papers
CT-based radiomics to predict peri-device leakage after left atrial appendage closure.
Lu X, Zhang J, Li J … +13 more , Ma W, Wang B, Guo W, Zeng D, Wang C, Chu Y, Gao C, Fang W, Yang Z, Niu X, Wen H, Wang Q, Li Y
BMC Med Imaging · 2026 Apr · PMID 42056934 · Full text
BACKGROUND: Peri-device leakage (PDL) is one of the major complications of left atrial appendage closure (LAAC). However, there is a lack of predictive models in clinical practice. The aim of this study was to develop an...
BACKGROUND: Peri-device leakage (PDL) is one of the major complications of left atrial appendage closure (LAAC). However, there is a lack of predictive models in clinical practice. The aim of this study was to develop and validate a prediction model for PDLs after LAAC based on preoperative CT and CT-based radiomic features. METHODS: This retrospective cohort study included 100 patients with nonvalvular atrial fibrillation who underwent LAAC between August 2023 and August 2024 at two centers. A clinical model was constructed via binary logistic regression with clinical information, and a radiomic model was constructed via conventional CT measurements and radiomic features. A combined model was also constructed by combining clinical information and imaging features. The performance of all the models was evaluated and compared, and internal validation was performed via the bootstrap method. RESULTS: Multivariate analysis revealed that the least axis length, larger diameter, and hypertension grade 3 were independent risk factors for PDLs. The combined model constructed based on these three factors (AUC: 0.796, 95% CI: 0.704–0.888) was superior to the clinical model (AUC: 0.711, 95% CI: 0.605–0.817) and the radiomic model (AUC: 0.743, 95% CI: 0.641–0.845). CONCLUSIONS: This study is the first to find that the least axis length of the LAA was an independent risk factor for the PDL. The combined model demonstrated a high degree of reliability in predicting the PDL at 3 months following LAAC. This model exhibited a greater predictive capacity than the clinical model and radiomic model did.
Unsupervised radiomics-driven endotyping of chronic rhinosinusitis with nasal polyps: a multimodal characterization of CT imaging, clinical phenotypes, and proteomic profiling.
Zhao H, Wang Q, Hao Y … +9 more , Yu P, Li J, Wang J, Zhang H, Zhang Y, Yang Y, Mou Y, Mao N, Song X
BMC Med Imaging · 2026 Apr · PMID 42056929 · Full text
BACKGROUND: The current phenotyping of chronic rhinosinusitis with nasal polyps (CRSwNP) into eosinophilic (eCRSwNP) and non-eosinophilic (non-eCRSwNP) subtypes is increasingly insufficient to address complex clinical ch...
BACKGROUND: The current phenotyping of chronic rhinosinusitis with nasal polyps (CRSwNP) into eosinophilic (eCRSwNP) and non-eosinophilic (non-eCRSwNP) subtypes is increasingly insufficient to address complex clinical challenges, especially as some non-eCRSwNP patients have poor prognoses. Identifying intrinsic endotypes non-invasively is crucial for precision therapy. To identify CRSwNP endotypes via unsupervised clustering of paranasal sinus computed tomography (CT) radiomics features and analyze their clinical and biological significance. METHODS: Retrospective study of CRSwNP patients undergoing functional endoscopic sinus surgery (FESS) (Jan 2016-Apr 2021) with preoperative CT. Clustering analysis was performed on patients using 1409 radiomic features. Proteomics analyzed nasal polyps from 41 patients. Clinical characteristics and prognoses were compared. RESULTS: In total, 661 patients were included (median age, 50 years; interquartile range, 40–60 years; 213 [32%] women). Three radiomic clusters were identified. Endotype 3 had the worst prognosis, followed by endotype 1; endotype 2 had the best prognosis (log-rank test, P = 0.026 / 0.0026). Endotype 3 exhibited the lowest lymphocyte count (Kruskal-Wallis test, P = 0.049), highest CT scores (P < 0.0001) and nasal endoscopic scores (P < 0.0001). Endotype 3 showed enrichment in inflammatory pathways (complement activation, immune signaling, humoral response). Endotype 2 correlated with lipoprotein processes. Endotype 1 (intermediate prognosis) showed co-enrichment of lipoprotein and inflammatory pathways. CONCLUSION: Unsupervised CT radiomics clustering identified three prognostically distinct CRSwNP endotypes with differing clinical and biological features. This provides a novel non-invasive method for endotyping and prognostic assessment.
Should pre-measurement physical activity be standardized in muscle thickness and stiffness evaluations? - A randomized controlled four arm cross-over study.
Warneke K, Plöschberger G, Oraze M … +2 more , Jochum D, Siegel SD
BMC Med Imaging · 2026 Apr · PMID 42056910 · Full text
BACKGROUND: High standardization is of crucial relevance for reliability in imaging diagnostics. When quantifying muscle properties (muscle thickness and stiffness) by ultrasound or myotonometry, internal validity can be...
