BMC Medical Imaging[JOURNAL]
Sun
200 papers
End-to-end 2.5D multisequence-multichannel fusion model for preoperative survival prediction in glioma: a retrospective study.
Lv P, Ren Y, Chang J … +4 more , Wang M, Liu Y, Miao Y, He X
BMC Med Imaging · 2026 Jul · PMID 42401824 · Full text
Gliomas are the most common primary malignant tumors of the central nervous system and show marked imaging heterogeneity, making accurate preoperative prediction of 24-month overall survival status important for individu...
Gliomas are the most common primary malignant tumors of the central nervous system and show marked imaging heterogeneity, making accurate preoperative prediction of 24-month overall survival status important for individualized treatment planning and prognostic counseling. Summary receiver operating characteristic analyses were used to systematically compare combinations of convolutional backbone depth and input channel counts to identify an optimal sequence-channel configuration. Guided by these findings, we developed an end-to-end multisequence-multichannel fusion aggregation (EMFA) framework that integrates deep transfer learning image representations with radiomics features. A multi-instance learning (MIL) module was further incorporated to enable adaptive within-sequence weighting and cross-sequence feature aggregation, improving model interpretability. In the held-out test cohort, the EMFA framework achieved an area under the receiver operating characteristic curve (AUC) of 0.899 and an accuracy of 0.935 (93.5%), showing competitive performance relative to corresponding single-sequence baselines. The configuration analysis indicated that nine-channel inputs for T1-weighted and T2-weighted imaging and a three-channel input for contrast-enhanced T1-weighted imaging provided the best overall performance. These results suggest that EMFA offers an effective and scalable strategy for fusing multisequence magnetic resonance imaging (MRI) information for 24-month glioma survival-status prediction, supporting imaging-informatics-driven clinical decision support and potential future translation into neuroradiology workflows.
Association between 3 D CT-based volumetric fat infiltration of lumbar paraspinal muscles and bone mineral density: a clinical study.
Feng R, Pan B, Ye L … +4 more , Yin B, Pan G, Chen J, Wang D
BMC Med Imaging · 2026 Jul · PMID 42401813 · Full text
OBJECTIVE: This study investigated the association between the fat infiltration rate (FIR) of lumbar paraspinal muscles-specifically the multifidus, erector spinae, and psoas major-and bone mineral density (BMD) in postm...
OBJECTIVE: This study investigated the association between the fat infiltration rate (FIR) of lumbar paraspinal muscles-specifically the multifidus, erector spinae, and psoas major-and bone mineral density (BMD) in postmenopausal women. By examining muscular and segmental (L2-L4) adiposity variations, we aimed to identify imaging biomarkers for early osteoporosis detection and to inform integrated "bone-muscle" therapeutic strategies. METHODS: A retrospective analysis was performed on 120 postmenopausal women who underwent dual-energy X-ray absorptiometry (DXA) and lumbar computed tomography (CT) between 2024 and 2025. Participants were categorized into three cohorts-normal bone mass, osteopenia, and osteoporosis-based on DXA T-scores. Demographic data, including BMI, were recorded and subsequently utilized as a covariate in the regression models to ensure robust statistical adjustment. The L2-L4 paraspinal musculature was automatically segmented in 3D using 3D Slicer software with the MuscleMap plug-in to determine the FIR for each muscle group. Data normality was assessed using Shapiro-Wilk tests, while inter-group differences were compared via one-way ANOVA or Kruskal-Wallis H tests. Correlations were quantified using Pearson or Spearman analyses. RESULTS: Age was comparable across groups ( P = 0. 160 ), but BMI and average BMD differed significantly. Multifidus FIR showed a significant inverse correlation with BMD ( P < 0. 001 ). After adjusting for BMI, multifidus fat infiltration remained an independent predictor of BMD. At the segmental level, the L3 multifidus displayed the greatest inter-group variance and the strongest negative correlation with BMD ( r = -0. 689, P < 0. 001 ). CONCLUSIONS: Lumbar paraspinal muscle adiposity, particularly in the L3 multifidus, is inversely associated with BMD in postmenopausal women. Consequently, L3 multifidus FIR is a statistically significant auxiliary imaging metric for proactive osteoporosis screening and the assessment of osteosarcopenia risk.
A study to measure the utility of an AI-enhanced reporting tool in assisting busy CCTA readers with REPORT generation (SMART-REPORT).
O'Neal WT, Gudi H, Chandrasekhar M … +5 more , Schumann C, Birch A, Mullen S, Ng N, Morris M
BMC Med Imaging · 2026 Jul · PMID 42393584 · Full text
PURPOSE: This study aimed to determine if an integrated structured reporting (SR) tool incorporating results from artificial intelligence (AI) enabled coronary stenosis quantification (AI-CSQ), AI-enabled coronary plaque...
PURPOSE: This study aimed to determine if an integrated structured reporting (SR) tool incorporating results from artificial intelligence (AI) enabled coronary stenosis quantification (AI-CSQ), AI-enabled coronary plaque analysis (AI-CPA), and fractional flow reserve CT (FFR) can increase reader efficiency for coronary computed tomography angiography (CCTA), while reducing cognitive load. METHODS: Eighty CCTAs from patients with stable chest pain or symptoms suggestive of coronary artery disease (CAD) were analyzed by five independent readers from two clinical sites. CCTA datasets were used to generate combined reports for AI-CSQ, AI-CPA, and FFR. Each reader randomly interpreted 40 cases with the SR tool (which provided a prepopulated fully editable template) and 40 cases without the SR tool using site-specific clinical workflows (400 total cases). Reporting times from initiation of case interpretation through report signing were measured with and without the SR tool and stratified by CAD-Reporting and Data System (CAD-RADS) 2.0 category. Reader confidence was recorded using a 5-point Likert Scale, and cognitive load (e.g., working memory load) was determined using a 9-point scale, with comparison by SR tool use. Inter-reader agreement associated with CCTA interpretation with versus without the SR tool also was compared. RESULTS: Reporting time per case was 40.2% shorter with versus without the SR tool (6.0 ± 2.0 vs. 10.0 ± 3.3 min; p < 0.001), independent of CAD-RADS 2.0 category. Readers were more likely to report very confident reads with the SR tool than without (48% vs. 29%; p < 0.001). Cognitive load also was significantly lower with the SR tool than without (3.8 ± 1.2 vs. 6.1 ± 1.5; p < 0.001). Report agreement was significantly higher with versus without the SR tool (94.6%±2.3% vs. 45.3%±9.2%; p < 0.001). CONCLUSION: The SR tool significantly improved reader efficiency and decreased work effort when interpreting comprehensive CCTA exams.
