Academic Radiology最新文献

RATIONALE AND OBJECTIVES: To evaluate correlations between a standardized vascular obstruction score and invasive hemodynamic parameters in chronic thromboembolic pulmonary hypertension (CTEPH), using dual-energy CT (DECT), cone-beam CT (CBCT), and digital subtraction angiography (DSA).

Materials and methods: In this retrospective single-center study, 109 patients with CTEPH underwent DECT, CBCT, and DSA within a 3-month interval. A standardized vascular obstruction score was applied independently to each modality. Linear regression models were constructed to assess associations with mean pulmonary arterial pressure (mPAP), pulmonary vascular resistance (PVR), cardiac output (CO), and cardiac index (CI), quantified by adjusted R². Score distributions were compared using Friedman and Wilcoxon tests, and interobserver agreement was assessed with Cohen's κ.

Results: DSA demonstrated the highest degree of association with mPAP and PVR among the evaluated modalities (adjusted R² = 0.089 and 0.126), followed by DECT (0.075 and 0.098) and CBCT (0.050 and 0.062). DSA also correlated with CO and CI. Mean obstruction scores differed significantly across modalities (p < 0.001), with DECT yielding higher values than CBCT (p < 0.001) and DSA (p < 0.001). Interobserver agreement was highest for CBCT (κ = 0.76) and DECT (κ = 0.74), and lowest for DSA (κ = 0.57). None of the modalities correlated significantly with NYHA class or 6MWD.

Conclusion: A unified morphologic vascular obstruction score applied across DECT, CBCT, and DSA demonstrates reproducible associations with invasive hemodynamic parameters in CTEPH. While not a replacement for right heart catheterization, it provides a standardized framework for multimodality assessment and may support methodological integration across imaging modalities.

Critical relevance statement: This study presents a systematic application of a unified vascular obstruction scoring system across dual-energy CT, cone-beam CT, and digital subtraction angiography in patients with chronic thromboembolic pulmonary hypertension. The results demonstrate significant correlations with invasive hemodynamic parameters, with dual-energy CT and cone-beam CT providing higher reproducibility than angiography. These findings support the use of a standardized scoring framework to enable consistent multimodality assessment, improve reproducibility in structured multimodality imaging assessment, and facilitate cross-institutional comparisons in chronic thromboembolic pulmonary hypertension.

Rationale and objectives: Accurate prediction of axillary lymph node metastasis (ALNM) after neoadjuvant chemotherapy (NAC) remains challenging in breast cancer. This study aimed to develop an interpretable machine learning model integrating MRI-based radiomics, deep learning features, and the Node-RADS score for noninvasive ALNM prediction after NAC.

Materials and methods: In this multicenter retrospective study, 641 patients with pathologically confirmed breast cancer who underwent surgery between April 2017 and December 2024 across three institutions were enrolled. Preoperative dynamic contrast-enhanced MRI and clinicopathologic data were analyzed. Quantitative radiomics and ResNet50-derived deep learning features were extracted. Patients were divided into a training cohort (n = 397), an internal validation cohort (n = 99), and two external validation cohorts (n = 90 and n = 55). Three models-a clinical model, a deep learning-radiomics (DLR) model, and a combined clinical-deep learning-radiomics (CDLR) model-were constructed using five machine learning algorithms. Model performance was evaluated by ROC analysis, AUC, calibration, and decision curve analysis. SHapley Additive exPlanations (SHAP) were used to interpret feature importance.

Results: The CDLR model demonstrated superior predictive performance, with AUCs of 0.879, 0.805, 0.737, and 0.781 in the training, internal, and two external cohorts, respectively, outperforming both the DLR and clinical models. The CDLR model also showed good calibration and the highest net clinical benefit. SHAP analysis identified Node-RADS, lbp_3D_m1_glcm_Correlation, and DL_50 as the most influential predictors.

Conclusion: The interpretable CDLR model enables accurate, noninvasive prediction of ALNM after NAC in breast cancer and may assist in individualized clinical decision-making.

Rationale and objectives: Early and accurate staging of rectal cancer is essential for selecting optimal treatment strategies. This study aimed to evaluate the utility of a combined clinical, tumoral, and peritumoral radiomics model for predicting T1 and T2 rectal cancer staging.

Materials and methods: We retrospectively enrolled patients with pathologically confirmed rectal cancer from three medical centers between August 2018 and December 2024. Radiomics features were extracted from both tumoral and peritumoral regions using preoperative magnetic resonance imaging scans. The radiomics model with the highest area under the curve (AUC) was combined with a clinical model to construct a fusion model for distinguishing T1 and T2 stages.

Results: A total of 392 patients were included and allocated to a training set (n = 208), an internal test set (n = 90), and an external test set (n = 94). The fusion model (clinical+Com-T2WI) demonstrated robust performance, achieving AUCs of 0.91, 0.82, and 0.88 in the training, internal, and external test sets, respectively. Tumor thickness (P =.034) and tumor length (P <.001) were identified as independent predictors, further enhancing the model's staging accuracy.

Conclusion: The proposed fusion model provides a noninvasive, effective tool for preoperative differentiation of T1 and T2 rectal cancer. While the model achieved the best predictive performance in this study, prospective validation is required before clinical implementation.

Rationale and objectives: This study aims to evaluate the effect of ChatGPT-assisted reflective reasoning on guideline-concordant procedural decision-making among early-career interventional radiologists using standardized clinical scenarios based on the American College of Radiology Appropriateness Criteria.

Materials and methods: This prospective simulation-based study included 128 scenarios across common interventional radiology indications. Two expert interventional radiologists served as the reference standard. Three early-career radiologists completed all scenarios twice: first independently (pre-ChatGPT) and, after a two-month washout period, with access to ChatGPT-generated reasoning before recording final decisions (post-ChatGPT). Guideline concordance was assessed using a three-tier scoring system (appropriate = 2, may be appropriate = 1, inappropriate = 0) and a binary score reflecting avoidance of inappropriate decisions. Predifferences and postdifferences were analyzed with Wilcoxon signed-rank and McNemar tests. Agreement with experts was measured using Cohen's kappa.

Results: ChatGPT-assisted reflective reasoning significantly improved guideline-concordant decision-making. The mean detailed compliance score increased from 1.697 to 1.900, and minimal compliance enhanced from 90.89% to 98.70%. A total of 30 scenario-level corrections shifted from inappropriate to guideline-concordant selections (McNemar χ² = 27.03; p < 0.0001). Detailed compliance improved significantly for all radiologists (p < 0.01). Weighted Cohen's kappa increased from 0.08-0.13 to 0.21-0.30, indicating better agreement with expert consensus. Performance variability decreased, narrowing the gap between early-career radiologists and experts.

Conclusion: ChatGPT-assisted reflective reasoning enhanced guideline alignment and reduced inappropriate procedural selections among early-career interventional radiologists. These findings support the role of large language models as cognitive support tools during early clinical practice and warrant prospective evaluation in real-world settings.

Rationale and objectives: By 2033, the U.S. may face a shortage of up to 139,000 physicians, including radiologists. Many international medical graduates (IMGs) pursue the American Board of Radiology (ABR) Alternate Pathway, four years of U.S.-based training, research, or faculty experience, to achieve board eligibility. This study evaluated the performance of neuroradiology fellows in the ABR Alternate Pathway compared to U.S. DR residency graduates.

Materials and methods: Data were obtained from the ABR and neuroradiology fellowship program directors via a survey distributed in January-December 2025, with five reminders for non-respondents. The survey assessed clinical performance, research productivity, and board examination outcomes. Participation was voluntary.

Results: Of 80 neuroradiology fellowship program directors surveyed, 59 responded (74%). Among respondents, 64% reported accepting ABR Alternate Pathway Candidates (APCs). Research performance was rated as stronger in 34%, comparable in 22%, and weaker in 8% of programs (36% no response). Clinical skills were rated stronger in 7%, comparable in 34%, and weaker in 24%. Teaching ability was rated stronger in 17%, comparable in 31%, and weaker in 17%. Board examination performance was largely comparable between APCs and U.S. DR graduates for both Core (44%) and CAQ (39%) exams, with similar first-attempt pass rates.

Conclusion: ABR Alternate Pathway candidates perform comparably to U.S. DR graduates in clinical, research, teaching, and examination metrics. Integrating IMGs through the Alternate Pathway can help address the projected radiologist shortage while maintaining high educational and clinical standards.

Rationale and objectives: With the emergence of disease-modifying therapies, precise staging of dementia is urgent. This study aimed to develop a machine learning model integrating multimodal data to achieve objective staging of dementia severity in patients with cognitive impairment.

Materials and methods: A total of 149 patients (100 with Alzheimer's disease) were recruited. Demographic data, neuropsychological scores, and multimodal PET features were collected. Subjects were randomly split (7:3) into training and validation cohorts. PET features were screened using Boruta and LASSO to generate composite SUVR scores, while key demographic and neuropsychological predictors were identified through univariate and multivariate logistic regression analyses. Seven machine learning algorithms (logistic regression, decision tree, random forest, XGBoost, LightGBM, support vector machine, and artificial neural network) were trained using grid search and fivefold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration plots, and decision curve analysis (DCA), with SHAP analysis employed for interpretability.

Results: The cohort comprised 80 very mild-to-mild (CDR 0.5-1) and 69 moderate-to-severe (CDR 2-3) dementia cases. Key predictors included years of education, MMSE, and composite amyloid and FDG SUVR scores. The XGBoost model demonstrated robust performance, achieving an AUC of 0.888 (95% CI: 0.777-0.967) in the independent validation cohort. SHAP analysis identified MMSE and composite FDG SUVR scores as the most significant contributors to disease staging.

Conclusion: This study constructed and internally validated an interpretable multimodal model for dementia severity staging. While the results are promising, the developed web-based tool currently serves as a proof-of-concept to demonstrate how such models could potentially assist in optimizing patient management and screening candidates for novel therapies, pending further external validation.