BACKGROUND: High standardization is of crucial relevance for reliability in imaging diagnostics. When quantifying muscle properties (muscle thickness and stiffness) by ultrasound or myotonometry, internal validity can be compromised by examiner-related factors and participant biologic variability. A frequently neglected source of bias is pre-measurement activity, which may acutely alter muscle perfusion and muscle blood inflow. METHODS: The acute influence of different physical activity routines on tissue parameters was investigated in 30 healthy participants (16 m, 14f). Ten minutes before, immediately before, immediately after and 10 min retention of cycling, jogging, calf raises or control, muscle thickness and stiffness measurements via shear wave elastography (SWE) and myotonometry were measured. RESULTS: Reliability was excellent for muscle thickness (ICC = 0.94-1.00; CV = 1.7-9.1%), good-excellent for SWE stiffness (ICC = 0.68-0.97; CV = up to 26% for inter-day) and myotonometry (muscle ICC = 0.77-0.98; CV = 4.0-17% tendon 0.86-0.93 (CV = 11-17%). Muscle thickness significantly increased after calf raises (d = 1.60, 10.3%) and jogging (d = 0.60, 3.0%), without effects after cycling or control. Shear-wave elastography showed muscle stiffness decreased after calf raises (d=-0.73, -16.7%). Myotonometry indicated a stiffness increase (d = 1.04, 20.1%). The 10-minute retention showed consistent effects for muscle thickness (d = 0.80, 5.3%) and stiffness (SWE: d = 0.78, 21.1%, myotonometry: d=-0.82, -13.0%). CONCLUSION: Pre-measurement activity could systematically affect muscle thickness and stiffness with dependence on activity type and intensity. This highlights the importance of monitoring pre-measurement activity to minimize potential reliability issues as this, depending on several potential moderators, could enhance the random error if within sample pre-measurement activity is not standardized. Before ultrasound evaluation, for some activity (i.e. calf raises), > 10 min of rest was required to diminish this bias.
Assessment of cardiac allograft vasculopathy in heart transplant patients using multidimensional dynamic CTA and principal components analysis.
Zhang X, Yang M, Huang T … +4 more , Qin Q, Qian P, Luo Y, Wang J
BMC Med Imaging · 2026 Apr · PMID 42050431 · Full text
BACKGROUND: Cardiac allograft vasculopathy (CAV) is a major cause of late graft failure post heart transplantation. While coronary angiography remains the gold standard, non-invasive techniques, such as CT angiography (C...
BACKGROUND: Cardiac allograft vasculopathy (CAV) is a major cause of late graft failure post heart transplantation. While coronary angiography remains the gold standard, non-invasive techniques, such as CT angiography (CTA), are emerging alternatives. Electrocardiogram-gated multidimensional dynamic CTA (MD CTA) allows to track dynamic motions of coronary artery throughout the cardiac cycles, potentially revealing valuable insights into coronary abnormalities. METHODS: Principal component analysis (PCA) is employed to analyze the left anterior descending artery (LAD) motion, aiming to assess CAV in heart transplant patients. The motions were determined through registration of MD CTA images, and the incremental displacement of LAD between adjacent phases in a complete cardiac cycle was used as input in PCA. Two-sample t-test and logistic regression were used to compare and differentiate the control and CAV group based on PCA results, and a linear regression was used to correlate PCA results with the degree of stenosis. RESULTS: The resulted contribution rate of the first principal component (PC1) in control group (0.61 ± 0.05) is significantly higher than the value observed in CAV group (0.46 ± 0.06, p < 0.05). A univariate logistic model (AUC = 0.97) based on contribution rate can sharply discriminate the control and CAV group. Importantly, a negative correlation was found between the contribution rate of PC1 and the degree of stenosis in CAV group. CONCLUSION: This study employs PCA and multidimensional CTA to analyze LAD dynamic motion for assessment of CAV. The contribution rate of the first principal component (PC1) was identified as a promising indicator for evaluating CAV and tracking stenosis progression. These findings offer a quantitative, non-invasive approach that may enhance clinical decision-making in post heart transplantation care.
Amyloid-beta statuses prediction with free water MR imaging features in Alzheimer's disease using machine learning models.
Zhou R, Sun X, Chen S … +12 more , Zhao S, Zhao C, Qu J, Zeng W, Li C, Zhang X, Li Z, Wang Y, Zhang T, Xu X, Jia J, Liang Y
BMC Med Imaging · 2026 Apr · PMID 42046023 · Full text
PURPOSE: This study aimed to identify the effectiveness of free water MRI (FW-MRI) features for predicting amyloid-beta (Aβ) statuses in Alzheimer’s disease (AD) by constructing diagnostic models using machine learning a...
PURPOSE: This study aimed to identify the effectiveness of free water MRI (FW-MRI) features for predicting amyloid-beta (Aβ) statuses in Alzheimer’s disease (AD) by constructing diagnostic models using machine learning analysis. METHODS: This study retrospectively included 96 patients of mild cognitive impairment (MCI) and AD (69 Aβ-positive and 27 Aβ-negative). Clinical characteristics, FW-corrected and standard diffusion indices, and structural MRI indices were collected. Three supervised machine learning algorithms, including random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB), were adopted to construct a diagnostic model for distinguishing Aβ deposition in AD. SHapley Additive exPlanation (SHAP) value was used as an interpretable algorithm to identify influential characteristics based on the best-performing model. RESULTS: In the single-modality model, FW-DTI achieved better classification performance than conventional DTI, which obtained accuracies all above 80% among three machine learning approaches on the internal dataset (RF = 0.800, SVM = 0.867, XGB = 0.800). In the multi-modality model, the XGB model integrated FW-DTI, voxel-based morphometry, and clinical features outperformed the RF and SVM models, achieving an accuracy of 86.7% and an area under the curves (AUC) value 93.2% in the training cohort, and an accuracy of 77.8% and AUC value of 83.1% in the external testing cohort. The model demonstrated high sensitivity but relatively low specificity, indicating a tendency toward positive predictions. Furthermore, FW-DTI indices were shown to have the highest predictive value for Aβ deposition. CONCLUSION: Integrating FW-DTI with structural and clinical features effectively differentiated Aβ positivity in AD, with FW-DTI indices contributing the highest predictive risks, demonstrating the potential of FW-DTI in AD diagnosis.
Associations among thyroid function, white matter microstructure, and emotional disturbances in thyroid eye disease.
Zhang Z, Zhang H, Song X … +5 more , Song Y, Zhou H, Zhu L, Tao X, Jiang M
BMC Med Imaging · 2026 Apr · PMID 42046014 · Full text
OBJECTIVES: To characterize white matter microstructural abnormalities in thyroid eye disease (TED) using tract-based spatial statistics (TBSS) of diffusion tensor imaging (DTI) and diffusion kurtosis imaging (DKI), and...
OBJECTIVES: To characterize white matter microstructural abnormalities in thyroid eye disease (TED) using tract-based spatial statistics (TBSS) of diffusion tensor imaging (DTI) and diffusion kurtosis imaging (DKI), and to explore their associations with thyroid function and emotional disturbances. METHODS: 24 TED patients and 27 age-, sex-, and education-matched healthy controls (HC) were enrolled. DTI- and DKI-derived maps were analyzed using TBSS to obtain voxel-wise measures of fractional anisotropy (FA) and mean kurtosis (MK). Correlation analyses were performed among diffusion metrics, thyroid function indices, and psychological state assessment questionnaire scores. Mediation analysis was conducted to further understand the relationships between these variables. RESULTS: Compared with DTI, TBSS analysis of DKI revealed more extensive white matter abnormalities in TED after correction for multiple comparisons. TED patients showed significantly reduced MK values in multiple predominantly left-sided white matter tracts and the corpus callosum, whereas FA showed no significant corrected differences. Lower MK values were associated with lower thyroid function levels and greater emotional disturbances. Exploratory mediation analyses identified significant indirect effects linking thyroid function with emotional disturbances through altered white matter tracts. CONCLUSIONS: TED is associated with subtle white matter microstructural abnormalities detectable by DKI, particularly in predominantly left-sided tracts. Altered white matter integrity showed exploratory associations with thyroid hormone levels and emotional disturbances and may represent a potential correlate of their relationship in TED.
The diagnostic value of virtual unenhanced spectral CT images in acute pancreatitis: a comparison with true unenhanced images.
Wang H, Lin L, Yu H … +4 more , Fan C, Sun A, Yang Y, Sun H
BMC Med Imaging · 2026 Apr · PMID 42045865 · Full text
OBJECTIVE: To investigate whether virtual unenhanced (VUE) images generated from spectral CT using the rapid kVp switching technique can serve as a substitute for true unenhanced (TUE) images in the assessment of acute p...
OBJECTIVE: To investigate whether virtual unenhanced (VUE) images generated from spectral CT using the rapid kVp switching technique can serve as a substitute for true unenhanced (TUE) images in the assessment of acute pancreatitis (AP). METHODS: This retrospective study included 90 patients diagnosed with AP according to the Revised Atlanta Classification who underwent unenhanced non-spectral CT scan and dual-phase spectral CT scans of the abdomen. VUE images for the arterial (AP-VUE) and venous phases (VP-VUE) were reconstructed using material decomposition techniques. Objective and subjective assessments were conducted on TUE, AP-VUE, and VP-VUE images. The objective evaluation included CT attenuation, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). The subjective evaluation was performed by three readers independently and assessed image quality, lesion identification, severity grading of pancreatitis, and diagnostic confidence. RESULTS: CT attenuations differences between the TUE and VUE images were not significant for AP lesion (P = 0.110 (Pancreatitis) & 0.052 (Peripancreatic effusion)), with over 76.67% of anatomical regions showing differences within 10 HU. Compared with VP-VUE and AP-VUE images, TUE images demonstrated higher SNR and CNR overall. Diagnostic confidence scores were comparable across TUE, AP-VUE, and VP-VUE (confidence scores from 5(4, 5) to 5(5, 5), P from 0.439 to 1.000) and the diagnostic conclusions showed no significant differences as well (P > 0.05). The Balthazar CT grading system (A–E) demonstrated high concordance (Intraclass correlation coefficients > 0.8) across all image types, with no difference in AP severity assessment (P > 0.05). CONCLUSION: The diagnostic value of VUE images in AP is comparable to that of TUE images, suggesting that VUE images can serve as an alternative to reduce radiation exposure.
Exploring the key clinical and computed tomography features for distinguishing high-grade lung cancers from morphologically similar benign tumors: a two-center case-control study.
Zhao M, Gan H, Ding C … +3 more , Zhang WT, Lv FJ, Chu ZG
BMC Med Imaging · 2026 Apr · PMID 42045864 · Full text
BACKGROUND: Some high-grade lung cancers (HGLCs) and pulmonary benign tumors (PBTs) exhibit similar computed tomography (CT) features. This study aims to determine the key clinical and CT features for identifying HGLCs....
BACKGROUND: Some high-grade lung cancers (HGLCs) and pulmonary benign tumors (PBTs) exhibit similar computed tomography (CT) features. This study aims to determine the key clinical and CT features for identifying HGLCs. METHODS: From 2011 to 2025, 102 HGLCs presenting as regular solid nodules (SNs) and 102 size-matched PBTs were retrospectively enrolled. Clinical data and qualitative and quantitative CT indicators of lesions were analyzed and compared. Firth’s penalized logistic regression was used to identify independent predictors, and model performance was evaluated using 5-fold cross-validation. RESULTS: Compared to patients with PBTs, those with HGLCs were more likely to be male, older, and symptomatic individuals and smokers, with greater smoking amount (all P < 0.05). HGLCs also demonstrated higher rates of mediastinal/hilar lymph node enlargement, vascular encasement, and bronchial obstruction, while calcification was absent (all P < 0.001). The male (odds ratio [OR], 42.340; 95% confidence interval [CI], 2.283 -785.149; P = 0.012), high pack-years (> 27.6) (OR, 1.161; 95%CI, 1.022–1.321; P = 0.022), mediastinal/hilar lymph node enlargement (OR, 208.631; 95% CI, 6.477–6720.607; P = 0.003), bronchial obstruction (OR, 34.375; 95% CI, 2.314–196.064; P = 0.010), and vascular encasement (OR, 39.650; 95% CI, 9.089–67.752; P = 0.002) were revealed as independent predictors of HGLCs. CONCLUSIONS: In male heavy smokers, SNs with morphological features of PBTs but exhibiting vascular encasement, bronchial obstruction, or mediastinal/hilar lymph node enlargement warrant closer monitoring or further evaluation.