Age-specific MRI patterns in pediatric epilepsy: insights from a sudanese cohort and implications for low-resources settings.
Mohamed Y, Babiker A, Elamin H … +1 more , Abdalla E
BMC Med Imaging · 2026 Jul · PMID 42393563 · Full text
OBJECTIVE: Epilepsy is a global health problem affecting the quality of life of many children. Neuroimaging, particularly Magnetic Resonance Imaging (MRI), plays a crucial role in precisely identifying epileptogenic foci...
OBJECTIVE: Epilepsy is a global health problem affecting the quality of life of many children. Neuroimaging, particularly Magnetic Resonance Imaging (MRI), plays a crucial role in precisely identifying epileptogenic foci that are potentially amenable to surgical resection for a possible cure. This study aims to evaluate the diagnostic yield of MRI in a Sudanese pediatric epilepsy cohort and to identify age-specific neuroimaging patterns to optimize resource allocation in low-income settings. METHODS: A descriptive cross-sectional study enrolled 100 pediatric patients (≤ 17 years) diagnosed with epilepsy from June 2023 to June 2024. Data on age, gender, and MRI findings were collected and analyzed using SPSS v26.0. Chi-square tests assessed univariate associations, and multivariate logistic regression evaluated the combined impact of age and gender on abnormal MRI findings, with significance set at P < 0.05. RESULTS: Of 100 patients (55% male, 45% female), 87% were aged 1-12 years. MRI abnormalities were detected in 31%, with mesial temporal sclerosis (35.4%) most frequent, followed by brain atrophy and periventricular leukomalacia (16% each). Age was significantly associated with abnormalities (P = 0.002), with higher odds in < 1-year-olds (OR = 9.35, 95% CI: 1.67-52.41, P = 0.011) and 13-17-year-olds (OR = 14.20, 95% CI: 1.47-137.32, P = 0.021) compared to 1-6-year-olds. Specific findings varied by age: periventricular leukomalacia in < 1-year-olds, atrophy in 1-6 years, mesial temporal sclerosis in 7-12 years, and tumors in 13-17 years (P < 0.002). Gender showed no significant effect (OR = 0.74, 95% CI: 0.32-1.71, P = 0.480). CONCLUSION: Approximately one-third of children with epilepsy showed MRI abnormalities, most commonly mesial temporal sclerosis. Abnormal MRI results were notably linked to patient age, particularly those under one year old and between 13 and 17 years old. Age-specific MRI protocols could enhance diagnostic efficiency in low-resources settings. These findings support for prioritized neuroimaging in infants and adolescents, informing public health strategies to address Sudan's epilepsy burden. CLINICAL TRIAL NUMBER: Not applicable.
Qualitative and quantitative assessment of intratumoral fat using chemical-shift MRI for predicting histological grade of hepatocellular carcinoma.
Dai X, Wang Q, Hao Y … +7 more , Ye J, Feng L, Xu J, Su J, Ji D, Liu C, Lang N
BMC Med Imaging · 2026 Jul · PMID 42393558 · Full text
OBJECTIVES: To evaluate intratumoral fat in hepatocellular carcinoma (HCC) using both qualitative and quantitative approaches based on routine chemical-shift magnetic resonance imaging (MRI), and to investigate its poten...
OBJECTIVES: To evaluate intratumoral fat in hepatocellular carcinoma (HCC) using both qualitative and quantitative approaches based on routine chemical-shift magnetic resonance imaging (MRI), and to investigate its potential value in predicting histological grade. METHODS: This retrospective study included 282 patients with pathologically confirmed HCC between January 2015 and November 2025. Tumors were classified into low-grade and high-grade groups according to the Edmondson-Steiner grade. Intratumoral fat was assessed on in-phase and opposed-phase MRI images. For qualitative assessment, intratumoral fat pattern was categorized as none, heterogeneous, or homogeneous. For quantitative assessment, regions of interest were manually delineated on three consecutive slices showing the largest tumor area, and the mean fat fraction (FF) was calculated. Logistic regression analysis was performed to identify risk factors associated with high-grade HCC. Furthermore, models incorporating clinicoradiological factors were developed for the preoperative prediction of HCC histological grade. RESULTS: Homogeneous intratumoral fat was more frequently observed in low-grade tumors than in high-grade tumors, and FF was significantly higher in low-grade tumors than in high-grade tumors. Both homogeneous intratumoral fat (odds ratio [OR] = 0.230 [0.097-0.514], P = 0.001) and FF (OR = 0.861 [0.811-0.907], P < 0.001) were identified as independent predictors of high-grade HCC. When combined with other clinicoradiological factors, the FF-based model showed better performance than the intratumoral fat pattern-based model (area under the receiver operating characteristic curve: 0.792 vs. 0.744, P = 0.024). CONCLUSION: Intratumoral fat assessed using chemical-shift imaging provides a simple and noninvasive imaging biomarker for predicting the histological grade of HCC. Both homogeneous intratumoral fat and higher FF were associated with a lower risk of high-grade HCC, and the model based on quantitative assessment outperformed that based on qualitative evaluation.
Gd-EOB-DTPA-enhanced MRI in the diagnosis of intrahepatic cholestasis in mice: an experimental study.
Wang S, Lu R, Wu M … +6 more , Li Q, Zhang Y, Liu L, Yang Y, Wu K, Wang S
BMC Med Imaging · 2026 Jul · PMID 42393538 · Full text
BACKGROUND: The diagnosis of intrahepatic cholestasis remains challenging due to a lack of reliable noninvasive biomarkers. This study aimed to define the diagnostic utility of semi-quantitative parameters from Gd-EOB-DT...
