Academic Radiology最新文献

Rationale and objectives: Accurate preoperative assessment of axillary lymph node (ALN) status and nodal burden is crucial for individualized management of patients with breast cancer. This study aimed to develop and validate a two-stage deep learning (DL) framework that leverages preoperative breast ultrasound videos to predict ALN status and nodal burden.

Materials and methods: In this multicenter retrospective study, 864 patients with pathologically confirmed breast cancer (July 2019-December 2024) were analyzed and divided into a training set (n=495), an internal test set (n=213), and two external test sets (n=120 and 36). A two-stage framework, based on a Temporal Shift Module (TSM) video model, was proposed to first predict ALN status (negative vs positive) and subsequently classify ALN-positive patients via nodal burden (1-2 vs ≥3 nodes). Model performance was evaluated using AUC, sensitivity, and specificity, along with subgroup analyses and comparisons with other DL and clinical models.

Results: For ALN status prediction, the TSM-ResNet50 yielded AUCs of 0.851 (95% CI, 0.803-0.894), 0.886, and 0.772 across the internal and two external test sets. Performance was consistent across key subgroups, including tumors >2 cm (0.870) and BI-RADS 4 C lesions (>0.875). For nodal burden prediction, the TSM-ResNet18 achieved AUCs of 0.937, 0.797, and 0.667 for internal and two external test sets, respectively.

Conclusion: A two-stage video-based DL model showed promising performance in predicting ALN status and moderate yet clinically meaningful performance in predicting nodal burden, supporting its potential value for preoperative axillary assessment and individualized management.

Rationale and objectives: The purpose of this study was to develop and validate preoperative and postoperative recurrence-free survival (RFS) prediction models for patients with gastrointestinal stromal tumors (GISTs) of all risk levels in the gastric, small intestine, colorectal and extragastrointestinal regions.

Materials and methods: In total, 269 patients with GIST from hospital 1 were included and randomly divided into a training cohort (n=215) and an internal validation cohort (n=54). Another 42 patients from hospital 2 comprised the external validation cohort. All patients were followed up for at least 60 months. The whole tumor was 3D segmented and delineated as the region of interest (ROI) slice by slice, and 851 radiomic features were extracted from each ROI. Preoperative models were established on the basis of radiomics, clinical characteristics and CT visual information for RFS prediction. After surgery, the postoperative prediction models were constructed on the basis of preoperative information and pathological information.

Results: In the external validation, the C indices of the preoperative prediction models based of nonenhanced CT and contrast-enhanced CT of arterial phase, venous phase, delayed phase and combined phase were 0.696, 0.75, 0.771, 0.744 and 0.787, respectively; while the C indices of the postoperative prediction model were 0.776, 0.764, 0.783, 0.766, and 0.818, respectively. Non-enhanced CT could achieve effects similar to those of contrast-enhanced CT. A multipredictor nomogram was constructed for individualized estimation of RFS.

Conclusion: This study established preoperative and postoperative models for the RFS prediction in a whole GIST population with high, medium, low and extremely low risk levels. It could help clinicians develop personalized treatment plans to improve patient prognosis.

Rationale and objectives: To develop and validate a three-tier model incorporating clinical, qualitative magnetic resonance imaging (MRI), and radiomics features for preoperative prediction of axillary lymph node (ALN) metastasis in breast cancer, using nested cross-validation to ensure unbiased performance estimates.

Materials and methods: This retrospective study included 494 patients with pathologically confirmed breast cancer who underwent preoperative MRI (dynamic contrast-enhanced [DCE] and T2FS-STIR sequences) between July 2018 and August 2024. Three progressive models were developed as follows: Model 1 (clinical variables: age, tumor size, estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2), Model 2 (Model 1 + qualitative MRI features: peritumoral edema, time-intensity curve pattern, multifocality, field strength), and Model 3 (Model 2 + radiomics scores from DCE and T2FS sequences). Radiomics features were extracted using PyRadiomics and selected through variance-correlation-univariate-LASSO filtering. True nested five-fold cross-validation repeated three times ensured that feature selection and model training were performed independently within each training fold. Model performance was compared using Delong test with Bonferroni correction. SHAP analysis provided model interpretability.

Results: Model 3 achieved significantly higher AUC (0.769, 95% CI: 0.724-0.808) compared to Model 1 (0.676, 95% CI: 0.626-0.722; ΔAUC = +0.093, p<0.001) and Model 2 (0.735, 95% CI: 0.686-0.776; ΔAUC = +0.034, p = 0.002). All comparisons remained significant after Bonferroni correction (α = 0.017). Feature selection demonstrated moderate stability, with 26.8 ± 4.0 (range: 16-32) DCE and 30.8 ± 4.3 (range: 25-38) T2FS features selected per fold. SHAP analysis revealed T2FS-derived radiomics (mean |SHAP| = 0.811) and DCE-derived radiomics (mean |SHAP| = 0.492) as the most important predictors. Bootstrap validation confirmed model stability (optimism = +0.001). The model showed good calibration (Brier score = 0.201). Decision curve analysis demonstrated clinical utility across threshold probabilities 0.20-0.60. Risk stratification achieved negative predictive value of 71.5% (95% CI: 63.2%-78.8%) for low-risk and positive predictive value of 80.2% (95% CI: 73.9%-85.4%) for high-risk groups.

Conclusion: The three-tier MRI radiomics model significantly improves preoperative ALN metastasis prediction. The nested cross-validation approach ensures credible performance estimates for potential clinical implementation.

Rationale and objectives: Sturge-Weber syndrome (SWS) is a sporadic neurocutaneous disorder marked by cerebral venous abnormalities, progressive parenchymal damage, and early-onset neuro-cognitive complications. Existing imaging assessments lack standardized, quantitative approaches to capture the full disease burden. Here we tested an magnetic resonance imaging (MRI)-based scoring system that comprehensively captures both vascular and parenchymal brain abnormalities in SWS.

Materials and methods: Twenty-five young patients (mean age, 9.5 years; range, 1-24 years) with unilateral SWS brain involvement underwent 3 T MRI using a standardized protocol (with pre- and post-contrast sequences) and formal neuro-cognitive evaluation. Six imaging features, four vascular and two parenchymal, were scored by two investigators across lobes using a 3-point scale. Interrater reliability was assessed using intra-class correlation coefficients (ICC), and associations with neuro-cognitive variables were tested using Spearman's rank correlations.

Results: Both the total MRI score and each MRI subscore demonstrated excellent interrater reliability (ICC range: 0.91-0.99). Motor functions showed strong inverse correlations with the total MRI scores (ρ = -0.82, p < 0.0001). Low verbal IQ correlated with extensive calcifications (ρ = -0.55, p < 0.01). High seizure frequency correlated with greater pial enhancement (p < 0.05) and choroid plexus scores (p < 0.01). The new multiparametric score outperformed a previously established asymmetry-based MRI score in its associations with cognitive outcomes and seizure frequency.

Conclusion: This reliable and user-friendly MRI scoring system, that integrates multiple vascular and parenchymal features relevant to SWS pathophysiology, can be highly suitable for longitudinal monitoring, prognostication, and standardized outcome assessment in multicenter research and therapeutic trials.

Modern radiology requires sustained attention, rapid decision-making, and emotional resilience amid increasing imaging volumes, diagnostic complexity, and fragmented workflows. Task switching and high throughput can contribute to cognitive overload, rigidity in reasoning, and clinician burnout. While technological innovations such as artificial intelligence and workflow optimization address external factors, comparatively little attention has focused on radiologists' internal cognitive state. Zen-informed principles-including mindful attention, beginner's mind, non-attachment to outcome, intentional pausing, and compassion/interconnectedness-offer a secular, evidence-aligned framework to support perceptual clarity, cognitive flexibility, and emotional steadiness. These principles align with findings from cognitive psychology, human factors engineering, and mindfulness research, and provide conceptual guidance for mitigating bias, structuring attention, and enhancing collaborative practice. Integrating micropractices-brief pauses between cases, deliberate reorientation of focus, and awareness of downstream impact-may help radiologists navigate uncertainty with equanimity, maintain systematic search patterns, and foster effective communication and learning. This Perspective situates Zen-informed approaches within a contemporary framework, highlighting their relevance to diagnostic performance and the cultivation of resilient, attentive, and mindful radiologic practice.

Rationale and objectives: To investigate the value of dual-energy computed tomography (DECT) quantitative parameters and their percentage changes for evaluating the therapeutic response to neoadjuvant chemotherapy (NAC) in breast cancer.

Materials and methods: Clinical data from 43 patients with histologically confirmed breast cancer who underwent contrast-enhanced DECT scans before and after NAC were retrospectively analyzed. Based on the postoperative Miller-Payne (MP) grading system, patients were classified into an effective group (MP grades 4-5, n = 18) and an ineffective group (MP grades 1-3, n = 25). Quantitative parameters-including normalized iodine concentration (NIC), spectral slope (λHU), and electron density (ED)-were measured before and after NAC. The percentage changes (ΔNIC%, ΔλHU%, and ΔED%) were calculated and compared between the two groups.

Results: Before NAC, no statistically significant differences were observed in any of the quantitative parameters (NIC, λHU, and ED) between the effective and ineffective groups (P > 0.05). After NAC, the effective group showed significantly greater ΔNIC% and ΔλHU% in the arterial phase, as well as significantly greater ΔNIC%, ΔλHU%, and ΔED% in the venous phase than did the ineffective group (all P < 0.05).

Conclusion: Changes in DECT-derived quantitative parameters, especially venous-phase ΔNIC%, ΔλHU%, and ΔED%, were associated with response to neoadjuvant chemotherapy in breast cancer, indicating the potential of DECT-based metrics to support imaging evaluation of treatment response.

Rationale and objectives: To evaluate the diagnostic value of amide proton transfer-weighted imaging (APTw) combined with modified Dixon fat quantitation technique (mDixon-Quant) for differentiating triple-negative breast cancer (TNBC) from non-TNBC.

Materials and methods: This retrospective study included 107 breast cancer patients who underwent preoperative MRI with APTw and mDixon-Quant. Based on immunohistochemistry, patients were classified into TNBC (n=24) and non-TNBC (n=83) groups. Two radiologists independently measured APTw, fat fraction (FF), and T2* values. Interobserver consistency was assessed using the intraclass correlation coefficient (ICC). Continuous and categorical variables were compared using the Mann-Whitney U and χ² tests, respectively. Diagnostic performance was evaluated and compared using receiver operating characteristic (ROC) analysis, with the DeLong test for AUC comparisons.

Results: The TNBC group showed significantly higher Ki-67 index, APTw, and T2* values, but lower FF values than the non-TNBC group (all p<0.05). The combined use of APTw and FF values demonstrated favorable diagnostic efficacy, with an AUC of 0.885. Adding Ki-67 index increased the AUC to 0.911, but the difference was not significant (p>0.05).

Conclusion: APTw combined with mDixon-Quant provides valuable non-invasive imaging evidence for preoperative TNBC diagnosis, guiding clinical treatment strategies and prognostic assessment.

Rationale and objectives: To investigate the performance of deep learning image reconstruction (DLIR) at an ultra-low dose of approximately 4.5 mGy for detecting focal liver lesions (FLLs), in comparison with adaptive statistical iterative reconstruction-V (ASIR-V) at standard doses (10-15 mGy), through both phantom and prospective patient studies.

Materials and methods: A Gammex CT phantom (simulating FLLs with iodine concentrations 2.0-20 mg/mL and normal liver density 1.06 g/cm³) was scanned using DLIR (4.5 mGy) and ASIR-V (10 and 15 mGy). Quantitative metrics included image noise, signal-to-noise ratio, contrast-to-noise ratio, noise power spectrum peak (NPS_peak), and detectability index. In a prospective single-center study, 84 participants (mean age 64 ± 12 years, IQR 60-69 years; 48 males) underwent triple-phase upper-abdominal CT (target dose: 4.5 mGy/phase). Images were reconstructed with DLIR and ASIR-V. Two radiologists blindly evaluated image quality (5-point scale), FLL detectability, sensitivity, and specificity. The reference standard for detectability included histopathology, 3-month standard-dose CT, or MRI.

Results: In the phantom study, DLIR at 4.5 mGy outperformed ASIR-V at 10 mGy across all quantitative metrics (P < 0.001) and exceeded ASIR-V at 15 mGy in noise and NPS_peak (P < 0.05). Clinically, 71 FLLs (mean size 12.8 ± 10 mm; 55 benign, 16 malignant) were identified. The median CTDI_vol was 4.64 mGy (50% to 70% lower than standard doses). DLIR showed superior qualitative image quality vs. ASIR-V (1.25 mm slices: 3.9 ± 0.6 vs. 2.2 ± 0.4, P < 0.001) and higher FLL detection rate (93.0% vs. 77.5%, P < 0.001), with sensitivity 90.1% and specificity 75.0% (both higher than ASIR-V, P < 0.001).

Conclusion: DLIR at 4.5 mGy achieves substantial radiation dose reduction while providing superior FLL detection performance compared to ASIR-V at 10-15 mGy. This protocol offers a safe and accurate option for FLL screening and follow-up.

Rationale and objectives: Neuroimaging studies have revealed that anxiety disorders (ADs) have been associated with altered cortical thickness (CTh) in some key brain regions, but findings are inconsistent and the molecular basis of structural damage remains unknown. Here, the aim of this study was to identify the most consistent CTh alterations in ADs and to characterize their underlying molecular features.

Materials and methods: A comprehensive meta-analysis was conducted to identify consistent CTh alterations in patients with ADs by using anisotropic effect-size seed-based d mapping (AES-SDM) software. On this basis, spatial correlations between neurotransmitter distribution data and the CTh alterations were investigated using the JuSpace toolbox, thereby revealing the neural mechanisms underlying ADs from a cross-modal perspective.

Results: A total of 9 studies comprising 264 patients with ADs and 286 healthy controls (HCs) were included. Compared with HCs, ADs showed increased CTh in the left precentral gyrus (PreCG) and left insula and decreased CTh in the dorsolateral region of the right superior frontal gyrus. In meta-regression analyses, the CTh alterations in the left PreCG were negatively correlated with the mean age and percentage of males in the patients, respectively. The pattern of structural alterations associated with ADs was also correlated with the distribution of serotonergic, GABAergic, cholinergic, and glutamatergic neurotransmitters.

Conclusion: By linking abnormal CTh to specific neurotransmitter systems, this work advances an integrative understanding of the morphological alterations in ADs and their molecular basis, which provides clues to potential therapeutic targets.

Take-home message: This meta-analysis revealed consistent CTh abnormalities in sensorimotor cortex, limbic system, and dlPFC regions in ADs, correlating with multiple neurotransmitter distributions. While offering initial insights into complex neuropathogenesis and potential therapeutic targets, these findings remain preliminary and need to be validated through larger-scale, multimodal studies in well-defined phenotypic cohorts.

Rationale and objectives: To develop and validate an ultrasound-based habitat radiomics model for the preoperative prediction of pathologic upgrade in patients with ductal carcinoma in situ (DCIS) diagnosed by core needle biopsy (CNB).

Materials and methods: This retrospective study included 167 female patients (September 2015-August 2025) diagnosed with DCIS by CNB, randomly divided into training (n = 116) and test (n = 51) cohorts. Clinical risk factors were identified using univariate and multivariable logistic regression. Tumor regions of interest were segmented into three habitat subregions (K = 3) via K-means clustering based on 39 pixel-level features. Four predictive models (clinical, radiomics, habitat, and combined) were constructed using a Random Forest classifier. Performance was evaluated using the receiver operating characteristic curves, DeLong's test, calibration curves, and decision curve analysis (DCA).

Results: Adler blood-flow grade (OR = 1.201) and high Ki-67 expression (OR = 1.469) were independent clinical predictors (P < 0.05). In the test cohort, the Habitat model (AUC = 0.891) significantly outperformed the conventional Radiomics model (AUC = 0.741, P < 0.05). The Combined model (Habitat+Clinical) achieved the highest predictive performance (AUC = 0.925; accuracy = 0.863; specificity = 0.920). The combined model showed excellent calibration (Hosmer-Lemeshow P > 0.05) and provided the greatest net benefit in DCA.

Conclusion: An ultrasound-based habitat radiomics model, combined with clinical predictors, provides a robust noninvasive tool for predicting pathologic upgrade in CNB-diagnosed DCIS. This model may assist clinical decision-making and support individualized treatment planning.