Diagnostic performance of conventional ultrasound and shear wave elastography of the parotid glands in primary Sjögren's syndrome.
Zhao S, Zhang W, Huang H … +3 more , Xiao D, Zhang S, Mao F
BMC Med Imaging · 2026 Apr · PMID 42045848 · Full text
OBJECTIVE: To evaluate the diagnostic performance of the OMERACT scoring system and shear wave elastography (SWE) of the parotid glands (PG) for primary Sjögren’s syndrome (pSS). METHODS: A total of 190 subjects were enr...
OBJECTIVE: To evaluate the diagnostic performance of the OMERACT scoring system and shear wave elastography (SWE) of the parotid glands (PG) for primary Sjögren’s syndrome (pSS). METHODS: A total of 190 subjects were enrolled, comprising 79 patients diagnosed with pSS according to the 2016 ACR/EULAR classification criteria and 111 non-pSS individuals. After obtaining OMERACT scores (0–3) and SWE values using parotid gland ultrasonography, we defined an adjusted OMERACT score as the original score plus 1 point (not exceeding 3) when the PG SWE mean value exceeded the optimal cut-off of 12.3 kPa. The diagnostic performance of the OMERACT score, SWE mean, and the adjusted OMERACT score was then assessed. RESULTS: Receiver operating characteristic (ROC) curve analysis revealed that the area under the curve (AUC) for the OMERACT score in predicting pSS was 0.825, while the AUC for PG SWE mean was 0.852, with optimal cut-off values of 1 and 12.3 kPa, respectively. The corresponding sensitivities were 79.7% and 89.9%, and the specificities were 74.8% and 78.4%. Using this adjustment, the adjusted OMERACT score achieved an AUC of 0.912, with an optimal cut-off of 2, a sensitivity of 75.9%, and a specificity of 91.9%. The adjusted OMERACT score effectively distinguished the pSS group from the non-pSS group. CONCLUSIONS: The OMERACT scoring system and SWE demonstrate considerable clinical value in diagnosing pSS. They provide a practical and effective non-invasive approach that can be integrated into the diagnostic workflow, offering a more convenient and efficient examination for patients.
Fine-tuned lightweight language models for structured extraction of liver cancer imaging free-text report: a comparative analysis with existing large language models.
Luo YD, Sun Y, Tang SL … +12 more , Shen LJ, Zou X, Zou HL, Shen HL, Yang TT, Wu JY, Lu RY, Li CF, Huang JH, Cai J, Jiang YQ, Ren G
BMC Med Imaging · 2026 Apr · PMID 42035041 · Full text
BACKGROUND: Organizing free-text patient data into a structured format is labor-intensive and time-consuming. This study aims to evaluate the effectiveness of a fine-tuned lightweight language model in structuring liver...
BACKGROUND: Organizing free-text patient data into a structured format is labor-intensive and time-consuming. This study aims to evaluate the effectiveness of a fine-tuned lightweight language model in structuring liver cancer imaging reports. METHODS: A retrospective dataset of 2,780 liver imaging reports from Sun Yat-sen University Cancer Center (2012–2022), including cases of primary liver cancer and benign liver disease, was collected. Three key entries—Number of Malignant Tumors (NMT), Diameter of the Largest Tumor (DLT), and Vascular Invasion (VI)—were annotated by three radiologists and subsequently reviewed and calibrated by a senior oncologist to ensure data reliability. The annotated dataset was randomly split into training, validation, and test sets at a ratio of 7:1:2. A T5-based lightweight model with 250 M parameters (Liver-T5) was fine-tuned using these data. Performance was evaluated using Accuracy and Macro-F1 metrics. Comparative analysis with LLMs such as ChatGLM4, Qianwen2.0, and Llama3.1 was conducted. RESULTS: The fine-tuned Liver-T5 model outperformed larger LLMs in Exact Match (EM) rate and key evaluation metrics, achieving an EM of 0.8907 and high accuracy for NMT (0.9355) and VI (0.9910). Specifically, for NMT extraction, Liver-T5 achieved an accuracy of 0.9355, outperforming large models such as Qianwen72B (accuracy 0.9140), LLaMA3 (accuracy 0.8961), and ChatGLM4 (accuracy 0.8226). In the VI extraction, Liver-T5 achieved the highest accuracy of 0.9910, significantly surpassing other models, with Qianwen72B, LLaMA3, and ChatGLM4 achieving accuracies of 0.9606, 0.9462, and 0.7581, respectively. A higher proportion of schema-nonconforming outputs was observed in large general-purpose models (e.g., LLaMA3), while Liver-T5 more consistently generated schema-compliant predictions. CONCLUSIONS: The fine-tuned lightweight language model demonstrates superior accuracy and efficiency in structuring liver cancer imaging reports compared to larger LLMs. This capability addresses critical challenges in clinical workflows by converting unstructured data into structured formats.
CCTA-derived trans-lesion fractional flow reserve gradient for predicting cardiovascular events after PCI: a prospective observational study.
Wang Y, Sun K, Gao Z … +2 more , Ji C, Wang X
BMC Med Imaging · 2026 Apr · PMID 42034988 · Full text
OBJECTIVE: This study aimed to evaluate the predictive value of pre-interventional coronary CT angiography-derived fractional flow reserve (FFR) in chronic coronary syndrome (CCS) patients after percutaneous coronary int...
OBJECTIVE: This study aimed to evaluate the predictive value of pre-interventional coronary CT angiography-derived fractional flow reserve (FFR) in chronic coronary syndrome (CCS) patients after percutaneous coronary intervention (PCI). Additionally, we investigated revascularization completeness using distal FFR (dFFR) and trans-lesion FFR gradient (ΔFFR). METHOD: This prospective observational study included 475 patients who underwent CCTA within 90 days before PCI, among whom 97 patients experienced major adverse cardiovascular events (MACEs). Patients were stratified based on dFFR ≤ 0.80 and ΔFFR ≥ 0.22. Completeness of revascularization was assessed by integrating dFFR and ΔFFR. MACE was defined as a composite of cardiovascular death, spontaneous myocardial infarction, and target vessel revascularization. Kaplan-Meier curves were used to describe the cumulative incidence of MACEs. RESULTS: Both dFFR (HRadjust: 0.006, 95%CI: 0.001-0.057; P < 0.001) and ΔFFR (HRadjust: 34.894, 95%CI: 5.323-228.722, P < 0.001) were independently associated with MACEs. The cumulative incidence of MACEs differed significantly between groups stratified by dFFR (45.3% vs. 20.8%; P = 0.014) and ΔFFR (52.1% vs. 27.4%; P < 0.001). Vessels with dFFR ≤ 0.80 and ΔFFR ≥ 0.22 exhibited a significantly higher risk of MACE than vessels with dFFR ≤ 0.80 and ΔFFR < 0.22 or dFFR > 0.80 (Log-rank: P < 0.001). The incomplete revascularization group, as assessed by both dFFR and ΔFFR, showed the highest risk of MACE (Log-rank: P < 0.001). CONCLUSION: ΔFFR effectively predicts MACEs in CCS patients after PCI. Integrating clinical features and hemodynamic characteristics enhances predictive accuracy. Additionally, ΔFFR aids in evaluating the completeness of revascularization.
A comparative assessment of age- and gender-related variations in brain morphometry using linear and volumetric MRI analysis.
Gupta S, Rastogi R
BMC Med Imaging · 2026 Apr · PMID 42032530 · Full text
OBJECTIVE: Morphometric evaluation of the human brain is essential for characterizing normal structural variability and establishing benchmarks for early detection of pathological changes. This study aims to assess gray...
OBJECTIVE: Morphometric evaluation of the human brain is essential for characterizing normal structural variability and establishing benchmarks for early detection of pathological changes. This study aims to assess gray matter, white matter, and ventricular morphology through both linear and volumetric MRI techniques and to identify structural differences associated with age and gender. MATERIALS AND METHODS: This retrospective study included 100 healthy adult volunteers with normal MRI findings (49 males, 51 females) aged 20–55 years. Linear measurements of the corpus callosum, ventricular system, and lobar gray and white matter were obtained using Radiant DICOM and OsiriX. Volumetric segmentation of brain tissues and lobes was performed with INMED NeuroShield software. Group differences were assessed using independent t-tests and one-way ANOVA, with false discovery rate correction (Benjamini–Hochberg) applied to control for multiple comparisons. RESULTS: Linear morphometric measures showed no significant gender differences, and age-related variations observed at the uncorrected level did not remain significant after multiple-comparison correction, whereas ventricular linear indices demonstrated significant gender differences that survived correction (p < 0.01). In contrast, volumetric analysis revealed higher global gray matter, white matter, and ventricular volumes in males, with age-associated gray matter reduction and ventricular enlargement remaining significant after false discovery rate correction (p < 0.01). CONCLUSION: The study establishes normative reference values for linear and volumetric morphometry, demonstrating age-dependent ventricular enlargement and reduction in GM and WM volumes, with measurable gender dimorphism. These findings support the clinical utility of MRI morphometry in distinguishing normal variation from early pathological change.
Predicting muscle-invasive bladder cancer with dual-layer detector spectral CT-derived extracellular volume fraction.
Lv J, Yang P, Zheng W … +10 more , Huang D, Feng Y, Huang B, Mu R, Dai S, Li P, Yang P, Li X, Zhu X, Qin X
BMC Med Imaging · 2026 Apr · PMID 42032498 · Full text
OBJECTIVES: To evaluate the diagnostic performance of dual-layer detector spectral CT (DLCT)-derived extracellular volume fraction (ECV) for pre-operatively distinguishing muscle-invasive bladder cancer (MIBC) from non-m...
OBJECTIVES: To evaluate the diagnostic performance of dual-layer detector spectral CT (DLCT)-derived extracellular volume fraction (ECV) for pre-operatively distinguishing muscle-invasive bladder cancer (MIBC) from non-muscle-invasive bladder cancer (NMIBC). MATERIALS AND METHODS: This retrospective study included 116 patients with pathologically confirmed urothelial carcinoma who underwent preoperative DLCT. Pathological results from transurethral resection or cystectomy served as the reference standard, classifying patients into MIBC (n = 57) and NMIBC (n = 59) groups. Two radiologists independently measured morphological features and spectral CT parameters, including iodine density and ECV. Statistical analyses involved univariate comparisons, multivariable logistic regression, and receiver operating characteristic (ROC) analysis. RESULTS: Multivariable analysis identified the longest tumor contact length (CL) ≥ 3 cm (Odds Ratio [OR] = 5.827; p = 0.006) and ECV (OR = 1.378 per 1% increment; p < 0.001) as independent predictors of MIBC. The area under the ROC curve (AUC) for ECV (0.886) was significantly superior to that of CL ≥ 3 cm (0.726) (DeLong test, p = 0.040). A combined model (ECV + CL) achieved an AUC of 0.869, which was not significantly better than ECV alone (p = 0.146). At the optimal cut-off of 72.3%, ECV predicted MIBC with 84.2% sensitivity and 88.1% specificity. Excellent inter-reader agreement was observed for all quantitative measurements (ICC ≥ 0.86). CONCLUSION: DLCT-derived ECV is a robust, non-invasive, and reproducible quantitative biomarker that outperforms conventional morphological assessment for the pre-operative prediction of MIBC, offering significant potential for improving clinical decision-making and personalized treatment planning.
Association between deep learning-based coronary artery calcium score on non-gated chest CT and progression of chronic kidney disease: a retrospective observational cohort study.
Yang K, Li M, Wang J … +5 more , Yu Y, Yu L, Dai X, Wu D, Zhang J
BMC Med Imaging · 2026 Apr · PMID 42032494 · Full text
BACKGROUND: Coronary artery calcification (CAC) is a pathological manifestation of coronary atherosclerosis in chronic kidney disease (CKD) patients. CAC on non-gated chest CT images can be precisely quantified through d...
BACKGROUND: Coronary artery calcification (CAC) is a pathological manifestation of coronary atherosclerosis in chronic kidney disease (CKD) patients. CAC on non-gated chest CT images can be precisely quantified through deep learning algorithms. Nevertheless, the relationship between deep learning-based coronary artery calcium score (DL-CACS) and the progression of CKD remains unclear. METHODS: Between January 2017 and June 2022, data from individuals with CKD were retrospectively collected. All enrolled participants had undergone non-gated chest CT scans and were stratified by DL-CACS at baseline: 0, 1-100, 101–400, and > 400 Agatston units (AU). The primary outcome of this study was a composite endpoint related to CKD progression, defined as either a ≥ 50% decrease in eGFR from baseline or the initiation of kidney replacement therapy during follow-up. The secondary outcome was major adverse cardiovascular events (MACEs), including cardiac death, non-fatal myocardial infarction, revascularization, rehospitalization resulting from heart failure or aggravated angina and all-cause mortality. RESULTS: Among the 509 patients with CKD (median age: 64.00 [57.00-70.50] years old; 317 men) finally included in this study, 155 (30.5%) patients achieved primary outcome during the follow-up period of 2152 person-years. Compared to individuals without CAC, higher DL-CACS was greatly associated with CKD progression. In the fully adjusted hazard models, the hazard ratio of DL-CACS of 1-100 was 2.27 (95% confidence interval [CI], 1.26–4.10), 3.75 (95% CI, 2.01-7.00) for DL-CACS of 101–400, and 4.52 (95% CI, 2.45–8.33) for DL-CACS > 400. The sensitivity analyses yielded similar results with primary findings. Of the 48 patients experienced the secondary outcome of MACEs, DL-CACS of 1-100, 101–400, and > 400 were associated with HRs of 1.65 (95% CI, 0.39–7.06), 5.46 (95% CI, 1.41–21.14), and 11.60 (95% CI, 3.09–43.58), respectively, in the final hazard models. CONCLUSIONS: Higher DL-CACS is associated with an increased risk of CKD progression. Associations with MACE were directionally consistent but imprecise, reflecting the limited events and wide confidence intervals.
Comparison of multistage and single-stage framework for automated landmark localization and radiologic measurement: a case of the C1-2 complex on cervical spine lateral radiographs.
Rhee W, Lee J, Chae I … +5 more , Kim BS, Park SC, Chang SY, Chang BS, Kim H
BMC Med Imaging · 2026 Apr · PMID 42032488 · Full text
BACKGROUND: Despite recent advances, applying deep learning to X-rays remains challenging due to the need for substantial downsampling and limited dataset sizes. While multistage frameworks have been proposed to promote...
BACKGROUND: Despite recent advances, applying deep learning to X-rays remains challenging due to the need for substantial downsampling and limited dataset sizes. While multistage frameworks have been proposed to promote stable performance, their advantages over single-stage designs remain unclear. This study aims to systematically compare the two designs for analyzing the C1-2 complex on cervical spine lateral radiographs. METHODS: A total of 1200 dynamic cervical spine X-ray and MRI pairs were collected from a tertiary care institution, and 300 studies from another institution were set aside for external validation. Two deep learning-based frameworks for localizing 8 landmarks and measuring 14 radiologic indices of the C1-2 complex were developed. The first of which is a multistage pipeline that implements coarse-to-fine localization, and the other is a single-stage model that processes full-region images. Five-fold cross-validation and an anatomy-aware multi-term loss were implemented, and performance was compared in terms of localization, segmentation, and measurement accuracy. RESULTS: The multistage framework demonstrated significantly smaller errors for the localization of the basion and C2 landmarks on the test sets (p < 0.001). For vertebral segmentation, the multistage framework achieved better performance (p < 0.001), yielding DSC and IoU of over 0.90 and 0.80, respectively. Radiologic measurements were more precise with the multistage design, exhibiting mean errors of 1–2° for angular measurements and 0.5–1.5 mm for distance measurements. CONCLUSIONS: We demonstrate that the multistage framework can consistently achieve superior performance over single-stage counterparts in analyzing cervical spine radiographs. It is expected to provide evidence for architectural design when developing deep learning models for the analysis of musculoskeletal imaging studies. CLINICAL TRIAL NUMBER: Not applicable.
Reduced right/left ventricular blood pool T2 ratio on cardiac magnetic resonance indicates cognitive impairment in heart failure secondary to ischemic heart disease.
Cui Y, Zheng C, Gu S … +6 more , Si J, Xiao K, Hu Y, Yang Y, Li J, Lu J
BMC Med Imaging · 2026 Apr · PMID 42021205 · Full text
PURPOSE: The right-to-left ventricular blood pool T2 ratio (RV/LV T2 ratio) derived from cardiac magnetic resonance (CMR) T2 mapping is a potential biomarker of blood oxygenation. This study investigated the association...
PURPOSE: The right-to-left ventricular blood pool T2 ratio (RV/LV T2 ratio) derived from cardiac magnetic resonance (CMR) T2 mapping is a potential biomarker of blood oxygenation. This study investigated the association between RV/LV T2 ratio and cognitive performance in heart failure (HF) secondary to ischemic heart disease (IHD). METHODS: This retrospective study included 52 patients with chronic HF and 26 healthy controls, all of whom underwent CMR and neuropsychological testing. Regions of interest were manually drawn in RV and LV blood pools to calculate RV/LV T2 ratio. Cognitive performance was assessed by the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA). Group comparisons and linear, multivariate, and mediation analyses were performed. RESULTS: HC demonstrated higher RV/LV T2 ratio than patients with HF with preserved or reduced and mildly reduced ejection fraction (p = 0.012, and p < 0.001). Patients with cognitive impairment (MoCA-defined) showed lower RV/LV T2 ratios than cognitively normal patients (p = 0.010). RV/LV T2 ratio was positively related to MMSE (β = 0.336, p = 0.020) and MoCA score (β = 0.563, p < 0.001). In multivariate linear regression analysis, RV/LV T2 ratio was the only CMR predictor for MoCA score (β = 0.340, p = 0.015). Mediation analysis showed RV/LV T2 ratio partially mediated the correlation between stroke volume and MoCA score, with a mediation effect ratio of 44.3%. CONCLUSION: RV/LV T2 ratio is an additional CMR biomarker to evaluate the cognitive performance in patients with HF secondary to IHD.
Deep learning prediction of contrast extravasation versus intracranial hemorrhage after thrombectomy in patients with acute stroke.
Ma D, Yang C, Jia Y … +5 more , Yan Q, Zhang F, Wei P, Xu Z, Zhang H
BMC Med Imaging · 2026 Apr · PMID 42021187 · Full text
BACKGROUND: It is difficult to distinguish the nature of contrast extravasation (CE) vs. intracranial hemorrhage (ICH) on immediate postprocedural computed tomography (CT) after endovascular thrombectomy. This differenti...
BACKGROUND: It is difficult to distinguish the nature of contrast extravasation (CE) vs. intracranial hemorrhage (ICH) on immediate postprocedural computed tomography (CT) after endovascular thrombectomy. This differentiation between CE and ICH carries important clinical implications, and this study aims to explore and validate the feasibility of approaches based on machine learning algorithm. METHODS: Patients with the hyperdensity on the immediate post endovascular treatment (EVT) CT scans were enrolled. Clinical and radiologic data of these patients were collected and nature of hyperdensity was identified based on the comparison of the immediate post-EVT CT scans vs. CT scans obtained 24 h post-EVT. After images with hyperdense lesion were labelled, two deep learning models based on convolutional neural network (CNN) were derived and validated. Model 1 was derived and internally evaluated with five folds cross-validation, and model 2 was fitted and evaluated on radiographs of whole patients who were randomly divided into the training set, validation set, and testing set. Moreover, the performance of model 2 on the testing set was compared with a support vector machine (SVM) model and a recursive partitioning and regression trees (RPART) model based on CT Hounsfield Units (HU) values. RESULTS: A total of 106 patients were enrolled, 63 patients were identified as CE, and 43 as ICH. Model 1 accomplished classification performance with the mean AUC of 0.955 ± 0.024 on the validation set. The performance of model 2 reached an AUC of 0.956 on testing set, which was higher than those of SVM model and RPART model. CONCLUSION: CNN-based deep learning algorithm demonstrated favorable classification performance in distinguishing between CE and ICH on post-EVT brain CT scans, and it provide a feasible exploratory tool for this clinical differentiation task.
Geometric feature analysis for identifying and characterizing normal variations on chest radiographs.
Fujimoto S, Hara T, Tanaka M … +5 more , Uesaka H, Itoh H, Sakai T, Tateishi T, Tsujikawa T
BMC Med Imaging · 2026 Apr · PMID 42021183 · Full text
PURPOSE: To develop and evaluate a quantitative method using geometric features of anatomical shadows on chest X-rays (CXRs) to identify normal variant cases and to elucidate the factors characterizing these variations....
PURPOSE: To develop and evaluate a quantitative method using geometric features of anatomical shadows on chest X-rays (CXRs) to identify normal variant cases and to elucidate the factors characterizing these variations. MATERIALS AND METHODS: This study included 548 normal CXRs confirmed by CT within one month. Shadows of six key anatomical structures (trachea, aorta, right atrium, left ventricle, right/left diaphragm domes) were manually labeled. Geometric features—center of gravity (x, y), length (l), and count (c)—were calculated for each label after image standardization. Cases with features (x, y, l) in the upper/lower 2.5% distribution or with label count (c) ≥ 2 were defined as normal variants. Associated factors were investigated using corresponding CT images and patient age/BMI. RESULTS: Normal variants were identified in 291/548 cases (53.1%). Analysis of 532 outlier feature instances showed associations with adjacent anatomical structures (41.0%, e.g., vertebrae, pulmonary veins, accessory fissures), spinal curvature/vertebral levels (14.2%, trachea only), and significant correlations with age and/or BMI (44.8%). Age was primarily linked to aortic variations, while BMI correlated more with cardiac and diaphragmatic variations. Notable findings included partial right atrial shadow disappearance near the middle lobe pulmonary vein and a wide normal superior tracheal range (C6-Th2). CONCLUSION: This quantitative geometric feature method successfully identifies and characterizes a wide range of normal variations on CXRs. It effectively linked feature outliers to specific anatomical structures and patient factors like age and BMI. This approach provides a foundation for automated large-scale analysis, potentially enhancing radiological training and diagnostic accuracy.
Enhancing mesenteric vascular imaging with dual-energy CTA: a comparison of DLIR and ASIR-V using low contrast agent and low radiation dose protocols.
Wu Y, Long J, Wang C … +8 more , Wang Z, Fan L, Liu X, Wang X, Song H, Sun A, Xu K, Meng Y
BMC Med Imaging · 2026 Apr · PMID 42021174 · Full text
BACKGROUND: Timely diagnosis of mesenteric vascular diseases, especially acute mesenteric ischemia (AMI) due to embolism in the superior mesenteric artery (SMA), is crucial for effective intervention. Dual-energy compute...
BACKGROUND: Timely diagnosis of mesenteric vascular diseases, especially acute mesenteric ischemia (AMI) due to embolism in the superior mesenteric artery (SMA), is crucial for effective intervention. Dual-energy computed tomography angiography (DE-CTA) is a key diagnostic tool; however, concerns about contrast-induced nephropathy and radiation exposure persist. OBJECTIVES: This study assesses the diagnostic performance of DE-CTA for detecting mesenteric vascular diseases and compares the effectiveness of image reconstruction algorithms, specifically ASIR-V and DLIR, in enhancing image quality while minimizing the risks associated with contrast agents and radiation exposure. METHODS: DE-CTA with virtual monoenergetic imaging (VMI) at 40 keV was performed on 50 patients. Image quality was evaluated using contrast-to-noise ratio (CNR), signal-to-noise ratio (SNR), and standard deviation (SD). The performance of DLIR and ASIR-V algorithms was compared, with an emphasis on minimizing radiation dose and contrast agent use through optimized low-dose protocols. RESULTS: DLIR-H significantly outperformed both ASIR-V 50% and DLIR-M in terms of CNR and SNR, with CNR improving by 25% and SNR by 30% compared to ASIR-V 50%. DLIR-H also demonstrated superior noise reduction, with a 35% reduction in SD compared to ASIR-V. Furthermore, by effectively suppressing the elevated image noise inherent to low-dose scanning protocols, DLIR maintained excellent diagnostic image quality. This highlights its potential for dose reduction in clinical practice, which is crucial for minimizing the risks of radiation exposure and contrast-induced nephropathy. CONCLUSION: For mesenteric DE-CTA imaging at 40 keV, DLIR significantly improves objective image quality and subjective reader confidence compared to ASIR-V. By preserving fine vascular details at higher noise levels, DLIR demonstrates the potential to facilitate low-radiation and low-contrast-dose protocols in clinical practice.
Explainable machine learning-based mortality prediction in critically Ill patients with rheumatoid arthritis-associated lung disease: a radiomics and clinical data integration study.
He Y, Liang R, Zhang Y … +1 more , Li X
BMC Med Imaging · 2026 Apr · PMID 42015074 · Full text
BACKGROUND: Intensive Care Unit (ICU) mortality risk is high in rheumatoid arthritis (RA) patients, yet effective prognostic tools remain scarce. OBJECTIVES: To develop and evaluate a machine learning (ML)-based prognost...
BACKGROUND: Intensive Care Unit (ICU) mortality risk is high in rheumatoid arthritis (RA) patients, yet effective prognostic tools remain scarce. OBJECTIVES: To develop and evaluate a machine learning (ML)-based prognostic model for predicting hospital mortality in severe RA patients. METHODS: This retrospective cohort study utilized data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) and the Medical Information Mart for Intensive Care Chest X-ray (MIMIC-CXR) databases, including 1,951 chest X-rays from 984 patients with RA. The primary outcome was all-cause in-hospital mortality. Radiomics features were extracted using PyRadiomics, with 74 features retained after quality control. Key features were selected using the Boruta algorithm and integrated with clinical variables to develop three modeling strategies: clinical-only, radiomics-only, and combined models. Nine ML algorithms were applied using a 60/40 training-test split with 10-fold cross-validation. To address class imbalance (mortality rate: 7.7%), the Synthetic Minority Over-sampling Technique (SMOTE) was applied. Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI) evaluated the predictive incremental value of imaging omics features combined with clinical data compared to single-modality data. SHapley Additive exPlanations (SHAP) was used to interpret model predictions. RESULTS: A total of 984 RA patients were included, of whom 76 (7.7%) experienced in-hospital mortality. The neural network model demonstrated superior performance, with an area under the curve (AUC) of 0.887 (95% CI: 0.752–0.934) in the training set and 0.800 (95% CI: 0.714–0.856) in the test set. The combined clinical-radiomics model showed significant incremental value compared to the clinical-only model (NRI: 0.3773, 95% CI: 0.3499–0.4047, P < 0.001; IDI: 0.2303, 95% CI: 0.2197–0.2409, P < 0.001) and the radiomics-only model (NRI: 0.2249, 95% CI: 0.1510–0.2988, P < 0.001; IDI: 0.1939, 95% CI: 0.1655–0.2222, P < 0.001). Key predictive features included blood urea nitrogen (BUN), Wavelet-LL 10th Percentile, and Wavelet-Haar HH Entropy. CONCLUSION: The integrated ML model combining chest radiomics and clinical data effectively predicts mortality risk in critically ill RA patients, offering good generalizability and interpretability. It provides a practical, interpretable framework for clinical risk stratification and lays the foundation for the development of intelligent prognostic systems for patients with severe RA and resource allocation.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe