RPS 602 - Advances and emerging trends in breast MRI

Author Block: N. Mesropyan¹, C. Katemann², A. Isaak¹, J. A. Luetkens¹; ¹Bonn/DE, ²Hamburg/DE
Purpose: To clinically implement and evaluate a 4D diamond-shaped pseudo–golden angle stack-of-stars (SoS) acquisition with k-space weighted image contrast (KWIC) reconstruction for breast dynamic contrast-enhanced MRI (4D-DCE), assessing image quality, diagnostic confidence, and BI-RADS agreement across conventional and ultrafast protocols.
Methods or Background: This retrospective study included female patients who underwent breast MRI at 3T using the 4D-DCE sequence. Three protocol types were generated: (1) conventional (4 × 60 s), (2) ultrafast (20× 3s), and (3) combined (ultrafast followed by 3 × 60 s). Two readers independently or in consensus rated image quality (overall quality, artifacts, sharpness, lesion conspicuity, and morphology) and diagnostic confidence using a 5-point Likert scale. BI-RADS scores were compared to the final reference standard (histology or ≥2-year imaging follow-up). Agreement was assessed using Cohen’s kappa and ICC.
Results or Findings: A total of 167 patients (mean age: 59 ± 11 years) were included. Despite high temporal resolution of the ultrafast 4D-DCE, image quality was good to excellent and was comparable to the standard-resolution post-contrast T1 mDixon sequence (e.g., overall quality: 4.8±0.4 vs. 4.8±0.3, P=.99).The combined 4D-DCE protocol yielded the highest diagnostic confidence by BI-RADS assignment in both readers, with the most pronounced improvement observed in patients with high background parenchymal enhancement (e.g., reader 1: 3.2 ± 0.6 [conventional] vs. 4.2 ±0.4[ultrafast] vs. 4.9 ± 0.3[combined],P<.001). BI-RADS agreement with the final reference standard was good to excellent across all DCE protocols, with the highest agreement achieved using the combined 4D-DCE(e.g., reader 1: κ=0.89, 95% CI:0.84–0.95).
Conclusion: The proposed 4D-DCE technique enables robust breast DCE-MRI with high spatial and temporal resolution. Combining ultrafast and conventional acquisitions within a single protocol improves diagnostic confidence and BI-RADS agreement.
Limitations: Intraindividual study comparing different 4D-DCE approaches within the same examination
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Number 2025-317-BO, Ethic committee of the University Hospital Bonn

Author Block: L. Y. Chan, P. L. LAM, D. Fenn, W. P. CHEUNG, L. W. LO, W. S. Mak, K. M. L. WONG, E. P. Y. Fung; Hong Kong/HK
Purpose: Ultrafast dynamic contrast-enhanced MRI (UF-DCE MRI) is a novel sequence which captures contrast wash-in in breast lesions in the very early post-contrast period with high spatial and temporal resolution, allowing analysis of an early wash-in kinetic curve. This study aims to evaluate the correlation between kinetic parameters in UF-DCE MRI and histopathological characteristics of invasive breast cancers.
Methods or Background: This is a retrospective study performed in an acute general hospital. 49 consecutive female adult patients (median age, 54.2 years; range, 30.0-85.4 years) with 53 pathologically proven invasive breast cancers who underwent pretreatment MRI breast imaging were included from June 2023 to June 2025. The ultrafast MRI protocol consisted of a pre-contrast and 12 T1-weighted post-contrast high-temporal-resolution images in the first minute using TWIST-VIBE technique. Kinetic parameters of index breast cancers were obtained from ultrafast sequences, including time-to-enhancement (TTE), maximum slope (MS) and arterial-venous interval (AVI). TTE, MS and AVI among breast cancers with different histological grade (Modified Bloom-Richardson grade), ER/PR/HER2 positivity and tumour subtype were compared using the Mann-Whitney U test and Kruskal-Wallis test as appropriate.
Results or Findings: High histologic grade cancers had larger MS compared to low to intermediate grade cancers (high grade=14.9%/s vs low to intermediate grade=11.4%/s, p=0.049). ER-negative cancers showed shorter TTE compared to ER-positive cancers (ER-negative=4.5s vs ER-positive=9.0s, p=0.037). HER2-positive cancers showed shorter TTE than HER2-negative cancers (HER2-positive=4.5s vs HER2-negative=9.0s, p=0.032).
Conclusion: Ultrafast MRI–derived kinetic parameters TTE and MS were associated with histopathologic characteristics in invasive breast cancers. UF-DCE demonstrated potential to identify more aggressive breast cancers and provide prognostic value to guide breast cancer management.
Limitations: The limitations of the study are single-centre retrospective study with limited sample size.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Central Institutional Review Board (CIRB) of the Hospital Authority (HA) in Hong Kong Special Administrative Region (reference number: CIRB-2025-249-5). The requirement for informed consent was waived by CIRB.

Author Block: W. Yang, X-G. Peng; Nanjing/CN
Purpose: This study aims to develop predictive models for tumor-infiltrating lymphocyte (TIL) levels in breast cancer patients using multi-phase dynamic contrast-enhanced MRI (DCE-MRI) radiomics and semantic features from tumor and peritumoral regions.
Methods or Background: A total of 298 pathologically confirmed BC patients were stratified into low (<10%, n=102) and high (≥10%, n=196) TIL groups. Tumor and peritumoral volumes were manually delineated across six DCE-MRI phases using 3D-Slicer, with radiomics features extracted via FAE software. Ten machine learning algorithms were evaluated using five-fold cross-validation. Traditional imaging features were selected through univariate and multivariate logistic regression. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), and statistical comparisons were performed with DeLong’s test.
Results or Findings: 1,688 radiomics features were extracted from tumor and peritumoral VOIs. Among intratumoral models, DCE_Phase 5 (DCE_P5) showed the best performance (training AUC = 0.835; validation AUC = 0.714), and DCE_Phase 3 (DCE_P3) was optimal for peritumoral models (training AUC = 0.793; validation AUC = 0.703). The combined model using DCE_P5 and DCE_P3 achieved AUCs of 0.878 (training) and 0.772 (validation). In semantic feature analysis, axillary lymph node status (P = 0.017) and fibroglandular tissue content (P = 0.019) were significant predictors. The combined model using DCE_P5 tumor and semantic features achieved AUCs of 0.853 (training) and 0.706 (validation), while the DCE_P3 and semantic feature model reached AUCs of 0.792 and 0.705. The final integrated model achieved the highest performance (AUC_train = 0.881; AUC_valid = 0.773).
Conclusion: This study presents a noninvasive, clinically applicable model combining multi-phase DCE-MRI and semantic features to predict TIL levels in BC, aiding in immunotherapy decision-making.
Limitations: The retrospective study with a limited sample size may affect generalizability. External validation in larger cohorts is needed.
Funding for this study: This study has received funding by National Natural Science Foundation of China (82272064), Jiangsu Provincial Science and Technique Program (BK20221461), Zhongda Hospital Affiliated to Southeast University, Jiangsu Province High-Level Hospital Pairing Assistance Construction (zdlyg08), Postgraduate Research & Practice Innovation Program of Jiangsu Province (KYCX23_0323, and KYCX22_0297).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Ethics Committee of the Affiliated Zhongda Hospital of Southeast University (Approval Number: 2024ZDSYLL369-P01). All participants provided informed consent, and the study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki.

Author Block: N. Mesropyan¹, O. Weber², A. Isaak¹, H. Peeters³, C. Katemann², J. A. Luetkens¹; ¹Bonn/DE, ²Hamburg/DE, ³Best/NL
Purpose: To evaluate, a fully volumetric deep-learning(DL) super-resolution research prototype combining compressed sensing with denoising and resolution upscaling for 3D T2-weighted (T2w) breast MRI, and to quantify its impact on acquisition time, image quality, and BI-RADS agreement.
Methods or Background: This prospective single-center study was conducted on a 1.5-T MRI. Two Cartesian 3D T2w acquisitions with low and normal resolution were acquired and reconstructed with the DL framework to yield T2DL and T2NR+DL. Acquisition time, quantitative image quality (apparent signal-to-noise (aSNR), apparent contrast-to-noise (aCNR)) and qualitative scores (5-point Likert scale: overall quality, sharpness, conspicuity, morphology, artifacts) were obtained in randomized, blinded reading sessions. BI-RADS agreement between different protocol types with the final score were assessed. Kruskal–Wallis one-way analysis of variance followed by Dunn’s post hoc test, intraclass correlation coefficients (ICCs), and Cohen k test were used for statistical analysis.
Results or Findings: 64 women (mean age 55±13 years; range 28–84) were investigated. Acquisition time decreased by 41% for T2DL vs T2NR (126 vs 215 seconds). Quantitative quality improved with DL (e.g., for aSNR: 28.5±7.3 [T2NR] vs 36.6±10.9 [T2DL] vs 45.7±10.3 [T2NR+DL]; for aCNR: 21.8±5.6 [T2NR] vs 25.9±8.6 [T2DL] vs 35.3±9.1 [T2NR+DL]; all P<.001). Qualitative image quality was highest for T2NR+DL (for reader 2: overall quality: for T2NR+DL 5 [IQR 5–5] vs 4 [4-5] for T2 DL vs. 4 [3–4] for T2NR; P<.001). Inter-reader agreement for image quality assessment was good to excellent, ICCs rages 0.79–0.96 (CI:0.65-0.98). BI-RADS agreement was strong across all 3D DL protocols, Cohen k ranges 0.98–0.99 (CI: 0.98-0.99).
Conclusion: This volumetric DL super-resolution approach enables faster 3D T2w breast MRI while enhancing image quality. Alternatively, at unchanged acquisition time, it can further improve the quality of normal-resolution images.
Limitations: Research prototype was used in this study.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Approved by the ethic committee of the University Hospital Bonn.

Author Block: A. Zeynalova, G. Esen, A. DOLU, M. Akin, N. Önder Denizoğlu, F. Tokat, C. Uras; Istanbul/TR
Purpose: To evaluate the association between molecular subtypes and magnetic resonance imaging (MRI) findings in ductal carcinoma in situ (DCIS), with a focus on identifying MRI features characteristic of the HER2-positive subtype.
Methods or Background: This retrospective observational study included 55 female patients who underwent preoperative breast MRI and had pure or microinvasive DCIS confirmed by surgical pathology between 2010 and 2024. Lesion size, lesion type, nipple extention, morphology and kinetic features were evaluated by two experienced breast radiologists in consensus. Age, clinical presentation, lesion location, tumour grade, necrosis, tumour size, surgical margin status, axillary lymph node involvement, receptor status were recorded. Receptor status was re-evaluated by a pathologist if initially unavailable.
Results or Findings: HER2 positivity was detected in 43.6% of cases. HER2-positive DCIS showed a larger mean MRI lesion size (44.7 mm, p=0.004) and more frequent nipple extension (p=0.032). ER negativity was higher in HER2-positive cases (p=0.003), and high nuclear grade was more common (p=0.052). Receiver operating characteristic (ROC) analysis identified a 22 mm lesion size threshold for predicting HER2 positivity (AUC=0.740). MRI lesion size, nipple extension, ER negativity and high nuclear grade were independent predictors of HER2 positivity, while no other imaging or pathological variables showed significant differences.
Conclusion: Recent studies have shown that HER-2 positivity may be associated with a worse prognosis and more frequent progression to invasive disease in DCIS. There is very limited data in literature about MRI features of different molecular subtypes of DCIS. Our results show that MRI features, particularly large lesion size and nipple extension are associated with HER2 positivity in DCIS and may aid molecular subtype prediction and surgical planning.
Limitations: The limitations of the study are its retrospective single-centre design and the small sample size.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The ethics committee notification can be found under the number 2023-21/704.

Author Block: A. h. Shah, A. Chandra, S. Sen, P. Ghosh; Kolkata/IN
Purpose: Axillary nodal status is critical in breast cancer staging, yet MRI assessment relies on subjective criteria such as cortical thickening or hilum loss, which are only moderately accurate. We evaluated whether radiomics of axillary nodes visible on routine breast MRI can provide a reproducible, objective predictor of nodal metastasis.
Methods or Background: We retrospectively analyzed 212 women with biopsy-proven breast cancer who underwent preoperative breast MRI (2013–2024) and subsequent nodal sampling or dissection. The most conspicuous axillary node per patient was segmented on post-contrast T1- and T2-weighted sequences. PyRadiomics extracted first-order, shape, and texture features. Machine-learning models were trained to classify nodal status and compared against conventional MRI descriptors (short-axis >10 mm, cortical thickness >3 mm, hilum loss).
Results or Findings: Of 212 patients, 118 had metastatic and 94 benign nodes. Radiomics features reflecting shape irregularity, enhancement heterogeneity, and cortical entropy were most predictive. The radiomics model achieved an AUC of 0.87, outperforming conventional MRI criteria (AUC 0.71, p<0.01). Sensitivity and specificity were 82% and 80%, respectively. Radiomics also improved prediction of extranodal extension (AUC 0.79 vs 0.62). Interobserver reproducibility was excellent for radiomics scores, compared with only moderate agreement for visual assessment (κ=0.46).
Conclusion: Radiomics of axillary nodes on standard breast MRI provides a reproducible, quantitative biomarker of nodal metastasis, clearly outperforming conventional visual descriptors. This is the largest dedicated study to date demonstrating that data already embedded in routine breast MRI can be unlocked to refine nodal staging, with direct implications for surgical and systemic treatment planning.
Limitations: Single-center retrospective design; segmentation restricted to one node per patient; no external validation.
Funding for this study: None
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: