Insights into Imaging最新文献

Reporting and Data Systems (RADS) aim at standardizing imaging acquisition, interpretation, lexicon, and reporting standards in specific patient populations, facilitating the communication between radiologists and clinicians. While the adoption of RADS has been supported by several studies and guidelines, with some of them endorsed by the American College of Radiology, the clinical adoption of the RADS algorithm remains heterogeneous among general practice radiologists worldwide, being lower in non-academic and young radiologists. This article aims to provide an updated review, aimed at young and general radiologists, of the RADS alphabet, discussing the main applications and imaging criteria with tips for their correct use in clinical practice. The following RADS will be discussed: BI-RADS, Bone-RADS, C-RADS, CAD-RADS, LI-RADS, Lung-RADS, MET-RADS-P, MY-RADS, NI-RADS, Node-RADS, O-RADS, ONCO-RADS, PI-RADS, ST-RADS, TI-RADS, and VI-RADS. CRITICAL RELEVANCE STATEMENT: A comprehensive guide aimed at young and general radiologists featuring all of the major RADS with the objective to foster their implementation in clinical practice, which could be beneficial in a further standardization of the medical reports and in the communication between radiologists and clinicians. KEY POINTS: RADS are outlined to enhance communication efficacy between radiologists and clinicians. Updated overview of RADS frameworks, detailing applications, imaging criteria, and advancements. RADS' implementation remains a challenge, but can be addressed.

Objectives: To assess the correlation between the functional liver imaging score (FLIS) and FibroScan^®-derived fibrosis stage, and to determine whether incorporating parenchymal heterogeneity (FLIS-H) improves its association with fibrosis and clinical scores.

Materials and methods: This retrospective single-centre study included 113 patients who underwent FibroScan^® and hepatocyte-specific contrast-enhanced MRI within a median interval of 4 days. FLIS was calculated, and the parenchymal heterogeneity score was added to FLIS (FLIS-H; range 0-8). Inter-reader agreement was evaluated using a two-way random-effects intraclass correlation coefficient (ICC). Correlations between FLIS/FLIS-H and fibrosis stage/clinical scores (Child-Pugh, MELD, ALBI) were assessed using Spearman's rank correlation. Steiger's z-test and Zou's method were used to compare correlations.

Results: A total of 113 patients (67 men; mean age 56.6 ± 13.5 years) were evaluated. Inter-reader agreement was excellent for FLIS (ICC 0.994; 95% CI: 0.975-1.000), heterogeneity (ICC 0.949; 95% CI: 0.901-0.984), and FLIS-H (ICC 0.974; 95% CI: 0.957-0.989). FLIS showed significant negative correlations with Child-Pugh (ρ = -0.2664, p = 0.0087), ALBI (ρ = -0.3076, p = 0.0022), and fibrosis stage (ρ = -0.3207, p < 0.001). FLIS-H demonstrated stronger correlations with Child-Pugh (ρ = -0.4167, p < 0.001), ALBI (ρ = -0.5243, p < 0.001), MELD (ρ = -0.2360, p = 0.020), and fibrosis stage (ρ = -0.5270, p < 0.001). Steiger's z-test confirmed that correlations were significantly improved with FLIS-H for ALBI (z = -3.03, p = 0.0025), Child-Pugh (z = -2.01, p = 0.045), and fibrosis stage (z = -2.90, p = 0.0038).

Conclusion: FLIS correlates significantly with fibrosis stage and clinical scores. Incorporating parenchymal heterogeneity into FLIS enhances these associations and may provide a superior method for liver assessment.

Critical relevance: This study introduces a modified FLIS version (FLIS-H) that integrates parenchymal heterogeneity and demonstrates superior correlation with elastography-derived fibrosis stages and clinical scoring systems, offering a practical improvement for non-invasive assessment in routine practice.

Key points: FLIS has no reported correlation with elastography-based liver fibrosis staging. Parenchymal heterogeneity is not included as a parameter in the standard FLIS. Integrating heterogeneity improves correlation with fibrosis stage and clinical scores. FLIS-H enables fast, reliable, structure-function liver assessment in clinical radiology.

Objectives: To determine the rate of malignancy upgrade in MRI-only B3 lesions and to identify clinical, imaging, and histological features that can predict upgrade.

Materials and methods: This retrospective single-center study included MRI-only lesions diagnosed as B3 after MRI-guided vacuum-assisted biopsy and excised between January 2007 and March 2023. We calculated upgrade rates for the entire series and for subgroups based on possible risk factors. To analyze variables considered risk factors for upgrade, we used logistic regression, calculating odds ratios (OR) and their 95% confidence intervals (CI).

Results: Of 592 lesions biopsied, 89 (15.03%) were classified as B3. After excluding 30 lesions because excisional specimen results were unavailable, we analyzed 59 lesions in 51 patients. Biopsy classified 15 (25.4%) lesions as pure atypical ductal hyperplasia (ADH), 27 (45.8%) as pure flat epithelial atypia (FEA), 12 (20.3%) as mixed lesions, and 5 (8.5%) as lobular neoplasia. A total of 7 (11.9%) lesions were upgraded to malignancy (71.4% to ductal carcinoma in situ, 14.3% to invasive ductal carcinoma, and 4.3% to invasive lobular carcinoma). Although histological type was not associated with upgrade to malignancy (p = 0.47), 20% of pure ADH and only 3.7% of pure FEA lesions were upgraded. Larger lesion size on MRI was associated with upgrade [6.25% of lesions ≤ 20 mm vs. 36.4% of those > 20 mm, p = 0.02; OR 8.57 (95% CI: 1.57‒46.71) p = 0.01].

Conclusion: Lesion size may predict upgrade in MRI-only B3 lesions independent of histological type; imaging follow-up may suffice for FEA lesions measuring < 20 mm.

Critical relevance statement: Considering lesion size and histological type could help define the management of MRI-only lesions classified as B3 after MRI-guided vacuum-assisted biopsy.

Key points: The management of MRI-only B3 lesions has yet to be established. Lesion size is a relevant factor to consider when deciding clinical management in MRI-only B3 lesions. Conservative management appears to be safe in selected flat epithelial atypia lesions (< 20 mm).

Objectives: To evaluate the conspicuity of fast field echo resembling a CT using restricted echo-spacing (FRACTURE) in visualizing hand tendons and assess the utility of FRACTURE-derived volume rendering (VR) images using MRI in healthy individuals.

Materials and methods: This prospective observational study enrolled ten healthy volunteers who underwent MRI, including FRACTURE, three-dimensional proton density-weighted volume isotropic turbo spin-echo acquisition (PD-VISTA), and two-dimensional T2-weighted image (T2WI) in neutral and ulnar deviation positions. VR images depicting bones and tendons were created from FRACTURE data. Twenty-four flexor and extensor tendons were qualitatively evaluated by four experienced readers using a 5-point scale for cross-sectional images (including FRACTURE inversion) and a 3-point scale for VR images. Quantitative analysis included tendon cross-sectional area measurements and contrast-to-noise ratio (CNR) calculations. Inter- and intra-reader reliability and FRACTURE-inversion agreement were assessed using weighted kappa coefficients. Statistical analysis included an ordinal mixed-effects model, Bland-Altman analysis, correlation coefficients, and paired t-tests.

Results: Ten healthy volunteers (5 men, 5 women, mean age 37.4 ± 9.1 years) were evaluated. FRACTURE achieved the highest qualitative scores (3.30 ± 0.364) compared to PD-VISTA (3.09 ± 0.265) and T2WI (2.60 ± 0.509), showing statistically significant superiority by ordinal mixed-effects modeling (p < 0.001). FRACTURE inversion showed high agreement with FRACTURE (weighted kappa = 0.975). Tendon cross-sectional area measurements showed strong correlations between sequences (r = 0.680-0.740) but significant systematic bias (p < 0.017), with FRACTURE measuring consistently smaller areas. FRACTURE demonstrated significantly higher CNR for muscle-tendon comparisons (12.63 ± 1.088 vs 7.911 ± 1.746, p < 0.017).

Conclusion: FRACTURE provides superior hand tendon visualization compared to conventional MRI sequences, with potential value for clinical practice.

Critical relevance statement: FRACTURE showed superior hand tendon visualization compared to T2WI and PD-VISTA, potentially helping assess anatomical variations. VR images provide a three-dimensional understanding of the hand tendon structure. These capabilities could enhance surgical planning and procedure selection in hand surgery.

Key points: FRACTURE performs better than T2WI and PD-VISTA for evaluating hand tendons. FRACTURE provides excellent contrast, enabling the creation of VR images. FRACTURE could serve as an aid in surgical planning and procedure selection, with the potential to improve hand surgery practice.

Background: Pediatric rhabdomyosarcoma (RMS), the most common soft-tissue sarcoma in children, exhibits heterogeneous responses to neoadjuvant chemotherapy (NAC), necessitating reliable biomarkers for early prediction. This multicenter study evaluates MRI-derived radiomic features of intratumoral and peritumoral regions to predict NAC response in the largest pediatric RMS cohort to date.

Materials and methods: A retrospective analysis included 519 RMS patients from three Chinese centers. Radiologists manually segmented tumors and 2-mm peritumoral regions on standardized T1-weighted contrast-enhanced (T1CE) and T2-weighted fat-saturated (T2Fs) MRI sequences. PyRadiomics extracted 1015 radiomic features, with robustness ensured (ICC ≥ 0.80) and predictive features selected via LASSO regression. Twelve XGBoost models (intra-/peritumoral, multisequence) were developed, validated internally/externally, and compared using DeLong's test, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). SHAP analysis interpreted feature contributions. Clinical variables (age, fusion gene) were assessed for incremental value.

Results: The T1CE-based combined intratumoral-peritumoral model (T1CE_IntraPeri2mm) demonstrated the best generalizability, achieving AUCs of 0.917 (training), 0.760 (internal validation), 0.837 (external test1) and 0.843 (external test2). It significantly outperformed intratumoral-only and multisequence fusion models in DeLong, NRI, and IDI analyses (all p < 0.05). The combined clinical-radiomic model did not provide incremental benefit (AUC: 0.843 vs. 0.838, p = 0.891). SHAP analysis indicated that features reflecting peritumoral structural irregularity and enhancement heterogeneity were key predictors of NAC resistance.

Conclusion: T1CE-based peritumoral radiomics robustly predicts NAC response in pediatric RMS, emphasizing tumor-microenvironment interactions. This approach offers a non-invasive tool for personalized therapy stratification.

Critical relevance statement: This study establishes peritumoral MRI radiomics as a critical predictor of chemotherapy response in pediatric rhabdomyosarcoma, addressing the unmet need for non-invasive biomarkers and advancing precision oncology through tumor-microenvironment interaction analysis in clinical radiology practice.

Key points: Integrated tumor/peritumoral MRI features enhance neoadjuvant chemotherapy (NAC) response prediction. T1CE MRI best captures tumor-microenvironment treatment interactions. Non-invasive radiomics model outperforms clinical factors for therapy adjustment.

Objectives: This study aims to develop a radiomics model based on carotid perivascular adipose tissue (PVAT) from CT angiography to identify histologically confirmed vulnerable plaques in patients with carotid artery stenosis (CAS).

Materials and methods: In this prospective cohort study, we enrolled patients with CAS scheduled for carotid endarterectomy between 2014 and 2023. Histological plaque assessment served as the reference standard for vulnerability. We developed three models: the PVAT attenuation model, the conventional plaque feature model, and the PVAT radiomics model using features extracted from segmented CT images and machine learning. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration, and decision curve analysis across training, validation, and independent testing from three different scanners. Shapley Additive exPlanations (SHAP), a tool that quantifies the contribution of each feature to the model's predictions, was used to enhance model interpretability.

Results: We included 122 patients (mean age 66.45 years, 81.97% male, 63.11% vulnerable). In the training, validation, and testing sets, the PVAT radiomics model predicts an AUC of vulnerability of 0.945, 0.819, and 0.817, respectively, while the plaque score model showed an AUC of 0.688, 0.799, and 0.497, and the PVAT attenuation model showed an AUC of 0.667, 0.708, and 0.493, respectively. The PVAT radiomics model outperforms the PVAT attenuation model (p = 0.01) and plaque score models (p = 0.03) in the test set. SHAP analysis highlighted significant predictors such as logarithm_firstorder_RootMeanSquared.

Conclusions: The PVAT radiomics model is a promising non-invasive tool for identifying vulnerable carotid plaques, offering superior diagnostic efficacy and generalizability across different imaging equipment.

Critical relevance statement: The carotid PVAT radiomics identified histologically vulnerable plaques before surgery through an interpretable and generalizable machine-learning model, beneficial for risk stratification and surgical decision-making.

Key points: Noninvasive and effective identification of histological carotid vulnerable plaques is challenging. The PVAT radiomics outperforms conventional imaging biomarkers in identifying vulnerable plaques. The PVAT radiomic model is generalizable across scanners and interpretable, assisting clinical decision-making.

Objectives: To investigate the prognostic value of artificial intelligence (AI) quantified emphysema and interstitial lung abnormality (ILA) in patients with non-small cell lung cancer (NSCLC).

Materials and methods: This retrospective study used AI to quantify emphysema and ILA in patients diagnosed with NSCLC between January 2015 and December 2017. Associations between AI-quantified emphysema and ILA severity and overall survival (OS) were evaluated using Cox proportional hazards models. The ability of AI-quantified emphysema and ILA severity to predict OS was explored via concordance index (C-index) and area under the time-dependent receiver operating characteristic curve (AUC). Furthermore, exploratory OS analyses were performed on subgroups stratified by chronic obstructive pulmonary disease status, treatment type, and tumor-node-metastasis (TNM) staging.

Results: Of 1675 patients, 830 (49.6%) survived, and 845 (50.4%) died. Whole emphysema (mild: HR, 1.66 [95% CI: 1.26, 2.18]; p < 0.001; more than mild: HR, 2.55 [95% CI: 1.88, 3.48]; p < 0.001) and ILA (equivocal ILA: HR, 1.63 [95% CI: 1.15, 2.32]; p = 0.006; definite ILA: HR, 2.33 [95% CI: 1.61, 3.35]; p < 0.001) severity were independent prognostic factors for NSCLC, while regional emphysema and regional ILA severity were not. The model combining AI-quantified whole emphysema severity and ILA severity outperformed the TNM staging-only model in predicting NSCLC patient outcome (C-index, 0.80 vs. 0.75; AUC, 0.90 vs. 0.85).

Conclusions: Increased AI-quantified whole emphysema and ILA severity were associated with worse OS in NSCLC. The model combining AI-quantified emphysema and ILA showed improved performance for predicting patient survival versus TNM staging alone.

Critical relevance statement: AI-quantified emphysema and ILA severity are associated with NSCLC patient outcome and can provide information complementary to TNM staging for predicting NSCLC patient survival and promoting the development of individualized management strategies.

Key points: The study explores artificial intelligence (AI) quantified emphysema and interstitial lung abnormality (ILA) severity in non-small cell lung cancer (NSCLC) prognosis. The AI-quantified whole emphysema severity and ILA severity were independent prognostic factors for NSCLC patient outcome, while regional emphysema and regional ILA severity were not. AI-quantified emphysema and ILA severity may help predict the survival of NSCLC patients and help clinicians make informed treatment decisions.

Objective: Development of a preoperative mesorectal lymph node metastasis (LNM) prediction model for rectal cancer (RC) based on intratumoral and multiregional peritumoral radiomics features extracted from super-resolution multiparametric MRI.

Materials and methods: This multicenter study included preoperative MRI data from 243 rectal cancer patients (194 from center A, 49 from center B) with SR reconstruction and scoring. Radiomic features were extracted from tumor, peri-3mm and peri-5mm on SR-DWI and SR-T2WI images. The least absolute shrinkage and selection operator (LASSO) and the maximum relevance minimum redundancy (mRMR) were used for feature selection and dimensionality reduction. DWI_T2WI_INTRA, DWI_T2WI_IntraPeri3mm, DWI_T2WI_InterPeri5mm models were developed employing Logistic regression. Independent clinical risk factors identified through univariate and multivariate stepwise regression analyses were used to construct a clinical model. The optimal IntraPeri model integrated with clinical model design the combined model. Predictive performance was evaluated using ROC curves, calibration curves, and decision curve analysis (DCA).

Results: Qualitative evaluation demonstrated superior scores for SR-T2WI across five metrics compared to original images (all p < 0.001). For DWI, SR images achieved significant improvements in all parameters (p < 0.001), except lesion conspicuity [median (IQR): 3 (1) vs. 3 (1)]. Comparative analysis revealed the DWI_T2WI_IntraPeri3mm model's optimal predictive performance in training, validation, and test cohorts (AUCs: 0.880, 0.735, and 0.714, respectively). The AUC of the combined model, integrating radiomic (DWI_T2WI_IntraPeri3mm) model with clinical risk factors, was 0.933, 0.829, and 0.867 in each cohort, all exceeding those of the clinical and radiomic models.

Conclusion: Using GANs-based 3D-SR of multi-sequence MRI, our multiregional prediction model for preoperative mesorectal LNM in RC demonstrated good diagnostic performance.

Critical relevance statement: The integration of super-resolution-based tumor and peritumoral 3-mm predictive model with clinical risk factors enables performance in predicting mesorectal LNM, potentially aiding clinical therapeutic decision-making.

Key points: How do tumor and peritumoral (3-5 mm) models-based SR images perform in predicting lymph node metastasis (LNM)? The DWI_T2WI_IntraPeri3mm model, when combined with clinical factors, improves diagnostic accuracy. Multiparametric, multiregional super-resolution (SR)-MRI radiomics models exhibit good performance for LNM.

Objectives: The use of AI is gaining relevance in healthcare. There is limited information regarding the views of patients on AI in healthcare. The aim of our study was to assess the views of patients on the use of AI in healthcare with an on-site questionnaire.

Materials and methods: Patients in our tertiary hospital with a diagnostic imaging appointment were invited to complete a paper-based questionnaire between December 2022 and October 2023. We asked about socio-demographic data, experience, knowledge, and their opinion on the use of AI in healthcare, focusing on the fields (1) diagnostics, (2) therapy, and (3) triage.

Results: Out of a total of 198 patients (mean age 49.41 ± 17.6 years, 99 female), 91.5% stated that they expected benefits from the implementation of AI in healthcare, although 73.4% rated their knowledge of AI as moderate to none. The majority of patients were in favour of using AI in diagnostics (87.2%) and therapy (73.1%), while only 28.2% approved its use in patient triage. 84.0% wanted to be informed about the use of AI in at least one of the mentioned areas. Participants with higher education, higher self-assessed knowledge of AI and personal experience with AI showed greater approval for AI in healthcare.

Conclusion: Our interviewed patients have a rather open attitude towards AI in healthcare, with differentiated views depending on the topic; patients are in favour of the use of AI, especially in diagnostics and to a lesser extent also for therapy support, but they reject its use for triage.

Critical relevance statement: Overall, the results emphasise the need for widespread efforts to address patient concerns about AI in healthcare, including enhancing understanding and acceptance while protecting marginalised groups. This will help clinical radiology to adopt AI more effectively.

Key points: There is limited information on patients' views of AI in healthcare, often focused on specific groups, limiting generalizability. Patients are open to AI in healthcare, supporting its use in diagnostics and therapy, but rejecting its use for triage. Overall, patients want to be informed about AI usage and participants with higher education and AI experience showed more approval.

Objectives: To evaluate the stability of radiomic features under different CT acquisition settings and investigate its impact on diagnostic model performance and generalizability.

Materials and methods: 198 patients with 1227 pulmonary nodules underwent chest CT scans using varied settings (three slice thicknesses, two reconstruction matrices, six convolution kernels, two transmission methods). 1394 radiomic features were extracted per nodule. Feature stability was evaluated using the Intraclass Correlation Coefficient (ICC, stable: ICC ≥ 0.8, intermediate stable: 0.4 < ICC < 0.8, unstable: ICC ≤ 0.4). Four diagnostic models (Full-feature, Stable, Unstable, Intermediate stable) were developed using two datasets (lung cancer screening, n = 184; clinical scenarios, n = 1192). In addition, three combination models were constructed for ablation analysis. Model performance and generalizability were assessed via fivefold cross-validation and independent test sets with different CT parameters.

Results: Slice thickness and image transmission methods had the greatest and least impacts on feature stability (7.0% and 83.0% stable features, respectively). In training and validation sets, the Full-feature and Intermediate stable models showed higher AUCs than the Stable and Unstable models (p < 0.05). However, in test sets with varying CT parameters, the Stable model maintained consistent performance (AUC: 0.693-0.728), while the Unstable model exhibited the greatest variability (AUC: 0.523-0.800). Notably, the Full-feature and Intermediate stable models largely predicted nodules as benign, exhibiting limited ability to discriminate malignant cases.

Conclusion: Radiomic feature stability is significantly affected by CT reconstruction parameters, especially slice thickness. Models based on stable features demonstrate better generalizability across varying CT settings, underscoring the importance of assessing feature stability in radiomic-based diagnostics.

Critical relevance statement: Radiomic feature stability is significantly affected by CT acquisition parameters. Stable radiomic features enhance diagnostic model consistency and reliability across diverse CT settings. Therefore, feature stability analysis and selection of stable features are crucial to enhance model generalizability and stability.

Key points: How do CT settings affect radiomic feature stability and model performance? Feature stability varies with CT parameters, but stable features enhance model generalizability. Stable feature models boost diagnostic reliability and clinical applicability.