BACKGROUND: The diagnosis of intrahepatic cholestasis remains challenging due to a lack of reliable noninvasive biomarkers. This study aimed to define the diagnostic utility of semi-quantitative parameters from Gd-EOB-DTPA-enhanced MRI for assessing pathological severity in a murine model of 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC)-induced intrahepatic cholestasis. METHODS: Thirty female Kunming mice were divided into control (n = 5) and DDC-fed (n = 20) groups. Multi-timepoint (0-60 min post-injection) Gd-EOB-DTPA-enhanced MRI at 3.0T was performed to derive kinetic parameters-maximum relative enhancement (RE), time-to-peak (TTP), and maximum slope (Slope)-from liver parenchyma and gallbladder. Histopathological scoring of key features (bile duct proliferation, cholestasis, hepatocyte degeneration, periductular fibrosis, porphyrin emboli) and serum biochemistry served as reference standards. Correlations and diagnostic performance were analyzed statistically. RESULTS: DDC feeding induced progressive cholestatic injury, evident on histology from week 2 onward. MRI revealed distinct kinetic impairments: attenuated hepatic RE (strong inverse correlations with bile duct proliferation [rs = -0.661, P = 0.0003] and cholestasis [rs = -0.546, P = 0.005]) and suppressed gallbladder positive Slope (inversely correlated with cholestasis severity [rs = -0.663, P = 0.0003]). Gallbladder positive Slope demonstrated exceptional diagnostic accuracy (AUC = 0.950), outperforming hepatic parameters. Serum biomarkers (AST, ALT) were elevated but did not correlate with MRI kinetics. CONCLUSIONS: Gd-EOB-DTPA-enhanced MRI quantitatively captures functional impairment in mice intrahepatic cholestasis. Gallbladder excretion kinetics (Slope) and hepatic uptake (RE) serve as sensitive, non-invasive biomarkers strongly correlated with histopathological severity, offering a promising approach for functional assessment that complements conventional serology and morphology.
SWI combined with cMRI and CT in the differentiating of intracranial Rosai-Dorfman disease from fibrous meningioma.
Lin X, Wang X, Wang Y … +5 more , Su W, Yu F, Wang F, Cao D, Xing Z
BMC Med Imaging · 2026 Jul · PMID 42387449 · Full text
OBJECTIVE: Intracranial Rosai-Dorfman disease (RDD) and fibrous meningioma, despite exhibiting overlapping features on conventional MRI (cMRI) and CT, are frequently confounded in preoperative diagnosis. This clinical di...
OBJECTIVE: Intracranial Rosai-Dorfman disease (RDD) and fibrous meningioma, despite exhibiting overlapping features on conventional MRI (cMRI) and CT, are frequently confounded in preoperative diagnosis. This clinical dilemma is critical, as the two entities demand substantially different treatment strategies and are associated with divergent prognoses. The aim of this study was to identify key discriminative imaging features on cMRI, susceptibility-weighted imaging (SWI), and CT to enhance preoperative differentiation between intracranial RDD and fibrous meningioma. METHODS: This retrospective study included 7 patients with pathologically confirmed intracranial RDD (12 lesions) and 36 patients with fibrous meningioma (36 lesions). All patients underwent preoperative imaging including cMRI, SWI, and CT. Two neuroradiologists independently evaluated imaging features, including signal characteristics, enhancement, perilesional edema, SWI hypointensity, phase image signal, presence of calcification, and the morphological ratio. Interobserver agreement was evaluated using Cohen's kappa coefficient, and intergroup differences were analyzed using the χ2 test and t-test. RESULTS: Compared to fibrous meningiomas, RDD lesions demonstrated several distinctive imaging characteristics. A statistically significant difference in sex distribution was observed between the two groups (p = 0.035). In terms of imaging features, RDD lesions consistently showed: (1) different T1 and T2 signal; (2) perilesional edema (100% vs. 25%, p<0.001); (3) diffuse SWI hypointensity throughout the lesion (100% vs. 0%, p<0.001); (4) high phase image signal in the majority of RDD lesions (75% vs. 19.4%, p<0.001); (5) absence of calcification (0% vs. 52.8%, p<0.001); (6) a significantly higher and more variable longest-to-shortest diameter ratio (2.81±2.76 vs. 1.20±0.20, p<0.001). No significant differences were observed in enhancement homogeneity or meningeal enhancement degree. CONCLUSION: The integration of cMRI, SWI, and CT features may aid in differentiating intracranial RDD from fibrous meningioma.
Fractional anisotropy, perfusion, and metabolic correlates of peritumoural brain oedema in meningiomas: a cross-sectional observational multiparametric MRI study.
Varshney E, Chauhan U, Arora RK … +2 more , Bahurupi Y, Dev R
BMC Med Imaging · 2026 Jul · PMID 42387434 · Full text
OBJECTIVE: To evaluate perfusion, diffusion tensor, and spectroscopy correlates of peritumoural brain oedema (PTBO) in intracranial meningiomas and identify imaging parameters associated with PTBO volume using multiparam...
OBJECTIVE: To evaluate perfusion, diffusion tensor, and spectroscopy correlates of peritumoural brain oedema (PTBO) in intracranial meningiomas and identify imaging parameters associated with PTBO volume using multiparametric Magnetic Resonance Imaging (MRI). METHODS: In this cross-sectional study of 66 adults, PTBO presence was defined as a T2-hyperintense peritumoural volume ≥ 10 mL (n = 26) and absence as ≤ 1 mL (n = 35); intermediate volumes were excluded (n = 5). Preoperative 1.5T MRI assessed relative cerebral blood volume (rCBV), apparent diffusion coefficient (ADC), fractional anisotropy (FA), and magnetic resonance spectroscopy (MRS). Independent predictors were identified via multivariable logistic regression. Diagnostic performance (Area Under the Curve, AUC) was estimated using 5-fold cross-validation. RESULTS: PTBO-associated tumours demonstrated significantly larger volumes, more frequent intratumoural necrosis, higher rCBV, elevated ADC, and increased choline ratios (p < 0.001 for all). Multivariable regression identified FA, rCBV, and tumour volume as independent predictors of PTBO presence. FA demonstrated superior discrimination (cross-validated AUC = 0.96; 95% CI 0.92-0.99) compared to rCBV (cross-validated AUC = 0.82) and tumour volume (cross-validated AUC = 0.78). CONCLUSION: Multiparametric MRI correlates significantly with PTBO. FA shows the strongest independent association with the presence of PTBO. Elevated rCBV/ADC, reduced FA, and higher choline ratios characterise oedema-forming meningiomas. All proposed FA thresholds (including values around 0.45) are exploratory, cohort-derived, and require external validation before clinical use.
Identification of ischemic stroke risk in patients with left ventricular non-compaction using echocardiography, deep learning, and radiomics.
Wang J, Tao W, Wang Z … +1 more , Liu B
BMC Med Imaging · 2026 Jul · PMID 42387411 · Full text
BACKGROUND: There is a certain correlation between left ventricular non-compaction (LVNC) and ischemic stroke (IS); however, there are currently no predictive models available to assess the risk of IS in LVNC patients. M...
BACKGROUND: There is a certain correlation between left ventricular non-compaction (LVNC) and ischemic stroke (IS); however, there are currently no predictive models available to assess the risk of IS in LVNC patients. METHODS: A multicenter retrospective study included 309 LVNC patients from two institutions. Institution 1 (228 patients) provided a training and internal validation set, while institution 2 (81 patients) served as the external validation set. The deep transfer learning features were extracted using a ResNet101-based model, and the radiomics features were extracted using Pyradiomics. Feature selection was done with LASSO, the selected features were input into XGBoost to construct the Res101-RAD-XGB model. Model performance was evaluated using receiver operating characteristic (ROC) curves and decision curve analysis (DCA), and the model's interpretability was assessed using SHAP analysis and the Grad-CAM method. Finally, the performance of the Res101-RAD-XGB model was compared with that of clinical criteria to assess its effectiveness. RESULTS: The Res101-RAD-XGB model achieved area under the curve (AUCs) of 0.942, 0.913, and 0.921 for the training, internal validation, and external validation sets, outperforming models using only one type of feature and clinical criteria. CONCLUSION: The Res101-RAD-XGB model can accurately identify LVNC patients who are at risk of IS, enabling specialists to develop targeted preventive strategies for high-risk individuals.
Quantitative spectral parameters of photon-counting detector CT for noninvasive prediction of PD-L1 expression in non-small cell lung cancer.
Ren J, Fu Z, Li L … +2 more , Huang Y, Yin Y
BMC Med Imaging · 2026 Jul · PMID 42387400 · Full text
PURPOSE: This study aimed to investigate the value of quantitative photon-counting computed tomography (PCCT) spectral parameters for the non-invasive assessment and prediction of programmed death-ligand 1 (PD-L1) expres...
PURPOSE: This study aimed to investigate the value of quantitative photon-counting computed tomography (PCCT) spectral parameters for the non-invasive assessment and prediction of programmed death-ligand 1 (PD-L1) expression in non-small cell lung cancer (NSCLC). METHODS: This single-center retrospective study enrolled 63 NSCLC patients who underwent pretherapeutic PCCT. Pretherapeutic PD-L1 expression was evaluated using immunohistochemistry and classified into three groups: negative (TPS < 1%), intermediate (1%-49%), and high (≥ 50%). The arterial and venous-phase CT values (A/V-40 keV, A/V-70 keV, and A/V-100 keV) and iodine concentration of the lesions were measured on 40, 70, and 100 keV monoenergetic images. Normalized iodine concentration and the slope of the spectral curve (λHU) were calculated accordingly. Spearman and Pearson correlation, receiver operating characteristic (ROC), univariate and multivariate logistic regression analyses were conducted to assess the correlations, predictive performance, and the independent predictive value of PCCT parameters for PD-L1 expression. RESULTS: Age and gender demonstrated statistically differences among the groups (P = 0.036; P = 0.008). A-40 keV and A-λHU also exhibited significant differences. Both parameters were negatively correlated with PD-L1 expression (A-λHU: rs = - 0.270, P = 0.032; A-40 keV: rs = - 0.279, P = 0.027). Multivariate logistic regression confirmed that age and A-λHU as independent predictors of intermediate and high PD-L1 expression, respectively (OR = 1.014 and 0.031; both P < 0.05). ROC analysis revealed that the clinical model, PCCT model, and combined model demonstrated favorable predictive performance for high PD-L1 expression, with area under the curve values of 0.762 (95%CI: 0.637-0.860), 0.846 (95% CI: 0.733-0.925), and 0.889 (95% CI: 0.785-0.954), respectively (all P < 0.001). CONCLUSIONS: Age and A-λHU are valuable predictors of PD-L1 expression in NSCLC patients before treatment. The model, combining PCCT spectral parameters and clinical features, may serve as a non-invasive tool for individualized treatment in NSCLC.
Radiomics-based causal machine learning for exploratory treatment-effect estimation of neoadjuvant chemotherapy cycle intensity in osteosarcoma: a proof-of-concept study.
Yuan Z, Wang R, Zhang H … +9 more , Zheng Z, Zhou X, Xing J, Zhai W, Cao X, Zhao C, Zhou T, Fan L, Yang C
BMC Med Imaging · 2026 Jun · PMID 42380890 · Full text
BACKGROUND: Radiomics-based modeling has shown promise for characterizing tumor heterogeneity, but its integration with causal machine learning for treatment-effect estimation remains underexplored in osteosarcoma. OBJEC...
BACKGROUND: Radiomics-based modeling has shown promise for characterizing tumor heterogeneity, but its integration with causal machine learning for treatment-effect estimation remains underexplored in osteosarcoma. OBJECTIVE: This study aimed to develop a proof-of-concept radiomics-based causal machine learning framework for exploratory estimation of average and individual treatment effects associated with neoadjuvant chemotherapy cycle intensity in osteosarcoma. METHODS: This retrospective single-center study included 34 patients with osteosarcoma who underwent neoadjuvant chemotherapy followed by surgical resection. Radiomic features were extracted from pre-treatment T1-weighted magnetic resonance imaging and combined with baseline clinical variables. Three causal meta-learners-S-Learner, T-Learner, and X-Learner-were implemented to estimate counterfactual survival probabilities under high-cycle and low-cycle neoadjuvant chemotherapy strategies. Average treatment effects and individual treatment effects were derived from the predicted potential outcomes. RESULTS: The proposed framework enabled estimation of population-level and individualized treatment-effect measures using integrated radiomic and clinical covariates. The estimated average treatment effects differed in magnitude and direction across meta-learners, indicating instability of treatment-effect estimation in this small cohort. Confidence intervals crossed zero for two of the three learners, and model performance metrics were interpreted only as technical indicators of feasibility rather than evidence of generalizable predictive validity. CONCLUSION: This study demonstrates the methodological feasibility of combining radiomics with causal machine learning for exploratory treatment-effect estimation in osteosarcoma. Given the limited sample size, retrospective single-center design, treatment-group imbalance, missing outcome information, and uncertainty of causal assumptions, the findings should be regarded as hypothesis-generating rather than clinically actionable. Larger multicenter studies with standardized imaging protocols, adequate event counts, longer follow-up, and prospective validation are required before the translational relevance of radiomics-based causal treatment-effect estimation can be assessed.
Gestational age-specific MRI reference values for fetal renal morphology and ADC.
Pu Y, Ma S, Wang Q … +4 more , Li R, Zou X, Liu Y, Kang M
BMC Med Imaging · 2026 Jun · PMID 42374326 · Full text
BACKGROUND: Fetal renal developmental abnormalities are among the most common congenital malformations and are associated with adverse perinatal outcomes. Although fetal magnetic resonance imaging (MRI) can provide compl...
BACKGROUND: Fetal renal developmental abnormalities are among the most common congenital malformations and are associated with adverse perinatal outcomes. Although fetal magnetic resonance imaging (MRI) can provide complementary morphological and quantitative information, gestational age (GA)-specific reference values integrating fetal renal morphology and diffusion parameters remain limited. This study aimed to establish MRI-based reference values for normal fetal renal morphology and apparent diffusion coefficient (ADC) across gestation. METHODS: In this retrospective single-center observational study, 313 singleton pregnancies (GA 23-38 weeks) with normal postnatal urinary outcomes underwent 1.5T fetal MRI, including T2-weighted and diffusion-weighted imaging sequences. Morphological parameters [anteroposterior, transverse and longitudinal diameters, and mean renal parenchymal thickness (MRPT) and functional parameters T2 relative signal intensity (RSI-T2) and apparent diffusion coefficient (ADC)] were measured for both kidneys. Reproducibility was assessed using intraclass correlation coefficients (ICC). Associations between MRI parameters and GA were evaluated using correlation and linear regression analyses. RESULTS: All measured parameters showed good to excellent inter-observer reproducibility. Renal diameters and MRPT increased significantly with GA (r = 0.72-0.89; all P < 0.0001). In contrast, renal ADC values showed moderate negative correlations with GA (right kidney: r =-0.47; left kidney: r =-0.52; both P < 0.0001). The regression equations for ADC (×103 mm²/s) were ADC_right = 3.23 - 0.049 × GA and ADC_left = 3.46 - 0.058 × GA. CONCLUSIONS: This study provides GA-stratified reference ranges for key morphological and functional MRI parameters of normal fetal kidneys in the second and third trimesters. These quantitative benchmarks may provide a reference framework for fetal renal MRI assessment and for future studies of fetal renal abnormalities. TRIAL REGISTRATION: Not applicable.
MRI findings of intrahepatic cholangiocarcinoma with sarcomatoid differentiation: a retrospective case series.
Wu H, Hu C, Zhang H … +2 more , Zhao X, Zhu Z
BMC Med Imaging · 2026 Jun · PMID 42374279 · Full text
PURPOSE: To describe MRI findings of intrahepatic cholangiocarcinoma with sarcomatoid differentiation (s-iCCA), summarize associated clinical features, and explore imaging characteristics that may help distinguish it fro...
PURPOSE: To describe MRI findings of intrahepatic cholangiocarcinoma with sarcomatoid differentiation (s-iCCA), summarize associated clinical features, and explore imaging characteristics that may help distinguish it from conventional intrahepatic cholangiocarcinoma. MATERIALS AND METHODS: This retrospective single-center study reviewed surgically resected, pathology-proven s-iCCA between June 2018 and September 2025. Patients who underwent preoperative liver magnetic resonance imaging (MRI) with dynamic contrast enhancement and diffusion-weighted imaging (DWI) were included; tumors arising from the extrahepatic bile duct or gallbladder and those treated before MRI were excluded. Two radiologists evaluated lesion morphology, signal characteristics, enhancement pattern, and tumor spread independently. Clinical, laboratory, pathologic, treatment, and follow-up data were collected. RESULTS: Five patients (2 men, 3 women; mean age, 64 years) were included. Serum carbohydrate antigen 19-9 (CA 19-9) was elevated in all patients, whereas alpha-fetoprotein (AFP) and carcinoembryonic antigen (CEA) were within reference range in tested patients. All tumors presented as a dominant mass-forming lesion (3.4-9.4 cm; segments V-VII (right), n = 4; segments III/IV (left), n = 1). On MRI, lesions were hypointense on T1-weighted imaging and heterogeneously hyperintense on T2-weighted imaging; intratumoral hemorrhage suggested by T1 hyperintensity was observed in two patients, and extensive nonenhancing necrosis/cystic degeneration in three. All tumors showed marked diffusion restriction on DWI. Dynamic imaging demonstrated arterial peripheral rim enhancement with progressive and persistent heterogeneous enhancement without washout in all cases. Segmental bile duct dilatation was present in two patients, suspected portal vein tumor thrombus in one, and regional lymphadenopathy in three. Extensive necrosis or cystic degeneration was common and contributed to marked imaging heterogeneity. CONCLUSION: In this small series, s-iCCA frequently appeared as a necrotic and heterogeneous mass-forming lesion with diffusion restriction and progressive enhancement. Although these findings substantially overlap with those of conventional iCCA, prominent necrosis and marked heterogeneity may raise suspicion for sarcomatoid differentiation in the appropriate clinical context.
Multimodal deep learning for papillary thyroid carcinoma diagnosis using ultrasound and cytology.
Kalejahi BK, Rajabi MJ
BMC Med Imaging · 2026 Jun · PMID 42374268 · Full text
BACKGROUND AND OBJECTIVE: Papillary thyroid carcinoma (PTC) is the most common thyroid malignancy, and pre-operative diagnosis depends on integrating ultrasound (US) imaging with fine-needle aspiration cytology (FNAC). P...
BACKGROUND AND OBJECTIVE: Papillary thyroid carcinoma (PTC) is the most common thyroid malignancy, and pre-operative diagnosis depends on integrating ultrasound (US) imaging with fine-needle aspiration cytology (FNAC). Pang et al. (2025) recently released a paired US/cytology dataset and demonstrated that classical radiomics combined with classifiers such as support vector machines, random forests, and XGBoost can reach AUROC ≈ 0.99 on a single random split. We re-examine this result under a stricter evaluation protocol and contribute a calibrated multimodal deep learning model for PTC diagnosis. METHODS: Using 384 patients from the Pang cohort (220 PTC, 164 benign) we created a stratified, untouched 20% holdout (n = 77). On the development set (n = 307) we performed 5-fold cross-validation and trained a multimodal model (v2) combining a ConvNeXt-Tiny ultrasound encoder, a domain-pretrained CTransPath cytology encoder, gated-attention multiple-instance learning (MIL) over cytology patches, and bidirectional cross-attention fusion. We compared this against a self-attention multimodal baseline (v1), unimodal ablations, and a modified Pang-style classical comparator that, because lesion masks were not released with the public dataset, used whole-image radiomics-style features rather than ROI-based features. Evaluation included bootstrap and Wilson confidence intervals, paired DeLong tests, McNemar tests, and a four-method calibration analysis (none, temperature, Platt, isotonic) using ensemble out-of-fold predictions and equal-mass binning. RESULTS: Multimodal v2 achieved holdout AUROC 0.977 (95% CI 0.949-0.996), Brier 0.042 (95% CI 0.014-0.079), sensitivity 0.977 (Wilson 0.882-0.996), and specificity 0.939 (Wilson 0.804-0.983) at the cross-validation Youden threshold. Across three random seeds (42, 7, 123), AUROC was 0.977 ± 0.0004 (mean ± SD). Paired DeLong showed our model statistically outperformed Pang's reimplemented Random Forest (p = 0.017) and XGBoost (p = 0.022) on identical holdout patients. Paired DeLong vs. the v1 baseline showed identical patient ranking (p = 1.0), but v2 showed numerically better operating-point and probability quality despite the identical ranking: Brier was roughly halved (0.042 vs. 0.083), MCC at threshold 0.5 increased by 0.105 (0.894 vs. 0.789), and uncalibrated expected calibration error fell from 0.114 to 0.042. The difference in paired binary decisions between v1 and v2 was, however, not statistically significant (McNemar p = 0.221). Calibration analysis revealed that temperature scaling, the de facto standard, was inappropriate for cross-validated ensembles, while isotonic regression on out-of-fold predictions reduced ECE by 53% on the v1 baseline. CONCLUSIONS: A multimodal model with domain-pretrained encoders and proper MIL aggregation achieves discriminative performance comparable to a modified Pang-style classical comparator and shows improved probability quality (lower Brier and calibration error), although these gains over the simpler v1 baseline did not reach statistical significance on the present holdout. We argue that AUROC alone is insufficient for evaluating clinical AI: discrimination and calibration are distinct properties, and operating-point performance is more directly relevant to deployment. We acknowledge that the present ultrasound branch operates on whole-image crops without explicit lesion localization and therefore should not yet be regarded as a lesion-specific clinical decision tool. Code and analysis scripts are released on GitHub.
MonoGID: geometry and illumination aware enhancement with distillation for self-supervised monocular endoscopic depth estimation.
Wei G, Bao Q, Shi T … +5 more , Li Y, Miao Y, He W, Tian Y, Jiang Z
BMC Med Imaging · 2026 Jun · PMID 42374259 · Full text
BACKGROUND: Monocular depth estimation in endoscopic scenes is a fundamental prerequisite for intraoperative 3D reconstruction and surgical navigation. However, factors such as weak textures, specular reflections, local...
BACKGROUND: Monocular depth estimation in endoscopic scenes is a fundamental prerequisite for intraoperative 3D reconstruction and surgical navigation. However, factors such as weak textures, specular reflections, local overexposure, and dynamic illumination changes in endoscopic environments can undermine the effectiveness of photometric-consistency-based self-supervised learning methods, leading to insufficient structural representation and limited prediction accuracy. METHODS: To address this issue, this paper presents a task-specific architectural extension of the EndoDAC framework for self-supervised monocular endoscopic depth estimation. Specifically, the proposed method adapts and integrates geometry- and illumination-aware feature enhancement, offline multi-generation self-distillation, and inference-stage structural fusion to improve feature representation, prediction refinement, and deployment efficiency under weak-texture and adverse illumination conditions. Experiments are conducted on the SCARED dataset for training and in-domain evaluation, while zero-shot cross-domain testing is performed on the Hamlyn dataset. RESULTS: The results show that the proposed EndoDAC-based extension improves depth estimation performance on the SCARED dataset and achieves lower error-oriented metrics than the baseline on Hamlyn zero-shot evaluation, with threshold accuracy remaining comparable to the baseline. CONCLUSIONS: The proposed method demonstrates that task-specific adaptation of geometry- and illumination-aware feature enhancement, offline self-distillation, and inference-stage fusion can improve an EndoDAC-based self-supervised endoscopic depth estimation pipeline. These results support the effectiveness of the proposed architectural adaptation, while also indicating that further work is needed to improve cross-domain scale consistency and robustness under more challenging surgical conditions.
Application of transformer attention mechanism-based multimodal deep learning model in the diagnosis of papillary thyroid carcinoma.
Xu M, Zhu X, Lu Z … +3 more , Xu W, Shen C, Yang J
BMC Med Imaging · 2026 Jun · PMID 42374243 · Full text
OBJECTIVE: A Transformer-based multimodal deep learning model was developed to enhance ultrasound imaging diagnosis of PTC and benign nodules. METHODS: This study included 491 thyroid nodule patients from two centers, wi...
OBJECTIVE: A Transformer-based multimodal deep learning model was developed to enhance ultrasound imaging diagnosis of PTC and benign nodules. METHODS: This study included 491 thyroid nodule patients from two centers, with 406 from the Suzhou Ninth People's Hospital divided into training (n = 284) and validation (n = 122) sets, and 85 from Jiangsu Shengze Hospital as an external test set. The dataset comprised 232 benign nodules and 259 papillary thyroid carcinoma cases. A comparison of 34 deep learning architectures and four traditional models was conducted, leading to the proposal of an Efficientnetv2scbamTrans fusion model. Five ablation experiments assessed module contributions, while SHAP and Grad-CAM were used for interpretability, with energy concentration, peak center distance, and spatial IoU as performance indicators. RESULTS: DenseNet201 excelled on the internal validation set but showed overfitting with an AUC of 0.702 on the external test set. The unimodal imaging baseline had an AUC of 0.782, which improved to 0.811 with clinical features, reducing overfitting risk. Directly fusing radiomics features yielded no improvement, maintaining an AUC of 0.751. The proposed Efficientnetv2scbamTrans model reached an AUC of 0.985 on the internal validation set. More importantly, on the independent external test set (Jiangsu Shengze Hospital), the model achieved an AUC of 0.986 (95% CI: 0.967-1.000), with 96.4% sensitivity and 96.7% specificity. On this external test set, our proposed model significantly outperformed the traditional DLR model by an AUC margin of 0.199 (p < 0.001). SHAP analysis indicated clinical features altered radiomics decision weights, and Grad-CAM showed an increase in spatial IoU from 0.097 to 0.349, enhancing visual localization. CONCLUSION: The study developed a multimodal deep learning model using thyroid ultrasound, improving diagnostics with interaction and attention mechanisms via CBAM and Transformer.
Multi-scale deformable attention fusion network with global context modeling for chest X-ray lesion segmentation.
Li X, Huang X, Zhan J … +4 more , Liu Z, Yi Y, Luo Y, Xie Y
BMC Med Imaging · 2026 Jun · PMID 42374211 · Full text
Chest X-ray segmentation is a critical technique for the screening and diagnosis of common thoracic diseases such as pneumonia and COVID-19, providing essential support for clinical diagnosis and quantitative analysis. H...
Chest X-ray segmentation is a critical technique for the screening and diagnosis of common thoracic diseases such as pneumonia and COVID-19, providing essential support for clinical diagnosis and quantitative analysis. However, the inherent challenges of chest X-ray images, including blurry lesion boundaries and irregular contours, pose significant difficulties for accurate segmentation. To address this problem, we propose a Multi-Scale Deformable Attention Fusion Network (MSDAFNet) for lesion segmentation in chest X-rays. MSDAFNet adaptively perceives lesion regions of varying shapes and sizes, thereby enhancing the model's robustness to diverse pathological tissues. Specifically, we design a multi-scale deformable convolution attention module that employs channel grouping and dilated convolutions with different dilation rates to capture multi-scale feature maps without increasing parameters. Learnable offsets are introduced, allowing the convolution kernels to adaptively adjust sampling positions. Furthermore, considering the clustered distribution characteristics of small target lesion regions, a memory module is proposed before the decoder.This module establishes semantic correlations between image regions by introducing learnable key-value pairs and an attention mechanism, thereby supplementing convolutional features with the capability of modeling long-range dependencies.Extensive experiments conducted on two large-scale medical datasets, QaTa-COV19-v1 and QaTa-COV19-v2, demonstrate that MSDAFNet achieves higher accuracy and robustness in anatomical structure segmentation than existing methods, delivering competitive performance.
Association of AI-assisted quantitative coronary plaque burden and CT-derived fractional flow reserve with major adverse cardiovascular events.
Zhang X, Bao M, Wu Y … +2 more , Qi H, Xing Y
BMC Med Imaging · 2026 Jun · PMID 42366355 · Full text
RATIONALE AND OBJECTIVE: This single-center retrospective study evaluated the associations of AI-quantified coronary plaque parameters and CT-derived fractional flow reserve (CT-FFR) with major adverse cardiovascular eve...
RATIONALE AND OBJECTIVE: This single-center retrospective study evaluated the associations of AI-quantified coronary plaque parameters and CT-derived fractional flow reserve (CT-FFR) with major adverse cardiovascular events (MACEs) in patients with coronary artery disease, and derived optimal risk cutoff values for plaque burden. METHODS: A total of 381 patients who underwent CCTA were consecutively enrolled. MACEs were defined as a composite of all-cause death, myocardial infarction (fatal and nonfatal), heart failure death, malignant arrhythmia, coronary revascularization, and rehospitalization for angina exacerbation. Maximum follow-up was 18 months. Risk cutoff values were derived from receiver operating characteristic analysis. Univariate and multivariate Cox regression, Kaplan-Meier analysis, and five predictive models (plaque model, CT-FFR model, combined model, LASSO-Cox, and Cox survival neural network) were constructed. RESULTS: Among 381 patients, 67 (17.6%) developed MACEs. All six total plaque parameters showed significant associations with MACEs. In multivariate Cox regression, total noncalcified percent atheroma volume (NCPAV) > 4.68% emerged as the strongest predictor (HR 5.073, 95% CI 2.930-8.786, P < 0.001). Analyzed continuously, each 1-SD increase in total-NCPAV conferred an HR of 1.82 (95% CI 1.54-2.14, P < 0.001). The combined model C-index was 0.750 (95% CI 0.696-0.804; optimism-corrected 0.708), comparable to the plaque model alone (0.744, 95% CI 0.686-0.801; corrected 0.705). The LASSO-Cox and Cox survival neural network models achieved C-indices of 0.747 (95% CI 0.674-0.816) and 0.730 (95% CI 0.628-0.833), respectively. In landmark sensitivity analyses excluding early events, the combined model C-index rose to 0.792, with the likelihood ratio test P value narrowing from 0.117 to 0.061, suggesting a trend toward incremental value for CT-FFR after accounting for potential incorporation bias. CONCLUSIONS: AI-quantified total noncalcified plaque burden was the strongest predictor of MACEs. The addition of CT-FFR to plaque parameters did not provide a clinically meaningful or statistically significant improvement in overall model performance, including discrimination, model fit, reclassification, or discrimination slope. Although landmark analyses suggested a possible trend toward incremental value after exclusion of early revascularization-driven events, this finding should be considered exploratory and requires further validation. Vessel-specific analyses identified RCA plaque burden as having the greatest prognostic weight among the target vessels; however, this exploratory finding also warrants confirmation in independent cohorts.
Tract-based χ-separation imaging differentiates multiple sclerosis from neuromyelitis optica spectrum disorder by mapping iron and myelin signatures.
Wu S, Xie Y, Zhang Y … +6 more , Yan S, Zhu H, Duan B, Dimov AV, Zhu W, Wang Y
BMC Med Imaging · 2026 Jun · PMID 42365292 · Full text
BACKGROUND: Multiple sclerosis (MS) and neuromyelitis optica spectrum disorder (NMOSD) frequently present with overlapping clinical and radiological features, complicating differential diagnosis. Iron and myelin abnormal...
BACKGROUND: Multiple sclerosis (MS) and neuromyelitis optica spectrum disorder (NMOSD) frequently present with overlapping clinical and radiological features, complicating differential diagnosis. Iron and myelin abnormalities are central to their pathophysiology, yet their distinct contributions in vivo remain poorly characterized. This study applied tract-based χ-separation imaging to evaluate susceptibility-related microstructural changes in normal-appearing white matter (NAWM) and lesions in MS and NMOSD, aiming to identify imaging patterns between these demyelinating diseases. METHODS: Eighty patients (48 MS, 32 AQP4-IgG-positive NMOSD) and 39 healthy controls were included. Quantitative susceptibility mapping (QSM) with χ-separation was performed on a 3D multi-echo gradient-echo sequence to obtain bulk susceptibility and to derive model-based positive susceptibility (χ) and negative susceptibility (χ sources. After masking lesions, mean values of susceptibility, χ and χ were extracted from 20 JHU white matter tracts, and lesions were compared with tract-matched NAWM within each disease group. Key features were identified using LASSO with cross-validation, and receiver operating characteristic (ROC) analysis was used to assess discrimination between MS and NMOSD. RESULTS: The three groups were similar in age and sex distribution. Mean disease duration was 4.12 ± 4.58 years in MS and 3.61 ± 4.26 years in NMOSD. Compared with NMOSD, MS showed reduced χ in 12 NAWM tracts and increased χ in 5 tracts (all FDR-corrected P < 0.05), particularly in the forceps major, inferior fronto-occipital fasciculus, and inferior longitudinal fasciculus, whereas no significant NAWM differences were found between NMOSD and HCs. In lesion-NAWM comparisons, MS lesions showed reduced χ in 11 tracts and increased χ in 16 tracts, whereas NMOSD lesions showed corresponding changes in 4 tracts. Conventional QSM did not observe corresponding group differences. Tract-wise χ metrics showed exploratory discriminatory value, with the internally validated classification model yielding a pooled held-out AUC of 0.771 (95% CI: 0.660-0.868). CONCLUSIONS: Tract-based χ-separation sensitively captures susceptibility-related microstructural alterations and reveals distinct χ/χ patterns in MS and NMOSD. These findings support the potential of tract-wise χ metrics as imaging markers for characterizing disease-specific white matter changes.
CT-derived quantitative imaging biomarkers for hypoxemia risk in repeated lung resection: a retrospective study.
Peng J, Wang J, Dong X … +9 more , Miao J, Yang H, Wang C, Guo G, Wang Y, Yang Y, Wei Q, Xiong S, Zhao L
BMC Med Imaging · 2026 Jun · PMID 42365255 · Full text
BACKGROUND: This study sought to elucidate CT-derived quantitative imaging biomarkers and clinical parameters predisposing to arterial oxygen desaturation during one-lung ventilation (OLV) in individuals undergoing repea...
BACKGROUND: This study sought to elucidate CT-derived quantitative imaging biomarkers and clinical parameters predisposing to arterial oxygen desaturation during one-lung ventilation (OLV) in individuals undergoing repeated lung resections, to improve understanding of preoperative factors associated with hypoxemia risk during OLV in this population. METHODS: A retrospective analysis was performed on 139 patients who underwent repeated lung resections between January 2022 and December 2023 at Yunnan Cancer Hospital. Preoperative CT-derived quantitative imaging biomarkers, including functional lung volume percentage (FLV%) and low attenuation volume percentage (LAV%), were extracted from routine CT scans alongside other clinical variables. Thirteen potential risk factors were first examined using Pearson's correlation and univariate regression analysis to identify variables significantly associated with hypoxemia. Variance inflation factor (VIF) tests were applied to exclude multicollinearity before performing multivariable logistic regression to determine independent risk factors. RESULTS: The incidence of hypoxemia during OLV in repeated lung resections was 16.55%. Four independent risk factors were identified: proportion of functional lung volume in the dependent lung as a key quantitative imaging biomarker (FLV%-dependent lung: OR 0.92, 95% CI 0.85-0.99; p = 0.038), a history of prior hypoxemia (OR 7.64, 95% CI 1.65-35.31; p = 0.009), moderate to severe postoperative pain at 48 hours (OR 5.59, 95% CI 1.95-16.03; p = 0.001), and forced expiratory volume in one second (FEV1% predicted: OR 1.04, 95% CI 1.00-1.08; p = 0.038). CONCLUSION: CT-derived quantitative imaging biomarker FLV%-dependent lung and FEV% predicted, a history of prior hypoxemia and moderate-to-severe acute postoperative pain within 48 hours after the initial lung resection, are associated with hypoxemia risk in repeated lung resections.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe