European Radiology最新文献

Objectives: Plain abdominal radiography is a widely used imaging modality for diagnosing neonatal necrotizing enterocolitis (NEC), but the characteristic features of stage I NEC are often subtle, making early diagnosis challenging. This study explores the application of deep learning (DL) models to assist in the early diagnosis of stage I NEC.

Materials and methods: This retrospective study included 380 and 300 neonates who underwent abdominal radiography at two centers between June 2016 and December 2023. Neonates were grouped based on a diagnosis of stage I NEC. DL features were extracted from the radiographs using the DenseNet121 model, based on which radiomics models were constructed using logistic regression (LR) and random forest (RF) algorithms. Performance was evaluated through receiver operating characteristic (ROC) curves. Both the training and external validation cohorts were used to assess model accuracy in distinguishing stage I NEC. Additionally, a direct comparison with human expert diagnostic performance was conducted.

Results: In the training cohort, 25 DL features were selected for model development. The area under the ROC curve (AUC) for LR and RF models was 0.972 (95% CI: 0.956-0.988) and 0.961 (95% CI: 0.942-0.980), respectively. In the external validation cohort, the models demonstrated AUCs of 0.964 (95% CI: 0.943-0.986) and 0.951 (95% CI: 0.925-0.976), respectively. These models evidently outperformed human experts in diagnostic performance.

Conclusion: The DL model based on plain abdominal radiography effectively identified stage I NEC in neonates. This approach offers a non-invasive method to enhance early NEC diagnosis and support clinical decision-making.

Key points: QuestionDeep learning (DL) models applied to plain abdominal radiography can enhance the early diagnosis of stage I neonatal necrotizing enterocolitis (NEC). FindingsIn this retrospective study involving 680 neonates from two centers, DL-based radiomics models achieved much higher accuracy for diagnosing stage I NEC than human radiologists. Clinical relevanceDL models based on plain abdominal radiography have the ability to significantly improve the early identification of stage I NEC, offering a non-invasive tool to support radiologists in early diagnosis.

Objectives: Studies have reported promising results regarding artificial intelligence (AI) as a tool for improved mammographic screening interpretive performance. We analyzed AI malignancy risk scores from two versions of the same commercial AI model.

Materials and methods: This retrospective cohort study used data from 117,709 screening examinations performed in BreastScreen Norway 2009-2018. The mammograms were processed by two versions of the commercially available AI model, Transpara (version 1.7 and 2.1). The distributions of exam-level risk scores (AI score 1-10) and risk categories were evaluated for both AI versions on all examinations, including 737 screen-detected and 200 interval cancers. Scores between 1-7 were categorized as low risk, 8-9 as intermediate risk, and 10 as high risk of malignancy.

Results: Area under the receiver operating curve was 0.908 (95% CI: 0.986-0.920) for version 1.7 and 0.928 (95% CI: 0.917-0.939) for 2.1 when screen-detected and interval cancers were considered as positive cases (p < 0.001). A total of 87.1% (642/737) and 93.5% (689/737) of the screen-detected cancers had an AI score of 10 with version 1.7 and 2.1, respectively. Among interval cancers, 45.0% (90/200) had AI score 10 with version 1.7 and 44.5% (89/200) had AI score 10 with version 2.1.

Conclusion: A higher proportion of screen-detected breast cancers had the highest AI score of 10 with the newer version of the AI model compared to the older version. For interval cancers, there was no difference in the proportion of cases assigned to the highest score between the two versions.

Key points: Question Studies have reported promising results regarding the use of AI in mammography screening, but comparisons of updated versus older versions are less studied. Findings In our study, 87.1% (642/737) of the screen-detected cancers were classified with a high malignancy risk score by the old version, while it was 93.5% (689/737) for the newer version. Clinical relevance Understanding how version updates of AI models might impact screening mammography performance will be important for future quality assurance and validation of AI models.

Objectives: To develop and validate a deep learning (DL)-based algorithm for automated measurement of femoral head ossification center (FHOC) size and establish AI-derived growth charts.

Materials and methods: This retrospective study included 1705 healthy Korean children (mean age, 5.1 ± 3.3 years; 841 females, 864 males) with anteroposterior pelvic radiographs (2018-2024). A three-stage DL algorithm (region-of-interest detection, FHOC segmentation, landmark-based size computation) was used to automatically measure FHOC size. Agreement with radiologist measurements was evaluated using concordance correlation coefficient (CCC), Pearson correlation coefficient (r), mean absolute error (MAE), root mean square error (RMSE), and Bland-Altman analyses, supplemented by paired t-test and Fisher's Z-test. AI measurements were used to create FHOC growth charts via quantile polynomial regression, with predictive accuracy assessed by adjusted R², MAE, and RMSE.

Results: AI-derived FHOC size measurements showed close agreement with radiologist measurements, with mean differences within ±0.5 mm and 95% limits of agreement within ±3 mm in age-stratified analyses, and overall agreement was further supported by high CCC, r, and consistently low error metrics. Growth curves based on AI measurements demonstrated strong predictive accuracy (adjusted R² = 0.927 for females; 0.934 for males), with low errors across age groups (females: MAE 1.77-2.98 mm, RMSE 2.28-3.54 mm; males: MAE 1.60-3.01 mm, RMSE 2.00-4.10 mm). Reference percentiles (5th-95th) were established, providing standardized FHOC size ranges for clinical application.

Conclusion: Our DL-based approach provides precise and reproducible FHOC size measurement, offering a robust reference for standardized growth assessment and early pediatric hip joint evaluation.

Key points: QuestionThe timing of FHOC appearance is an important radiographic indicator; however, manual measurement is subjective, and studies on age-specific changes remain limited. FindingsA DL-based algorithm achieved high agreement with expert measurements, and age-based regression reliably predicted FHOC size in children. Clinical relevanceAI-derived FHOC growth charts may provide objective, standardized references for pediatric hip joint development, potentially enabling earlier detection of growth abnormalities and improving diagnostic consistency in clinical practice.

Objective: To assess moderate-dose statin effects on intracranial plaques with < 70% stenosis using high-resolution vessel wall imaging (HR-VWI).

Materials and methods: This retrospective study enrolled patients with intracranial atherosclerosis and initiating statin therapy. HR-VWI was used to evaluate the therapeutic effects of statins on two subgroups: the culprit plaques and significantly enhanced plaques. Plaque characteristics and their change rates before and after treatment were analyzed for each subgroup. Multiple linear regression analysis was used to evaluate the relationship between the effectiveness of statin therapy and baseline plaque characteristics.

Results: Fifty-two patients (age 60.4 ± 11.3 years, 40 males and 12 females) with 138 plaques were enrolled. Maximum plaque thickness (p = 0.007), plaque burden (p = 0.003), luminal stenosis (p < 0.001), and plaque enhancement (p < 0.001) were significantly reduced in culprit plaques. Culprit plaques showed greater reductions in plaque burden (p = 0.009) and luminal stenosis (p = 0.028) change rates compared to non-culprit plaques. There were significant reductions in maximum plaque length (p = 0.044), maximum plaque thickness (p = 0.036), luminal stenosis (p = 0.047), and plaque enhancement (p < 0.001) in significantly enhanced plaques. The significantly enhanced plaques also demonstrated a more pronounced decrease in enhancement change rate (p = 0.037) compared to mildly enhanced plaques. Higher plaque burden (%, B = -0.28, S.E = 0.13, p = 0.036) and plaque enhancement degree (%, B = -0.16, S.E = 0.07, p = 0.015) were positively correlated with statin treatment on plaque enhancement change rate.

Conclusion: Compared to non-culprit plaques and plaques with mild enhancement, moderate-dose statins demonstrate superior therapeutic efficacy for culprit plaques and significantly enhanced plaques with stenosis < 70%.

Key points: Question To assess the effects of moderate-dose statin therapy on intracranial atherosclerotic plaques with mild to moderate stenosis using high-resolution vessel wall imaging. Findings Compared to non-culprit plaques and plaques with mild enhancement, moderate-dose statins demonstrate superior therapeutic efficacy for culprit plaques and significantly enhanced plaques with stenosis < 70%. Clinical relevance Early statin intervention can stabilize and regress high-risk intracranial plaques, reduce luminal narrowing, and provide effective secondary prevention for stroke and transient ischemic attack in patients with < 70% stenosis.

Objectives: To assess whether phase-resolved functional lung MRI (PREFUL-MRI) can differentiate pulmonary arterial hypertension (PAH) from chronic thromboembolic pulmonary hypertension (CTEPH) and to compare PREFUL-derived perfusion and ventilation characteristics between the two entities.

Materials and methods: In this single-center study (January 2024-March 2025), patients with pulmonary hypertension (PH) who underwent PREFUL-MRI, ventilation/perfusion single-photon emission CT/CT (V/Q SPECT/CT), and right-heart catheterization within 1 week were retrospectively enrolled. PREFUL-MRI was acquired with a free-breathing fast spoiled gradient echo sequence to generate perfusion and ventilation maps. PREFUL maps and V/Q SPECT/CT were assessed by four blinded readers. Quantitative PREFUL parameters were compared between PAH and CTEPH, and correlations with hemodynamics were analyzed.

Results: Fifty-three PH patients were included (PAH 26, CTEPH 27). All manifested multiple perfusion defects on PREFUL maps. Visual assessment with PREFUL yielded 70% sensitivity and 38% specificity for CTEPH, lower than V/Q SPECT/CT (sensitivity 95%, specificity 95%). CTEPH had higher perfusion defect percentage and lower mean perfusion than PAH (median QDP_exclusive 51.9% vs. 24.4%, p < 0.001; QDP_total 58.4% vs. 30.85%, p < 0.001; mean perfusion 2.8% vs. 3.9%, p = 0.003). QDP_exclusive represents areas with perfusion defects but without ventilation defects, whereas QDP_total indicates the overall perfusion defect percentage. Across the cohort, QDP_exclusive correlated positively with pulmonary vascular resistance (PVR) (ρ = 0.313, p = 0.031).

Conclusions: PREFUL-MRI did not reliably distinguish PAH from CTEPH on visual assessment, although CTEPH exhibited a larger perfusion defect burden. It may be more suitable for functional evaluation of PH rather than for initial differential diagnosis.

Key points: QuestionCan phase-resolved functional lung MRI (PREFUL-MRI) noninvasively distinguish pulmonary arterial hypertension (PAH) from chronic thromboembolic pulmonary hypertension (CTEPH), and reflect their hemodynamic severity? FindingsPREFUL-MRI showed limited visual discrimination compared to V/Q SPECT/CT; CTEPH exhibited more severe perfusion defects; perfusion defect percentage correlated with pulmonary vascular resistance. Clinical relevancePREFUL-MRI provides radiation- and contrast-free perfusion metrics reflecting hemodynamic burden in pulmonary hypertension. Though suboptimal for initial discrimination in disease subtypes, it supports functional evaluation once the diagnosis is established.

Objectives: High-density areas (HDAs) are frequently observed in follow-up CT of large-vessel occlusions after endovascular therapy. Utilizing the established ASPECTS regions, incorporating the subarachnoid space and ventricles, we developed a novel HDA score to evaluate its correlation and predictive value for hemorrhagic transformations and clinical outcomes.

Materials and methods: This retrospective, multicenter study included consecutive patients who had HDA on follow-up CT after endovascular therapy. Multivariable logistic regression and area under the receiver operating characteristic curve (AUC) analyses assessed the associations and predictive value of HDA location and score with hemorrhagic transformations and unfavorable clinical outcomes.

Results: Among the 1130 consecutive patients treated with endovascular therapy, 542 patients (326 males; median age 70 years) had HDA were finally included. Multivariable logistic regression showed that HDA location in the lentiform nucleus (OR, 1.6; 95% CI: 1.1-2.5; p = 0.03) and ventricles (OR, 4.4; 95% CI: 1.2-16.6; p = 0.03) was associated with hemorrhagic transformations, whereas location in the lentiform nucleus (OR, 2.3; 95% CI: 1.4-3.8; p < 0.001), M1(OR, 3.9; 95% CI: 1.7-8.9; p = 0.001), and subarachnoid space (OR, 2.1; 95% CI: 1.2-3.8; p = 0.02) was associated with unfavorable clinical outcomes, as was the HDA score (OR = 1.35, 95% CI:1.19-1.54, p < 0.01). Including HDA indicators in the model significantly improved its unfavorable clinical outcome predictive power (AUC increased from 0.743 to 0.802; DeLong test; p < 0.01).

Conclusion: The HDA score, which reflects the number of HDA involved regions, significantly correlated with unfavorable clinical outcomes and effectively predicted the prognosis.

Key points: QuestionThe relationship between the location and extent of HDA on postoperative CT and hemorrhagic transformation and clinical outcomes in patients undergoing endovascular therapy is unclear. FindingsHemorrhagic transformation and unfavorable clinical outcomes were independently associated with the specific location of HDAs, and HDA score effectively predicted the prognosis. Clinical relevanceThe HDA score provided a simple quantitative measure that did not require specialized software. It significantly improved the prediction of unfavorable clinical outcomes and provided a risk stratification tool for patients after endovascular therapy.

Objectives: To evaluate the value of viscoelastic parameters in the differential diagnosis of parotid gland tumors.

Materials and methods: A prospective study was conducted on 80 patients with parotid gland tumors who underwent viscoelastic ultrasound examination from September 2024 to April 2025. Patients were divided into four groups: pleomorphic adenoma (PA), warthin tumor (WT), benign tumor (BT), and malignant tumor (MT). Regions of interest (ROI) were delineated within the tumor and a 2 mm surrounding area to measure quantitative parameters, including shear wave elastography (SWE), viscous, dispersion, and strain elastography (SE) parameters. Statistical analysis of the four tumor parameters and qualitative assessment of viscoelasticity maps were performed. The diagnostic performance of each variable in classifying parotid gland tumors was assessed by comparing the area under the curve (AUC).

Results: Parameters such as E-A-mean, Vi-A-max, and Disp-A-mean showed statistically significant differences between PA and WT. Dispersion parameters, including Disp-A-mean, Disp-A-max, and Disp-A-SD, were statistically significant in distinguishing BT from MT and WT from MT. Disp-A-mean had the best diagnostic value, with an optimal threshold of 4.70 m/s/kHz for PA and WT, and 7.08 m/s/kHz for BT and MT, as well as WT and MT. The proportion of Non-Edge Distribution of high dispersion values in the MT group, as shown by the dispersion heatmap, was significantly higher (75%, 9/12) than that in the BT group (14.1%, 10/68) (p < 0.001).

Conclusion: SWE parameters and the Viscous parameter Vi-A-max are useful in differentiating PA from WT. Dispersion parameters can effectively distinguish between PA and WT, WT and MT, and BT and MT. The distribution of the dispersion coefficient can also aid in the differential diagnosis of BT and MT.

Key points: QuestionWhat is the role of viscoelastic imaging in differentiating parotid gland tumors? FindingsDispersion parameters can effectively distinguish parotid gland tumors. The distribution of the dispersion coefficient can also aid in the differential diagnosis of BT and MT. Clinical relevanceEffective preoperative differentiation of BT and MT aids in predicting disease progression, selecting surgical options, and assessing prognosis.