Frontiers in radiology最新文献

This paper introduces an Ultra-Lightweight Uncertainty-Aware Ensemble (UALE) model for large-scale multi-class medical MRI diagnosis, evaluated on the 2024 Benchmark Diagnostic MRI and Medical Imaging Dataset containing 40 classes and 33,616 images. The model integrates five specialized micro-expert networks, each designed to capture distinct MRI features, and combines them using a confidence-weighted ensemble mechanism enhanced with variance-based uncertainty quantification for robust, reliable predictions. With only 0.05M parameters and 0.18 GFLOPs, UALE achieves high efficiency and competitive performance among ultra-lightweight models with an accuracy of 69.1% and an F1 score of 68.3%. Besides lightweight models, the paper offers an extensive analysis and performance comparison with fifteen state-of-the-art models, discusses various datasets, elaborates on uncertainty estimates pertaining to the clinical trustworthiness of the models and possible clinical deployment, and highlights trade-offs and avenues for future work in economically constrained settings. The extreme compactness and reliability of the UALE affords it unique utility in scalable medical diagnostics suitable for low-resource clinical settings and portable imaging devices, such as rural hospitals.

Objective: This study aims to evaluate the diagnostic performance of diffusion-weighted imaging (DWI) with a fractional order calculus (FROC) model for differentiating breast lesions and to explore the associations between FROC/apparent diffusion coefficient (ADC)-derived diffusion metrics and prognostic biomarkers and molecular subtypes in breast cancer.

Methods: This retrospective study included 147 patients with 159 histopathology-confirmed lesions who underwent multi-b DWI using simultaneous multi-slice (SMS) readout-segmented echo-planar imaging (rs-EPI) at 3.0 T. Whole-lesion histograms were computed for mono-exponential ADC and FROC parameters (D, β, μ). The Mann-Whitney U test was used to compare the histogram metrics of each diffusion parameter between the benign and malignant groups and between groups with different prognostic biomarkers and molecular subtypes. The Kruskal-Wallis test was used to compare the histogram metrics of each DWI-derived parameter among the different molecular subtypes. The Spearman rank correlation analysis was employed to characterize correlations between diffusion parameters and prognostic biomarkers. The diagnostic performance of each DWI-derived parameter in differentiating breast lesions was assessed using receiver operating characteristic (ROC) analysis.

Results: Interobserver reproducibility was excellent (intra-class correlation coefficient 0.827-0.928). Central tendency histogram metrics (10th, 90th percentiles, mean, median) of ADC and FROC parameters were higher in benign than malignant lesions, whereas skewness (all models) and entropy/kurtosis (ADC, D, μ) were lower in benign lesions (all p < 0.05, except β-skewness). The histogram metrics of ADC-median, D_FROC-mean, and D_FROC-median showed similar diagnostic performance. The values of ADC-mean, D_FROC-10%, D_FROC-mean, D_FROC-median, β_FROC-10%, β_FROC-mean, and β_FROC-median were significantly lower in the estrogen receptor (ER)-positive group compared with those in the ER-negative group. The tumors with progesterone receptor (PR)-negative status showed significantly higher β_FROC-10%, β_FROC-mean, and β_FROC-median values than those of tumors with PR-positive status. The values of D_FROC-skewness, β_FROC-10%, and β_FROC-mean exhibited significant differences in differentiating the triple-negative and luminal subtypes.

Conclusions: FROC-based histogram analysis yields diagnostic performance comparable to ADC for benign vs. malignant classification, while providing richer associations with ER/PR status, proliferation, and nodal involvement, reflecting microstructural heterogeneity not captured by mono-exponential diffusion.

Low-dose Computed Tomography (CT) imaging minimizes radiation exposure but often results in degraded image quality, making diagnosis challenging. Image Quality Assessment (IQA) is a process of quantitatively evaluating the visual quality of images and plays a crucial role in determining whether these CT scans meet the necessary standards for accurate diagnosis. IQA methods help identify issues such as noise, blurriness, or artifacts that may compromise the diagnostic value of the scans. Traditional quality assessment measures how closely an image matches an ideal or reference image. Since obtaining a high-quality reference image is often challenging, an automated quality assessment framework (diagnosis based IQA) using No-Reference Image Quality Assessment (NRIQA) techniques is proposed, allowing quality evaluation and eliminating the need for a high-quality reference image. In this approach, various statistical and structural features are extracted from low-dose CT scans and mapped to radiologist-assigned quality scores, which are subjective evaluations given by experts to train and compare various predictive models. The framework undergoes 100-fold validation, to ensure the reliability of the proposed model. CT images with predicted quality scores of 2 and below undergo spatial domain enhancement to improve their diagnostic value. These enhanced images are then reassessed using the diagnosis based IQA (trained Support Vector Regression) model, demonstrating an improvement in predicted quality scores. In addition, the enhanced images were verified by a radiologist, confirming the effectiveness of the enhancement process. This two-stage approach, automated NRIQA-based quality prediction and selective enhancement provides a reliable, and objective method for assessing and improving low-dose CT image quality.

Objective: To evaluate the efficacy of flow diverter implantation for treating CTA-negative giant vertebral artery dissection aneurysm (VADA) and to address the challenges in lesion characterization using MRI.

Methods: A 66-year-old male patient presented with a 3-month history of left facial numbness and dysarthria. Initial MRI-T1 revealed a mixed signal intensity lesion in the CPA region. However, both CTA and digital subtraction angiography (DSA) failed to identify any significant vascular abnormalities. Subsequently, CT-perfusion and dynamic contrast-enhanced computed tomography (DCE-CT) were performed to further characterize the lesion.

Results: DCE-CT revealed a giant VADA, which was significantly larger than the lesion initially detected by MRI and was identified as the cause of hypoperfusion in the posterior circulation. Based on these findings, a flow diverter implantation procedure was performed successfully without complications. Angiographic follow-up at 8 months demonstrated no recurrence of the lesion. At the 14-month clinical follow-up, the patient exhibited complete resolution of symptoms, with a mRS score of 0, indicating an excellent functional outcome.

Conclusion: Flow diverter implantation may be an effective treatment for CTA-negative giant VADAs. The limitations of MRI in accurately characterizing lesion size underscore the necessity of advanced imaging techniques, such as DCE-CT, for precise device selection and deployment.

Objective: To develop a radiomics-based predictive model for capsular invasion in thymomas by applying machine learning algorithms to non-contrast and contrast-enhanced CT imaging. This study aimed to assess the influence of intratumoural and peritumoural regions on capsular invasion prediction and to compare the performance of models derived from these regions within the same dataset, thereby identifying the optimal predictive model.

Methods: Clinical and imaging data were retrospectively collected from 151 patients with thymoma who underwent treatment at Tianjin Chest Hospital between June 2018 and January 2025. Based on pathological findings, patients were categorised into capsular invasion and non-invasion groups and subsequently randomised into a training set (n = 106) and a test set (n = 45) in a 7:3 ratio. Radiomic feature selection was performed using univariate logistic regression analysis followed by least absolute shrinkage and selection operator (LASSO) regression. Predictive models were developed employing multiple machine learning algorithms, including logistic regression. Model performance was evaluated through receiver operating characteristic (ROC) curve analysis, with sensitivity, specificity, F1 score, and decision curve analysis (DCA) used to assess diagnostic accuracy and clinical applicability. DeLong's test was applied to compare the area under the curve (AUC) values between different models. Calibration curves were generated to evaluate model calibration, and model interpretability was examined using the Shapley Additive exPlanations (SHAP) method.

Results: Comparative analysis of machine learning methods across different tumour regions revealed that the support vector machine (SVM) model, developed using radiomic features from the 4 mm peritumoural region on contrast-enhanced CT scans, demonstrated optimal predictive performance. This model achieved area under the curve (AUC) values of 0.890 [95% confidence interval (CI): 0.823-0.956] in the training cohort and 0.888 (95% CI: 0.792-0.983) in the test cohort.

Conclusion: CT-based radiomics demonstrates efficacy in predicting capsular invasion in thymomas, with the peritumoural region proving particularly significant. This methodology shows potential for supporting clinicians in preoperative treatment strategy formulation.

Background: The optimal management approach for ruptured intracranial aneurysms remains debated, with limited real-world evidence from Latin American populations. This study compared in-hospital outcomes between endovascular coiling and surgical clipping using a hierarchical win ratio (WR) analysis.

Methods: We conducted a single-center retrospective cohort study of 194 patients with ruptured intracranial aneurysms treated at a tertiary referral center (2011-2022). Patients were treated with either endovascular coiling (n = 73) or surgical clipping (n = 121). The primary outcome was the win ratio, analyzing a hierarchical composite endpoint of: (1) in-hospital mortality, (2) unfavorable functional outcome at discharge (modified Rankin Scale >2), (3) major complications, and (4) prolonged ICU stay (>10 days). Secondary analyses included multivariable logistic regression and prespecified subgroup analyses by clinical severity and aneurysm location.

Results: Baseline measured characteristics were balanced between groups. The win ratio significantly favored endovascular coiling (WR 1.75, 95% CI: 1.67-1.84, p < 0.001), indicating 75% more wins in the hierarchical outcome comparison. All individual components significantly favored coiling: mortality (WR = 1.35, p < 0.001), unfavorable functional outcome (WR = 1.53, p < 0.001), major complications (WR = 1.70, p < 0.001), and prolonged ICU stay (WR = 1.25, p < 0.001). Benefits were consistent across subgroups, including Hunt & Hess grades I-II (WR = 2.00) and III-V (WR = 1.96), and across most aneurysm locations. In contrast, multivariate logistic regression for poor outcome showed a favorable but non-significant trend for coiling (OR = 0.55, p = 0.102), while confirming Hunt & Hess ≥3 (OR = 5.54, p < 0.001) and modified Fisher ≥3 (OR = 3.85, p = 0.044) as dominant prognostic factors.

Conclusion: In this Colombian cohort, hierarchical outcome analysis suggested superior in-hospital outcomes for endovascular coiling vs. surgical clipping. However, the substantial attenuation of this association in adjusted analyses indicates that these apparent advantages may largely reflect case selection patterns rather than inherent treatment superiority, as residual confounding by aneurysm complexity cannot be excluded.

Background: The accurate separation of lung parenchyma, ground-glass opacity (GGO), and intrapulmonary vessels on high-resolution computed tomography (HRCT) in coronavirus disease 2019 (COVID-19) is challenging.

Methods: We conducted a cross-sectional study that analyzed 530 adults (20-40 years) with RT-PCR-confirmed COVID-19. For texture modeling, we sampled 597 regions of interest (ROIs) representing parenchyma, GGO, and intrapulmonary vessels. Region-of-interest-labeled HRCT patches representing parenchyma, GGO, and vessels were analyzed using first- and second-order texture features that were computed across different square window sizes (5 × 5-20 × 20 pixels). Feature selection with stepwise linear discriminant analysis yielded a three-class classifier. The primary endpoint was overall classification accuracy, with the secondary endpoints including the effect of window size and identification of the most informative features.

Results: The 20 × 20-pixel window produced the highest performance, with an overall accuracy of 88.6%. Five co-occurrence-based features (average difference, inverse difference moment, co-occurrence matrix standard deviation, sum entropy, and information correlation measure 1) were the most discriminative; the majority of the errors occurred at tissue boundaries where patches spanned mixed voxels.

Conclusion: Texture-based feature extraction achieved 88.6% ROI-level accuracy and can serve as a supplementary tool during radiological interpretation of chest CT.

Introduction: Early and accurate detection of viral diseases is vital for timely treatment and public health preparedness. However, most existing computer-based prediction systems depend on centralized data storage, which raises concerns about patient privacy, compatibility between different hospitals, and limited clarity on how predictions are made. To address these issues, this study introduces U-FDL-PPE, a new federated deep-learning framework designed to support early and reliable viral disease diagnosis while protecting patient confidentiality and offering clear and understandable prediction insights.

Methods: The framework uses a decentralized learning approach that allows hospitals to train models collaboratively without exchanging raw medical images. MobileNetV2 was used as the core model for classifying chest X-rays, and Grad-CAM was included to produce heatmaps that visually explain how the model arrived at its decisions. The system was tested using the publicly available COVID-19 Radiography Database in a simulated network of three healthcare institutions. Model performance was evaluated using standard measures such as accuracy, F1-score, AUC, and confusion matrix.

Results: Across five training rounds, U-FDL-PPE recorded 88% accuracy, an F1-score of 89.66%, and a multi-class AUC of 0.5192. The confusion matrix showed consistently correct predictions across the three diagnostic categories: COVID-19, Normal, and Viral Pneumonia. The Grad-CAM heatmaps highlighted medically relevant lung regions, confirming that the framework focused on features that clinicians would expect when diagnosing these conditions.

Discussion: The results indicate that U-FDL-PPE is a practical and scalable solution for early viral disease diagnosis, particularly in environments where patient privacy must be preserved. Its combination of decentralized training and visual explanation builds greater trust among clinicians while ensuring that sensitive medical data never leaves the originating institution. The lightweight MobileNetV2 architecture also supports faster processing, making the system suitable for hospitals and clinics with limited computing resources. Overall, U-FDL-PPE provides a privacy-conscious and transparent diagnostic framework that is well-positioned for real-world implementation across healthcare networks.

Introduction: Understanding the role of various brain regions of interest (ROIs) in various cognitive functions or tasks, across healthy or neurodegenerative conditions and multiple degrees of separation, remains a key challenge in neuroscience. Conventional network measures can only capture localized or quasi-localized features of brain ROIs. Topological data analysis (TDA), particularly persistent homology, provides a threshold-free, mathematically rigorous framework for identifying topologically salient features in complex networks. In this paper, we introduce a new metric, the Homological Vertex Importance Profile (H-VIP), designed to assess the relevance of vertices that participate in persistent topological structures (e.g., connected components, cycles or cavities) in brain networks. The H-VIP quantifies the topological features of the network at the ROI (node) level by compressing its higher-order connectivity profile using homological constructs.

Methods: Leveraging homological constructs of brain connectomes, we extend two of our previously defined network-level measures-average persistence and persistence entropy-to an ROI-level measure, i.e., the H-VIP. We then applied the H-VIP to two independent datasets: structural connectomes from the Human Connectome Project and functional connectomes from the Alzheimer's Disease Neuroimaging Initiative. Persistent homology was computed for each network, and H-VIP scores were derived to evaluate vertex-level contributions. Finally, H-VIP scores were used for the prediction of multiple cognitive measures.

Results: In both anatomical and functional brain networks, H-VIP values demonstrate predictive power for various cognitive measures. Notably, the connectivity of the frontal lobe exhibited stronger correlations with cognitive performance than the whole-brain network.

Discussion: H-VIP offers a robust and interpretable means to locate, quantify, and visualize region-specific contributions to network's topological, higher-order landscape. Its ability to detect potentially impaired connectivity at the individual level suggests possible applications in personalized medicine for neurological diseases and disorders. Beyond brain connectomics, the H-VIP can be used for other types of complex networks where topological features are of importance, such as financial, social, or ecological networks.

Introduction: Motion artifacts induced by atrial fibrillation (AF) present a substantial challenge in coronary computed tomography angiography (CCTA). Wide detectors, rapid scanning, and motion correction algorithms can effectively improve image quality in CCTA. This study aims to evaluate the impact of one-beat acquisition with a motion correction algorithm (Snapshot Freeze 1, SSF1) on the image quality of prospective CCTA in patients with AF, and its diagnostic performance using an artificial intelligence assisted diagnostic system (AI-ADS).

Materials and methods: A total of 91 consecutive patients with AF, who underwent one-beat CCTA were analyzed. Images were reconstructed with SSF1. The subjective and objective image quality of the coronary arteries were evaluated. Using the invasive coronary catheter angiography (ICA) as the reference standard, the diagnostic performance of AI-ADS and AI-ADS + radiologist for stenoses above moderate and severe degrees were calculated.

Results: Effective radiation dose was 2.43 ± 0.88 mSv. The average CT values of all major coronary arteries and branches were greater than 400 HU. All vessels were diagnosable (scores ≥ 3) with good or above ratings at 96.15% (350/364) and 96.70% (352/364). The diagnostic accuracy, sensitivity, specificity and AUC of AI-ADS vs. AI-ADS + radiologist for above moderate stenoses were: (84.62% vs. 91.21%), (89.61% vs. 98.70%), (57.14% vs. 50.00%) and (0.73 vs. 0.74) on patient level; (84.07% vs. 87.64%), (74.07% vs. 85.19%), (89.96% vs. 89.08%) and (0.82 vs. 0.87) on vessel level; (90.84% vs. 93.11%), (63.59% vs. 78.34%), (95.99% vs. 95.91%) and (0.80 vs. 0.87) on segment level. For severe stenoses, these values were: (62.64% vs. 82.42%), (58.62% vs. 91.38%), (69.70% vs. 66.67%) and (0.64 vs. 0.79) on patient level; (82.97% vs. 89.29%), (46.43% vs. 75.00%), (93.93% vs. 93.57%) and (0.70 vs. 0.84) on vessel level; (92.23% vs. 95.16%), (36.92% vs. 66.92%), (98.06% vs. 98.14%) and (0.68 vs. 0.83) on segment level.

Conclusion: One-beat CCTA with SSF1 provides high-quality coronary images for patients with AF. AI-ADS automatically distinguishes coronary images with different stenoses, but the sensitivity of AI-ADS is low, especially for severe stenoses. AI-ADS + radiologist further improves the diagnostic performance.