BMC Medical Imaging[JOURNAL]
Sun
200 papers
Novel markers on microvascular flow imaging for identifying holoprosencephaly.
Li Z, Cui Y, Guo C … +6 more , Zhang J, Wang L, Jia Z, Wu Y, Wu Q, Sun L
BMC Med Imaging · 2026 Jun · PMID 42251271 · Full text
OBJECTIVES: To establish gestational age-specific reference ranges for two new indices, the forehead-anterior cerebral artery angle (FACAA) and forehead-anterior cerebral artery distance (FACAD), and to objectively evalu...
OBJECTIVES: To establish gestational age-specific reference ranges for two new indices, the forehead-anterior cerebral artery angle (FACAA) and forehead-anterior cerebral artery distance (FACAD), and to objectively evaluate their diagnostic effectiveness, measured using Doppler technology with microvascular flow imaging, for identifying holoprosencephaly, even the semilobar and lobar types, in early pregnancy. METHODS: We evaluated FACAA and FACAD in 462 normal fetuses between 12 and 34 gestational weeks (GW) to generate normative reference ranges. Additionally, we evaluated these two indices in 41 fetuses with holoprosencephaly and 34 fetuses with similar two-dimensional ultrasound features (the study group), including 20 fetuses with agenesis of the corpus callosum and 14 with isolated absent septum pellucidum/septo-optic dysplasia. RESULTS: In the normal group, FACAA was almost stable throughout pregnancy, whereas FACAD increased with GW. All fetuses with holoprosencephaly had both smaller FACAA (13.99-38.00°, ≤5th percentile) and shorter FACAD (1.6-10.5 mm, ≤5th percentile). The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and negative likelihood ratio (LR-) of FACAA for the prediction of holoprosencephaly in fetuses with an intracranial malformation were 100% (95% confidence interval [CI]: 89.3-100.0%), 100% (95% CI: 87.4-100.0%), 100% (95% CI: 89.3-100%), 97.1% (95% CI: 82.9-99.8%), and 0, respectively. The sensitivity, specificity, PPV, NPV, and LR- of FACAD for the prediction of holoprosencephaly in fetuses with an intracranial malformation were 97.6% (95% CI: 85.9-99.9%), 100% (95% CI: 87.0-100.0%), 100% (95% CI: 89.3-100%), 97.1% (95% CI: 82.9-99.8%), and 0.02 (95% CI: 0.003-0.165), respectively. CONCLUSION: FACAA and FACAD are sensitive, easy-to-measure indicators for the prenatal identification of holoprosencephaly, even in early pregnancy. These indices provide objective assessment tools for the diagnosis of fetuses with holoprosencephaly. TRIAL REGISTRATION: Not applicable.
Ultrasound-based multiregional radiomics nomogram for predicting recurrence in HER2-positive breast cancer.
Fang X, Liu Y, Zhang X … +6 more , Fan W, Qin Z, Yang Z, Tian J, Zhang L, Cui H
BMC Med Imaging · 2026 Jun · PMID 42251269 · Full text
PURPOSE: Accurate prediction of recurrence risk is crucial for personalizing therapy in human epidermal growth factor receptor 2-positive (HER2-positive) breast cancer. We aimed to develop an interpretable machine learni...
PURPOSE: Accurate prediction of recurrence risk is crucial for personalizing therapy in human epidermal growth factor receptor 2-positive (HER2-positive) breast cancer. We aimed to develop an interpretable machine learning model integrating multimodal data to address this need. METHODS: This retrospective study enrolled 148 patients with human epidermal growth factor receptor 2 (HER2)-positive breast cancer between 2017 and 2021. On preoperative ultrasound images, the intratumoral region (region of interest B, ROIB) was manually delineated by experienced radiologists. Based on the manually segmented ROIB, the 5 mm inward inner peritumoral region (ROIC) and 5 mm outward outer peritumoral region (ROIA) were automatically generated via the built-in adaptive tool of 3D Slicer, with non-breast tissues excluded manually. Radiomic features were extracted from each ROI using PyRadiomics. A pre-fusion strategy (i.e. feature fusion) was used to construct combined feature sets (ROIA+B and ROIA+B+C [ROI All]), and the radiomic model based on ROI All was defined as Rad All. After feature selection via Student's t-test/Mann-Whitney U test, Pearson correlation analysis, maximum relevance minimum redundancy (mRMR) algorithm, and least absolute shrinkage and selection operator (LASSO) regression, multiple machine learning models were compared, and the support vector machine (SVM) was selected as the optimal algorithm for radiomic model construction. Independent clinical risk factors were screened by Cox regression analysis, and a post-fusion strategy integrated these factors with the radiomic signature to build a combined nomogram model. Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC), calibration curves, decision curve analysis (DCA), DeLong test, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Survival analysis was performed with the Kaplan-Meier method, and SHapley Additive exPlanations (SHAP) analysis was applied to enhance model interpretability. RESULTS: Among the tested radiomic models, the Rad All model that integrated intratumoral, inner peritumoral, and outer peritumoral features showed the best overall performance in the internal validation set, with an AUC of 0.820 (95% CI: 0.689-0.951). The combined nomogram incorporating the Rad All signature and independent clinical risk factors (irregular mass shape, abnormal posterior echo features, lymph node metastasis, and progesterone receptor[PR] negativity) achieved an AUC of 0.905 (95% CI: 0.817-0.992), which was higher than the radiomic‑only and clinical‑only models. Statistical assessments including DeLong test, NRI, and IDI supported the incremental predictive value of the combined model (all p < 0.05). The model exhibited acceptable calibration and favorable clinical net benefit in decision curve analysis. Survival analysis using the Rad All model enabled effective risk stratification in both training and internal validation sets. SHAP analysis identified lbp‑2D_Agrlrm_ARunLengthNonUniformityNormalized as the most important radiomic feature, with high values related to increased recurrence risk. CONCLUSION: In conclusion, integrating intratumoral, inner peritumoral, and outer peritumoral radiomic features (Rad All model) may provide more reliable predictive performance for postoperative recurrence in HER2-positive breast cancer compared with single-region radiomic models. The combined nomogram incorporating Rad All and clinical risk factors further improves predictive efficacy and may serve as a supplementary tool for clinical decision-making. SHAP analysis enhances model interpretability by identifying key predictive features. However, given the lack of independent external validation, the generalizability of our models needs to be further verified in future prospective multi-center studies.
Radiomics in thyroid nodule assessment.
Suo W, Guo C, Zhang X … +5 more , Li H, Lai W, Zhang J, Jin J, Zhang J
BMC Med Imaging · 2026 Jun · PMID 42249361 · Full text
This narrative review was conducted by searching PubMed and Google Scholar using keywords such as 'radiomics' AND 'thyroid nodules' from March 2019 to March 2024. Inclusion criteria focused on peer-reviewed studies in En...
This narrative review was conducted by searching PubMed and Google Scholar using keywords such as 'radiomics' AND 'thyroid nodules' from March 2019 to March 2024. Inclusion criteria focused on peer-reviewed studies in English on radiomics applications; 50 studies were ultimately evaluated after excluding non-relevant works. The purpose of this article is to systematize the utilization of radiomics in assessing thyroid nodules, aiming to furnish valuable theoretical insights. As an emerging interdisciplinary domain, radiomics integrates medical imaging with computer data analysis techniques, thus constituting a potent instrument for precise diagnosis and comprehensive evaluation of thyroid nodules. Through a scrutiny of the principles, methods, and applications of radiomics in thyroid nodule evaluation, we aim to elucidate its potential for enhancing diagnostic accuracy, assessing disease progression, and mitigating overdiagnosis. Additionally, this study offers a critical assessment of the inherent strengths and limitations of the employed methodology.Clinical trial number Not applicable.
Deformable image registration accuracy: impact of user-defined parameter selection on contour propagation for deep inspiration breath-hold and free breathing breast radiotherapy.
Knippen S, Pargmann L, Weimann S … +3 more , Hildebrandt G, Borm KJ, Duma MN
BMC Med Imaging · 2026 Jun · PMID 42249351 · Full text
PURPOSE: Deformable Image Registration (DIR) is increasingly used in breast radiotherapy planning, but the impact of different parameter settings on clinical accuracy remains unclear. This study systematically evaluates...
PURPOSE: Deformable Image Registration (DIR) is increasingly used in breast radiotherapy planning, but the impact of different parameter settings on clinical accuracy remains unclear. This study systematically evaluates how focus region selection and registration direction affect DIR performance, providing evidence-based recommendations for clinical implementation. MATERIALS AND METHODS: We analyzed 73 patients with left-sided breast cancer who underwent both free breathing (FB) and deep inspiration breath-hold (DIBH) planning CT scans. Clinical target volume (CTV), heart, left anterior descending artery, and both lungs were retrospectively contoured by one radiation oncologist on both datasets. Deformable Image Registration was performed bidirectionally (FB↔DIBH) using four different approaches: no focus region, CTV-focused, heart-focused, and surgical clip-focused registration. Volume differences and Dice Similarity Index (DSI) were calculated to assess registration accuracy. Generalized estimating equations analyzed possible correlations. RESULTS: Focus region selection critically impacted DIR accuracy. CTV-focused registration achieved highest CTV overlap (DSI: 0.96 vs. 0.91 without focus, p < 0.005) but compromised other structures. Heart-focused registration optimized cardiac structure accuracy but reduced CTV precision. Registration direction showed minimal impact on volume differences but affected DSI values. Tidal volume significantly affected both CTV (p < 0.005) and heart DSI (p < 0.007), with opposite effects on each structure. Manual contour adjustment remained necessary in all cases regardless of parameter selection. CONCLUSION: DIR performance in breast radiotherapy is highly dependent on user-defined parameters, with no single configuration optimizing all structures simultaneously. Focus region selection creates a trade-off between local accuracy and global registration quality. Clinical implementation requires structure-specific parameter optimization and mandatory manual review. These findings establish practical limitations and benefits of hybrid DIR algorithms for breast radiotherapy workflows and highlight the continued necessity of expert oversight.
Predicting STAS in peripheral stage I lung adenocarcinoma: the incremental value of CT-based tumor disappearance rate, with a focus on part-solid nodules.
Du H, Ji Y, Liu C … +5 more , Yao D, Jia C, Wang L, Xue X, Li X
BMC Med Imaging · 2026 Jun · PMID 42249345 · Full text
BACKGROUND: Tumor spread through air spaces (STAS) is a critical prognostic factor in peripheral stage I lung adenocarcinoma (LUAD) and substantially influences surgical decision-making. However, accurate preoperative pr...
BACKGROUND: Tumor spread through air spaces (STAS) is a critical prognostic factor in peripheral stage I lung adenocarcinoma (LUAD) and substantially influences surgical decision-making. However, accurate preoperative prediction of STAS remains challenging. This study aimed to investigate the value of computed tomography (CT)-derived tumor disappearance rate (TDR) as a novel imaging biomarker for STAS prediction, and to evaluate the incremental benefit of combining TDR with consolidation tumor ratio (CTR) in part-solid nodules. METHODS: This retrospective study enrolled 244 patients with peripheral stage I LUAD who underwent surgical resection at Xuzhou Hospital Affiliated to Jiangsu University between January 2022 and December 2023. The multivariate generalized additive model (GAM) was employed to characterize the adjusted dose-response relationship between TDR and STAS, with threshold effect analysis performed via the log-likelihood ratio test. The predictive performance of TDR, CTR, and the combined TDR-CTR index (COM) was compared using ROC curves, DeLong test, and bootstrap resampling validation. RESULTS: STAS was identified in 51 (20.9%) patients. The STAS-positive group exhibited a significantly lower TDR compared with the STAS-negative group (0.26 +/- 0.32 vs. 0.51 +/- 0.44, P < 0.001). Multivariate analysis identified TDR and spiculation sign as independent risk factors for STAS. TDR demonstrated a non-linear association with STAS (P < 0.001), with a threshold value of 0.610. In the part-solid nodule subgroup, the combined TDR-CTR index achieved an AUC of 0.793 (95% CI: 0.654-0.932), significantly outperforming CTR alone (AUC = 0.525, P = 0.040). Bootstrap validation confirmed favorable stability. While promising in this exploratory subgroup analysis, further validation in larger cohorts is warranted due to the wide confidence interval of the combined TDR-CTR index. CONCLUSIONS: CT-derived TDR is a valuable biomarker for preoperative STAS risk stratification in peripheral stage I LUAD. The combined TDR-CTR index significantly enhances STAS prediction in part-solid nodules, addressing the limitations of CTR alone and supporting individualized surgical planning.
Enhanced vertebrae localization in CT volumes: a two-stage deep learning framework.
Liu H, Han J, Jiang L … +10 more , Zhang Y, Wen X, Xi Y, Zhang Y, Yang C, Ge R, Tang H, Wang S, Feng Q, Chen Y
BMC Med Imaging · 2026 Jun · PMID 42249287 · Full text
BACKGROUND: Vertebral landmark localization in computed tomography (CT) volumes is crucial for spinal pathological diagnosis, postoperative assessment and surgical planning. However, vertebral localization within high-re...
BACKGROUND: Vertebral landmark localization in computed tomography (CT) volumes is crucial for spinal pathological diagnosis, postoperative assessment and surgical planning. However, vertebral localization within high-resolution three-dimensional CT volumes still poses prominent challenges. METHODS: This study introduces a novel two-stage deep learning framework that effectively addresses challenges such as large spatial spans and high morphological similarity of vertebrae. The first stage employs the improved squeeze and excitation V-Net (ISE-VNet) for coarse segmentation of the spinal column. The second stage utilizes a 3D spatial-generalized differential spatial-to-numerical transform (DSNT) module for precise localization of individual vertebrae. RESULTS: Our framework significantly improves identification rates from 81.15% to 96.32% and reduces localization errors from 7.6 mm to 2.1 mm, outperforming state-of-the-art methods. CONCLUSION: This approach provides a robust solution for clinical vertebral landmark localization in 3D medical images. This design effectively enables efficient and focused analysis of vertebral local regions and it holds significant value in both scientific research and clinical practice.
Very high frequency ultrasonographic characteristics of poroma: a retrospective observational study.
Dong B, Xia H, Xia Y … +3 more , Chai L, Xie B, Liu Y
BMC Med Imaging · 2026 Jun · PMID 42243726 · Full text
BACKGROUND: Poroma is a rare benign skin tumor, diagnostic challenges of which require advanced imaging. Herein we investigate the ultrasonographic manifestations and clinical value of the very high frequency(VHF) ultras...
BACKGROUND: Poroma is a rare benign skin tumor, diagnostic challenges of which require advanced imaging. Herein we investigate the ultrasonographic manifestations and clinical value of the very high frequency(VHF) ultrasound in the auxiliary diagnosis of Poroma. METHODS: Clinical data of 24 cases pathologically diagnosed as poroma were retrospectively analyzed by means of high-resolution VHF ultrasound, including size, location, shape, margin, boundary, internal echo and blood flow patterns. RESULTS: VHF ultrasound examination shows that most poroma are located in the epidermis and dermis. The maximum diameter was 2.6-30 (11.86 ± 6.31) mm. The tumor has clear borders, and the echo intensity of most lesions(22/24) was heterogeneous hypoechoic. Most lesions(18/24) showed small duct-like anechoic areas, 6 lesions(6/24) were detected with punctate calcification. In terms of blood flow, most lesions(18/24) showed abundant blood flow signals, which were dendritic. CONCLUSIONS: The VHF ultrasound lacks specificity in diagnosing poroma and currently makes accurate qualitative diagnosis difficult, but it holds significant value for assessment. It can clearly define the lesion's location, layers, number, size, boundaries, relationship to surrounding structures, and internal blood supply. This provides an objective basis for selecting the next diagnostic or treatment plan, and it can also serve as an evaluation. method for postoperative follow-up.
Optimizing shielding protocols in tube current modulation chest computed tomography: a phantom study.
Sun Q, Fan Y, Zhu P … +2 more , Wang J, Liang B
BMC Med Imaging · 2026 Jun · PMID 42243725 · Full text
BACKGROUND: Growing public radiation protection awareness has heightened clinical focus on shielding. This study examines the relationship between shielding material placement, the overranging (OR), and scout image visib...
BACKGROUND: Growing public radiation protection awareness has heightened clinical focus on shielding. This study examines the relationship between shielding material placement, the overranging (OR), and scout image visibility in chest tube current modulation (TCM) computed tomography (CT). It analyzes changes in overall radiation dose within the patient imaging area and image quality of tissues at the scanning boundary near the shield to provide evidence for selecting optimal clinical shielding protocols. MATERIALS AND METHODS: This study utilized an anthropomorphic phantom to perform chest CT scans under TCM on a GE Revolution CT scanner. Lead shielding was placed intra- and extra-OR, categorized as scout-visible or invisible, to assess its impact on imaging area dose and changes in boundary organ image quality versus an unshielded control. RESULTS: Intra-OR shields increased the radiation dose in the imaging area, with volume CT dose index (CTDI) rising by up to 22.9% (range: 0.43%-22.9%) depending on the collimation-pitch combination, and caused degradation of image quality at the scan boundary (SNR and CNR reductions approximately twice those of the extra-OR shields group, with 50% of image scores ≤ 2, indicating limited diagnostic value). In contrast, extra-OR shields produced minimal changes: CTDI increased by less than 3% in all cases, SNR/CNR reductions were below 10%, and all images scored ≥ 3 (meeting diagnostic requirements). Furthermore, scout-visible lead shielding was associated with increased radiation dose and reduced image quality. Conversely, scout-invisible lead shielding had a negligible effect on radiation dose but led to a more substantial degradation in image quality relative to the visible configuration. CONCLUSIONS: Based on this phantom study, to optimize radiation protection and image quality in clinical chest TCM CT examinations, it is suggested that the shielding device be placed in the extra-OR region. However, as these findings derive from phantom experiments, further validation in clinical settings is warranted before routine application.
Knowledge-guided brain tumor segmentation via synchronized visual-semantic-topological prior fusion.
Zhang MD, Pan KW
BMC Med Imaging · 2026 Jun · PMID 42237259 · Full text
Brain tumor segmentation requires precise delineation of hierarchical structures from multi-sequence MRI. However, existing deep learning methods primarily rely on visual features, showing insufficient discriminative pow...
Brain tumor segmentation requires precise delineation of hierarchical structures from multi-sequence MRI. However, existing deep learning methods primarily rely on visual features, showing insufficient discriminative power in ambiguous boundary regions. Moreover, they lack explicit integration of medical domain knowledge such as anatomical semantics and geometric topology. We propose a knowledge-guided framework, Synchronized Tri-modal Prior Fusion (STPF), that explicitly integrates three heterogeneous knowledge priors: pathology-driven differential features (T1ce-T1, T2-FLAIR, T1/T2) encoding contrast patterns, unsupervised semantic descriptions transformed into voxel-level guidance via spatialization operators, and geometric constraints extracted through persistent homology analysis. A dual-level fusion architecture dynamically allocates prior weights at the voxel level based on confidence and at the sample level through hypernetwork-generated conditional vectors. Furthermore, nested output heads structurally ensure the hierarchical constraint ET⊆TC⊆WT. STPF achieves a mean Dice coefficient of 0.868 on the BraTS 2020 dataset, surpassing the best baseline by 2.6% points (3.09% relative improvement). Notably, five-fold cross-validation yields coefficients of variation between 0.23% and 0.33%, demonstrating stable performance. Additionally, ablation experiments show that removing topological and semantic priors leads to performance degradation of 2.8% and 3.5%, respectively. By explicitly integrating medical knowledge priors-anatomical semantics and geometric constraints-STPF improves segmentation accuracy in ambiguous boundary regions while demonstrating generalization capability and clinical deployment potential.
Dual-stream token fusion with Swin Transformer and lesion-aware tokens for gastric metaplasia classification in IoMT-assisted deployment.
Rokhsati H, Fasihfar Z, Shahriari HR … +1 more , Rezaee K
BMC Med Imaging · 2026 Jun · PMID 42237245 · Full text
BACKGROUND: Gastric intestinal metaplasia (GIM) is often visually inconspicuous on routine endoscopy, while many artificial intelligence systems rely on dense supervision, lack calibrated probabilities, or provide limite...
BACKGROUND: Gastric intestinal metaplasia (GIM) is often visually inconspicuous on routine endoscopy, while many artificial intelligence systems rely on dense supervision, lack calibrated probabilities, or provide limited evidence of transfer across datasets and devices. We developed a single-frame, four-class endoscopic classifier that jointly models anatomic context and metaplasia status. METHODS: We propose a dual-stream architecture that combines an RGB Swin Transformer backbone with SHAP-guided lesion-aware multi-scale auxiliary tokens. The two streams are fused through class-token attention to obtain a compact and interpretable representation without relying on video context or pixel-level masks. To address class imbalance, training combines class-balanced focal loss, balanced-softmax/logit adjustment, class-aware sampling, and validation-tuned per-class thresholds. The model was evaluated on an internal four-class cohort of 666 endoscopic still frames and externally assessed, without retuning, on an unseen public endoscopy dataset recast as a binary normal-versus-abnormal task. RESULTS: On the internal cohort, the proposed model achieved macro-AUROC 0.950, macro-AUPRC 0.920, macro-F1 0.926, and accuracy 0.954, with expected calibration error 0.034 and inference latency of approximately 182 ms per 224 × 224 frame. On the unseen external dataset, the model retained AUROC 0.940, AUPRC 0.900, and F1 0.890 using frozen operating thresholds. Comparative and ablation analyses indicated that lesion-aware tokenization and token fusion contributed more strongly to performance gains than backbone choice alone, while calibration quality also improved. CONCLUSIONS: A dual-stream, single-frame token-fusion model can provide accurate, calibrated, and interpretable classification of gastric intestinal metaplasia while remaining compatible with low-latency edge-oriented inference. Although broader multicenter validation is still required, the results support the feasibility of deployment-oriented AI assistance for endoscopic GIM triage.
Prognostic value of [F]fluorodeoxyglucose PET/CT in esophageal cancer patients treated with anti-PD-1 inhibitors.
Whi W, Sun JM, Choi JY … +1 more , Lee H
BMC Med Imaging · 2026 Jun · PMID 42237244 · Full text
BACKGROUND: To evaluate the prognostic value of [F]fluorodeoxyglucose (FDG) PET/CT in patients with recurrent or metastatic esophageal cancer treated with anti-programmed death 1 (PD-1) inhibitors and chemotherapy. METHO...
BACKGROUND: To evaluate the prognostic value of [F]fluorodeoxyglucose (FDG) PET/CT in patients with recurrent or metastatic esophageal cancer treated with anti-programmed death 1 (PD-1) inhibitors and chemotherapy. METHODS: This retrospective study enrolled 43 patients treated with anti-PD-1 inhibitors plus chemotherapy between March 2022 and June 2025. We analyzed pre-therapeutic maximum and mean standardized uptake values (SUVmax, SUVmean), metabolic tumor volume (MTV), and total lesion glycolysis (TLG). Associations with overall survival (OS) and progression-free survival (PFS) were assessed using Kaplan-Meier and Cox regression analyses. Subgroup analyses evaluated early metabolic changes in patients with follow-up PET/CT (n = 15) and metabolic trajectory in those with initial staging PET/CT (n = 19). RESULTS: Forty-three patients (mean age, 65.0 ± 8.5 years; 34 men) were evaluated. High pre-therapeutic SUVmax, SUVmean, MTV, and TLG were significantly associated with poor OS based on the likelihood ratio test from univariate Cox regression (all p < 0.05). While pre-therapeutic parameters did not predict PFS, early reductions in volumetric parameters (Δ%MTV and Δ%TLG) on follow-up PET/CT were significantly associated with PFS (p < 0.05). Furthermore, patients exhibiting a "high-to-low" metabolic trajectory from initial staging to pre-therapeutic baseline demonstrated superior PFS compared to the "low-to-high" group (p = 0.02 for SUVmax; p < 0.001 for SUVmean). CONCLUSIONS: FDG PET/CT parameters were associated with survival outcomes; pre-therapeutic metabolic burden predicts OS, whereas dynamic metabolic changes and longitudinal trajectory showed associations with PFS in patients treated with chemo-immunotherapy.
Utility of deep learning for degree calculation of aortic arch calcification in chest-X ray.
Wu CK, Huang CY, Shen TX … +2 more , Tsai YS, Hsieh JW
BMC Med Imaging · 2026 Jun · PMID 42237240 · Full text
BACKGROUND: Aortic arch calcification (AoAC) is commonly classified into four grades according to the percentage of calcification observed in clinical practice, and the interpretation is typically based on visual assessm...
BACKGROUND: Aortic arch calcification (AoAC) is commonly classified into four grades according to the percentage of calcification observed in clinical practice, and the interpretation is typically based on visual assessment by clinicians. However, this manual evaluation process is time-consuming and may fail to detect subtle calcifications, potentially leading to grading inaccuracies. METHODS: This study presents a transformer-based model, termed Multi-Attention with Transformer Model (MATM), to improve the accuracy of AoAC grade classification. The proposed framework integrates multiple attention modules to enhance the representation of spatial features. In addition, the transformer mechanism incorporates positional information together with a hierarchical 16-block representation of the aortic arch, enabling fine-grained analysis of calcification distribution. RESULTS: The proposed method captures subtle calcification features and enables more accurate classification of AoAC grades. Experimental results demonstrate that the model can automatically estimate AoAC severity for both the traditional four-grade classification and the more detailed 16-grade classification, achieving an accuracy of up to 95.5%. CONCLUSIONS: The proposed method can reduce interpretation time and improve grading consistency for clinicians by minimizing variability caused by individual experience. Such AI-assisted assessment has the potential to standardize AoAC evaluation in future clinical practice.
Diagnostic accuracy of coronary CT angiography versus invasive coronary angiography for detecting coronary artery disease: a systematic review and Bayesian meta-analysis.
Alqahtani NG, Moawad MHED, Shati AA … +7 more , Bisht O, Abdul-Hafez HA, Shatoor AS, Alharasees M, Elkelish A, Alshanbari AS, Zabady AH
BMC Med Imaging · 2026 Jun · PMID 42231233 · Full text
BACKGROUND: The diagnostic performance of coronary computed tomography angiography, an emerging noninvasive modality used as an alternative to ICA for the assessment of coronary artery disease, is generally variable acro...
BACKGROUND: The diagnostic performance of coronary computed tomography angiography, an emerging noninvasive modality used as an alternative to ICA for the assessment of coronary artery disease, is generally variable across studies and clinical settings. OBJECTIVE: The objective of this study is to systematically assess the diagnostic accuracy of coronary computed tomography angiography (CCTA) in comparison with invasive coronary angiography (ICA) in the detection of anatomically significant CAD. METHODS: A systematic review and meta-analysis were conducted according to PRISMA guidelines. A search of PubMed, Scopus, and Web of Science for eligible studies through December 2025 was conducted. Studies were included if they reported CCTA diagnostic accuracy data with ICA as the reference standard. The risk of bias was assessed using QUADAS-2. The pooled sensitivity and specificity, along with the likelihood ratios and diagnostic odds ratio (DOR), were estimated using a Bayesian bivariate model. Summary receiver operating characteristic curves and Fagan nomograms were generated to assess overall performance and clinical utility. RESULTS: Twenty-seven studies encompassing 4461 patients were included in the quantitative synthesis. CCTA demonstrated high pooled sensitivity of 0.94 (95% posterior interval [PI] 0.892-0.969) and moderate specificity of 0.73 (95% PI 0.560-0.846). The pooled positive and negative likelihood ratios were 3.50 and 0.08, respectively, with a diagnostic odds ratio of 43.8, indicating strong overall discriminatory ability. Subgroup analyses showed higher accuracy in patients without prior coronary Intervention, while specificity was reduced in post-percutaneous coronary Intervention (PCI) or coronary artery bypass grafting (CABG) populations. A negative CCTA markedly reduced the post-test probability of CAD, supporting its value as a rule-out test. CONCLUSION: CCTA demonstrates high sensitivity and strong rule-out performance for anatomically significant CAD, particularly in appropriately selected low-to-intermediate-risk patients without prior coronary Intervention. However, its moderate specificity and reduced performance in complex post-PCI/CABG populations indicate that positive findings should be interpreted cautiously and may require confirmatory invasive or functional assessment. CLINICAL TRIAL NUMBER: Not applicable.
Deep learning-based neuroanatomical profiling reveals population-specific brain changes in multiple sclerosis: a large-scale Middle Eastern study.
Bawil MB, Shamsi M, Bavil AS
BMC Med Imaging · 2026 Jun · PMID 42226149 · Full text
BACKGROUND: Multiple sclerosis (MS) affects 2.8 million individuals worldwide, with Middle Eastern populations remaining underrepresented in neuroimaging research despite elevated regional prevalence rates. Large-scale c...
BACKGROUND: Multiple sclerosis (MS) affects 2.8 million individuals worldwide, with Middle Eastern populations remaining underrepresented in neuroimaging research despite elevated regional prevalence rates. Large-scale comparative studies between MS patients and healthy controls (HC) are essential for characterizing disease-specific brain changes and establishing normative biomarkers. This study aimed to perform comprehensive statistical characterization of brain structural changes in MS patients compared to HC using automated deep learning-based segmentation, establishing population-specific reference ranges for a Middle Eastern cohort. METHODS: This retrospective observational study analyzed 1,381 subjects (1,000 HC, 381 MS patients) from Northwest Iran. Four deep learning architectures were evaluated on multi-center training data (local cohort plus MSSEG dataset), with U-Net selected for automated FLAIR sequence segmentation. Neuroanatomically-informed lesion classification identified periventricular, deep, and juxtacortical white matter hyperintensities. Statistical analyses employed age group- and gender-stratified comparisons with comprehensive correlation assessments across ventricular and lesion load measures. RESULTS: U-Net achieved optimal segmentation performance (DSC = 88.8%, HD95 = 2.8 mm), supporting its selection for population-level analysis. Compared to healthy controls, MS patients exhibited markedly elevated structural burden across all age strata: ventricular load was 1.7-fold higher (normalized ratios: 1.39%→2.37%; t = - 18.92, Cohen's d = 1.139) and white matter lesion load was 2.4-fold higher (normalized ratios: 0.315%→0.749%; U = 11,963.0, rank-biserial r = 0.310), both with p < 10⁻⁶. Among lesion subtypes, periventricular lesions predominated (53.91 ± 20.62% of total burden), while anatomical distribution patterns showed no significant gender differences. Age-related structural changes were more pronounced in MS patients than in controls, with stronger correlations observed for both ventricular load (r = 0.403, p = 2.50 × 10⁻¹⁶) and lesion load (r = 0.266, p = 1.33 × 10⁻⁷). CONCLUSIONS: This study provides preliminary population-specific reference ranges for MS neuroimaging biomarkers in Middle Eastern populations, revealing lesion accumulation patterns with periventricular predominance. The automated segmentation and statistical framework address gaps in global MS research demographics.
Improving image quality and diagnostic confidence for PRETEXT staging in pediatric hepatoblastoma using thin-slice and low-energy virtual monochromatic images in dual-energy CT with deep learning image reconstruction algorithm.
Yin G, Guo T, Feng J … +5 more , Li H, Song Y, Zhang Y, Peng Y, Sun J
BMC Med Imaging · 2026 May · PMID 42226141 · Full text
BACKGROUND: Hepatoblastoma is the most common pediatric hepatic tumor, for which surgery is the primary treatment option. The PRETEXT (Pretreatment Extent of Disease) staging system, based on CT images, is a crucial basi...
BACKGROUND: Hepatoblastoma is the most common pediatric hepatic tumor, for which surgery is the primary treatment option. The PRETEXT (Pretreatment Extent of Disease) staging system, based on CT images, is a crucial basis for surgical planning. Therefore, improving the image quality and diagnostic confidence of PRETEXT staging impacts the overall therapeutic outcomes in pediatric hepatoblastoma. OBJECTIVE: To investigate whether thin-slice 40 keV dual-energy CT (DECT) images combined with a deep learning image reconstruction (DLIR) algorithm can improve image quality and diagnostic confidence for the evaluation of PRETEXT staging for pediatric hepatoblastoma. METHODS: This single-center retrospective study included 53 pediatric patients (mean age, 3.54 ± 2.26 years) with pathologically confirmed hepatoblastoma who underwent contrast-enhanced abdominal DECT. From the raw data, three distinct image series were reconstructed with a slice thickness of 0.625 mm for comparison: (A) standard-energy 68 keV VMI (virtual monoenergetic image) with 50% adaptive statistical iterative reconstruction-V (ASIR-V50%); (B) 68 keV VMI with high-level DLIR (DLIR-H); and (C) low-energy 40 keV VMI with DLIR-H. Objective image quality was quantified by the contrast-to-noise ratio (CNR) and Edge Rise Slope (ERS) of hepatic veins. Two independent radiologists performed PRETEXT staging and subjectively assessed image noise, hepatic vein visualization, and diagnostic confidence using a 5-point Likert scale. RESULTS: The final PRETEXT staging results showed no statistically significant difference among the three image groups. However, objectively, the 40 keV DLIR-H images demonstrated significantly superior ERS (83.73 ± 46.50), indicating the sharpest vessel boundaries (p < 0.001), and CNR values for the hepatic veins. Subjectively, the 40 keV DLIR-H images received the highest scores for hepatic vein visualization and diagnostic confidence (p < 0.001), and was the only group consistently deemed sufficient to meet all diagnostic requirements for staging. CONCLUSION: The 0.625 mm thin-slice 40 keV VMI in DECT combined with DLIR-H reconstruction provides superior image quality and significantly enhances the diagnostic confidence for PRETEXT staging, and may be considered for routine clinical use.
A Feasibility study of a two-step ADC-PSAD-GWR rule for stratifying prostate lesions prior to biopsy.
Chen H, Chen T, Li G … +5 more , Tian X, Han M, Li W, Xiang Q, Meng Z
BMC Med Imaging · 2026 Jun · PMID 42226128 · Full text
OBJECTIVES: To test the feasibility of a two-step rule integrating apparent diffusion coefficient (ADC), PSA density (PSAD), and gadolinium wash-in/wash-out rate (GWR) in stratifying prostate lesions for biopsy decision-...
OBJECTIVES: To test the feasibility of a two-step rule integrating apparent diffusion coefficient (ADC), PSA density (PSAD), and gadolinium wash-in/wash-out rate (GWR) in stratifying prostate lesions for biopsy decision-making, and to compare its performance with PI-RADS v2.1 within a retrospective single-center cohort. METHODS: This retrospective case series included 110 patients (mean age 70 ± 9 years) who underwent 3.0-T mpMRI and prostate surgery. ADC, PSAD, and GWR were measured from index lesions. Optimal cutoffs (ADC 0.747 × 10⁻³ mm²/s, PSAD 0.344 ng/mL², GWR 9.454%) were derived by 5-fold cross-validation. A two-step rule was constructed: step 1 assigned lesions to four ADC-PSAD quadrants (Q1-Q4); step 2 refined risk using GWR. Performance was compared descriptively with PI-RADS. RESULTS: The four-quadrant framework identified Q2 (low ADC/high PSAD) as the malignant-dominant quadrant and Q3 (high ADC/low PSAD) as predominantly benign. GWR refinement defined a subgroup (Q3 with low GWR) that contained no clinically significant prostate cancer (csPCa) among the 20 patients assigned. In the sensitivity analysis (all-grade PCa vs. benign), the same subgroup contained zero cancers. Compared with PI-RADS, the two-step rule avoided 35.1% (13/37) of unnecessary biopsies in PI-RADS ≥ 4 and detected all occult csPCa in PI-RADS ≤ 3 (12/12). Overall csPCa detection was 97.6% (41/42). Sensitivity analysis yielded consistent findings: unnecessary biopsy avoidance of 46.4% (13/28) and all-grade cancer detection of 98.0% (48/49). CONCLUSION: This feasibility study suggests that the two-step ADC-PSAD-GWR rule may help guide biopsy decisions by identifying a low-risk subgroup where biopsy may be safely deferred. However, due to the retrospective, single-center design and lack of external validation, the findings should be considered hypothesis-generating. No claim of superiority over PI-RADS or reduction in patient-important outcomes (e.g., mortality) is made.
Biventricular dysfunction and blood oxygenation deficits in obstructive sleep apnea: a prospective study with non-contrast cardiac MRI.
Wang J, Cheng Z, Du J … +7 more , Liu A, Ni Y, Yu H, Guo Y, Ma E, Zhang X, Liu M
BMC Med Imaging · 2026 Jun · PMID 42226066 · Full text
BACKGROUND: Conventional cardiac magnetic resonance imaging(CMR) metrics may remain normal in Obstructive Sleep Apnea (OSA) despite subclinical myocardial injury, limiting early risk stratification. This study aimed to e...
BACKGROUND: Conventional cardiac magnetic resonance imaging(CMR) metrics may remain normal in Obstructive Sleep Apnea (OSA) despite subclinical myocardial injury, limiting early risk stratification. This study aimed to evaluate whether biventricular function, mechanics, and tissue characteristics assessed by non-contrast CMR differ between patients with severe and non-severe OSA, and to explore their association with markers of nocturnal hypoxemia. METHODS: Seventy-five newly diagnosed OSA patients(62 male; age 43.7 ± 9.3 years) were prospectively included and underwent polysomnography followed by non-contrast CMR, including cine imaging, native T1 and T2 mapping. Patients were stratified by apnea-hypopnea index (AHI) into non-severe (n = 21, 4 mild and 17 moderate cases) and severe (n = 54) groups. Biventricular parameters were compared and correlated with AHI and the oxygen desaturation index (ODI). RESULTS: Compared with non-severe OSA group, severe OSA group exhibited higher left ventricular mass (LVM) and lower left ventricular global circumferential and radial strain (all p < 0.05). Right ventricular (RV) dysfunction was more pronounced, with significantly lower in global longitudinal, circumferential, and radial strain and strain rates (all p < 0.05). The RV end-systolic remodeling index (RVESRI) was higher in severe OSA (p = 0.002). The right ventricular blood pool T2 value (RVT2) was significantly lower in severe OSA than in the non-severe OSA group and moderately correlated with AHI (ρ=-0.435) and ODI (ρ=-0.425). Collectively, multiple CMR parameters showed weak‑to‑moderate correlations with AHI and ODI (ρ ranging from - 0.301 to 0.476), indicating that OSA severity is associated with a broad spectrum of subclinical biventricular alterations rather than a single dominant abnormality. CONCLUSION: Severe OSA is associated with subclinical biventricular dysfunction, disproportionately affecting the RV. Elevated RVESRI may indicate early systolic maladaptation, while lower RVT2 shows promise as a non-invasive imaging marker associated with hypoxemic burden, warranting further investigation. Non-contrast CMR enables detection of cardiac injury, offering valuable potential for risk stratification in OSA patients.
Deep learning-based cervical cancer T-staging using MRI: multi-structure segmentation and classification.
Xu S, Zou Y, Wu Z … +5 more , Xu Z, Wang R, Hou W, Wang Y, Wu Y
BMC Med Imaging · 2026 May · PMID 42218381 · Full text
AIMS: Cervical cancer has high incidence and mortality, seriously threatening women's survival and quality of life. Radiologists currently rely mainly on subjective clinical experience for cervical cancer T-staging, whic...
AIMS: Cervical cancer has high incidence and mortality, seriously threatening women's survival and quality of life. Radiologists currently rely mainly on subjective clinical experience for cervical cancer T-staging, which easily leads to misdiagnosis. OBJECTIVES: To develop a deep learning-based technique for automatic segmentation and T-staging of cervical cancer to improve clinical diagnostic accuracy and efficiency. MATERIALS AND METHODS: A dataset of 17,479 fT1WI MRI scans from 144 patients (T1-T4 stages) was constructed; tumors and adjacent structures were manually outlined with Amira 2019 to generate ground truth (GT). A novel segmentation network (CPANet) was designed by integrating global pyramid guidance (GPG) and atrous spatial pyramid pooling (ASPP) modules into CNN. CPANet's segmentation performance was validated against GT and compared with UNet, UNet++, DeepLabv3+, and UperNet-Swin. Based on CPANet-extracted ROIs, T-staging models were built using ResNet50, DenseNet121, and Swin Transformer (pathological T-staging as GT), and the optimal model was selected. RESULTS: CPANet outperformed other networks, with Dice similarity coefficients (DSCs) of 0.783 (tumor), 0.901 (uterus), 0.909 (bladder), and 0.892 (rectum), and average per-case processing time of 1.60 s. Swin Transformer achieved the best T-staging performance: AUCs of 0.713 (T1), 0.799 (T2), 0.845 (T3-T4) for main stages, and 0.623 (T1b), 0.673 (T2a), 0.897 (T2b) for sub-stages from MRI images. This not only improves diagnostic accuracy and efficiency but also conserves medical resources, thereby facilitating the establishment of intelligent healthcare systems in medically underserved areas. CONCLUSIONS: CPANet and Swin Transformer enable accurate automatic segmentation and T-staging of cervical cancer from MRI images, improving diagnostic accuracy and efficiency, saving medical resources, and facilitating intelligent healthcare in underserved areas.
Peripheral instability gradient in macular thickness measurements across five optical coherence tomography systems: a prospective cross-sectional study of inter-device agreement and clinical monitoring risk.
Akkan F, Gorgun E
BMC Med Imaging · 2026 May · PMID 42218369 · Full text
BACKGROUND: Cross-platform variability in optical coherence tomography (OCT) macular thickness measurements poses a clinically meaningful challenge for longitudinal patient assessment, particularly when patients are imag...
BACKGROUND: Cross-platform variability in optical coherence tomography (OCT) macular thickness measurements poses a clinically meaningful challenge for longitudinal patient assessment, particularly when patients are imaged on different devices over time. This study characterizes intra-device repeatability and inter-device agreement across five contemporary OCT systems, and introduces the Peripheral Instability Gradient (PIG)-the progressive deterioration in cross-device measurement concordance from the foveal center outward across ETDRS-defined macular zones. METHODS: In this prospective, cross-sectional study, 38 healthy adults underwent macular imaging with five OCT devices: HRA Spectralis (Heidelberg Engineering), BMIZAR (TowardPi, 400 kHz swept-source OCTA), TOPCON Maestro 2, NIDEK RS-1 Glauvas, and HUVITZ HOCT-1/1F. Macular thickness was measured in the central 1-mm ETDRS subfield and the 3-mm and 6-mm inner rings (superior, nasal, inferior, and temporal sectors). Three consecutive scans were acquired per device during a single visit by a single experienced operator. Intra-device repeatability was assessed using ICC(3,1) and repeated-measures ANOVA with Bonferroni correction. Inter-device agreement was evaluated using ICC(2,1), Bland-Altman analysis, and absolute percentage error (APE). HRA-anchored comparisons were used for directional bias interpretation, and all 10 direct pairwise device comparisons were additionally performed as cross-validation. RESULTS: All devices demonstrated excellent within-device repeatability (ICC > 0.85 across all regions). Inter-device agreement was strongly location-dependent, with PIG: ICC values declining from the central 1-mm subfield (HRA vs. HUVITZ: 0.942; HRA vs. NIDEK: 0.902) to the 6-mm ring (ICC < 0.40 for most pairs). This represents a relative decline of 40-67%. Central thickness ranged from 227.85 μm (TOPCON) to 290.73 μm (BMIZAR). Device-specific systematic biases were identified: BMIZAR overestimated HRA by a mean relative bias of 10.98%, and TOPCON underestimated HRA by a mean relative bias 13.65%, while HUVITZ (+ 2.60%) and NIDEK (+ 3.64%) showed smaller relative deviations (all biases as HRA minus comparator). Limits of agreement widened by a relative 30-50% from central to peripheral rings. Median APE was lowest for HUVITZ (2.39%) and highest for TOPCON (12.23%). CONCLUSIONS: Macular thickness measurements are highly repeatable within individual OCT systems but are not interchangeable across devices, with disagreement amplifying progressively toward peripheral macular zones. The Peripheral Instability Gradient demonstrates that cross-device follow-up carries the greatest risk in parafoveal and perifoveal monitoring. To ensure reliable longitudinal assessment, clinicians should maintain device consistency; when switching is unavoidable, device-specific baselines should be established and calibration adjustments applied.
MSTM-Net: a two-stage prostate cancer segmentation network based on swin-transformer-mamba architecture.
Chen J, Liu X, Wang S … +1 more , Ji W
BMC Med Imaging · 2026 May · PMID 42215932 · Full text
BACKGROUND: Magnetic resonance imaging (MRI) has become a core imaging modality for prostate cancer screening and diagnosis. Accurate and automatic segmentation of lesion regions is critical for subsequent staging assess...
BACKGROUND: Magnetic resonance imaging (MRI) has become a core imaging modality for prostate cancer screening and diagnosis. Accurate and automatic segmentation of lesion regions is critical for subsequent staging assessment and treatment planning. METHODS: To this end, this research proposes a two-stage segmentation framework for multimodal MRI. In the first stage, the prostate gland is segmented to extract the region of interest (ROI), thereby removing complex pelvic background structures. In the second stage, fine-grained prostate cancer lesion segmentation is performed within the ROI, enabling the model to focus on anatomically plausible lesion regions.A segmentation network, termed MSTM-Net, is developed based on this framework. The network adopts a Swin Transformer-based decoder architecture. At the input stage, T2-weighted images and apparent diffusion coefficient (ADC) maps are spatially aligned and concatenated along the channel dimension. During decoding, a Mamba module based on state-space modeling is introduced to jointly capture local structural information and long-range dependencies. Multi-head attention fusion and multi-scale feature fusion are further integrated into the skip connections to enhance the consistency between shallow spatial details and deep semantic representations. RESULTS: Experiments conducted on the cleaned PROSTATEx dataset demonstrate that the proposed method achieves a Dice score of 95.38% for prostate gland segmentation and 63.89% for lesion segmentation, outperforming the best comparative network by approximately 4% points, with an mIoU of 61.32%. Furthermore, cross-dataset validation on the PI-CAI dataset yields a Dice score of 63.14%, indicating good generalization ability and clinical feasibility for automated prostate cancer segmentation. CONCLUSION: The proposed MSTM-Net demonstrates effective performance for prostate cancer segmentation in multimodal MRI, achieving improved accuracy and feature representation compared with existing methods. The results indicate that the two-stage framework combined with multi-modal fusion and state-space modeling is a promising approach, although further validation on larger and more diverse datasets is required to enhance robustness and generalization.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe