EIBIR 2 - EIBIR Stage bonus session 2

Author Block: Y. Wang, Y. Xiao; Shanghai/CN
Purpose: This prospective study evaluated the feasibility of chemical exchange saturation transfer (CEST) MRI for assessing disease activity in thyroid-associated ophthalmopathy (TAO).
Methods or Background: A total of 88 patients with active TAO, 76 with inactive TAO, and 30 healthy controls were enrolled. CEST MRI-derived MTR and MTRasym at 1ppm, 2ppm, and 3.5ppm were calculated. Clinical data, MTR and MTRasym of extraocular muscles were compared across groups using the Wilcoxon signed rank test. Spearman's correlations were used to examine the associations between imaging parameters and the CAS. Logistic regression analysis was carried out to identify independent predictors, and ROC analysis (DeLong's test) was used to evaluate diagnostic performance for active TAO.
Results or Findings: Patients with active TAO showed lower MTR (P<0.001) and higher MTRasym_1ppm, MTRasym_2ppmand MTRasym_3.5ppm ( all P<0.001). MTR was negatively correlated with CAS (R=-0.402; P<0.001), while MTRasym_1ppm, MTRasym_2ppm and MTRasym_3.5ppm (R=0.369; R=0.350; R=0.349; all P<0.001) were positively correlated. Both MTR and MTRasym_1ppm were independent predictors of TAO activity. The AUC for MTR and MTRasym_1ppm in discriminating active from inactive TAO were 0.772 and 0.730, respectively. Combining MTR with MTRasym_1ppm significantly improved diagnostic performance, achieving an AUC of 0.805 ( P<0.05).
Conclusion: Our results showed MTR and MTRasym_1ppm independently distinguished active from inactive TAO. Their combination further enhanced diagnostic accuracy. These findings suggested that MTR and MTRasym_1ppm could serve as quantitative imaging biomarkers to guide treatments in patients with TAO.
Limitations: Firstly, it is a prospective single-center study. Secondly, the complex structure of the orbit, and the high risk associated with invasive surgery precluded the collection of histopathological samples, preventing direct confirmation of the relationship between GAGs/collagen content and MTRasym_1ppm of CEST in TAO patients
Funding for this study: This study has received funding from the National Natural Science Foundation of China [No. 82271994]; the Military Commission health care special project [NO. 22BJZ07]; the Shanghai Hospital Development Center [NO. SHDC22025311-A]; The Eastern Elite Program Leading Project of Shanghai Municipality [NO.LJ2023094] Navy Medical University teaching achievement cultivation project [No. JPY2022B15]; Shanghai Changzheng Hospital teaching achievement cultivation project [No. JXPY2021B10];
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: No.82170858

Author Block: A. Holden¹, R. Bradley¹, K. Rhode²; ¹Oxford/UK, ²London/UK
Purpose: This project aims to establish a standardised testing procedure for the emerging technology of contrast-to-noise driven exposure controls (CEC) as seen on the Siemens ARTIS Icono. Additionally, the project will compare the performance of ARTIS Icono with that of its predecessor, namely the Siemens ARTIS Zee.
Methods or Background: The Siemens ARTIS Icono is an advanced fluoroscopy system used in cardiac catheterisation laboratories, where high-quality imaging is essential for precise interventions, while ensuring patient dose is kept as low as reasonably practicable. Unlike conventional dose-driven automatic exposure control (DEC) systems that maintain a constant detector dose, the ARTIS Icono employs OPTIQ©, a CEC where the system selects imaging parameters based on a predefined target image quality. Additionally, Structure Scout enables material-specific optimisation.
Results or Findings: A phantom has been designed to test the systems capability, containing a range of reference materials, including iron, platinum, and iodine. Using this phantom we can establish if the system maintains the target CNR with varying parameters. Additionally, a figure of merit has been established to compare a metric of visibility to one of dose. By conducting measurements on both CEC and DEC systems we will be able to assess the success of CEC. Preliminary results suggest an increase in image quality at comparable doses.
Conclusion: CEC presents an opportunity to improve image quality in fluoroscopic procedures while minimising patient dose. This could lead to a major breakthrough in optimising patient dose. However, there is currently no established method available for conducting quality assurance on the system. This project demonstrates a successful possible testing method for sites using the ARTIS Icono
Limitations: Static phantom study. No measurements of effect of movement.
Funding for this study: King's College London, School of Biomedical Engineering and Imaging Sciences
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: F. Orzan¹, L. Dioșan², Z. Bálint¹; ¹Cluj Napoca/RO, ²Cluj-Napoca/RO
Purpose: We aimed to build an AI-based decision support system for automatic segmentation and characterization of MS lesions from 3D MRI, with the long-term goal of personalized treatment. This study focuses on preprocessing and segmentation, providing the basis for future characterization and clinical integration.
Methods or Background: Our study evaluated a 3D U-Net segmentation model with four encoding-decoding levels, integrating attention gates and skip connections to improve lesion localization. The architecture employs BatchNorm3d, ReLU activations, and Dropout for regularization. Evaluation was conducted on 40 cases from the MICCAI 2021 dataset. Preprocessing included intensity normalization, isotropic resampling, N4 bias field correction, skull stripping, and rigid registration. Training used Adam optimization with BCE loss over 50 epochs, with early stopping, 5-fold cross-validation, and an 80/20 train-validation split.
Results or Findings: This study focuses exclusively on image preprocessing, model development, and lesion segmentation. Textural analysis, lesion classification, and the development of a user interface for fine-tuning will be addressed in future work.
The model showed a steady decrease in both training and validation loss, with stabilization after epoch 40. The close alignment between losses suggests minimal overfitting and good generalization - an essential property for clinical applications. The Dice score increased consistently, reaching 0.78 on training and 0.72 on validation data, confirming the model’s ability to accurately segment MS lesions across diverse samples.
Conclusion: Our U-Net model with attention gates and skip connections shows promising performance for automated MS detection on 3D MRI. Future work will add textural analysis, lesion classification, user interface development, and validation on multi-vendor datasets for robustness and scalability.
Limitations: A key limitation of MS detection algorithms, including our study, is the scarcity of large imaging datasets, which limits performance relative to human experts.
Funding for this study: Not applicable
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: C. Verde, L. Tarotto, S. Stilo, F. Fiore; Naples/IT
Purpose: Our purpose is to describe clinical use of Artificial Intelligence (AI) in the current and future practice of Interventional Radiology (IR).
Methods or Background: AI has demonstrated great potential in a wide variety of applications in IR.
Support for decision-making and outcome prediction, new functions and improvements in fluoroscopy, ultrasound, computed tomography, and magnetic resonance imaging have all been investigated.
Furthermore, AI represents a significant boost for fusion imaging and simulated reality, robotics, touchless software interactions, and virtual biopsy.
The procedural nature, heterogeneity, and lack of standardisation slow down the process of adoption of AI in IR. The full range of possibilities is yet to be explored.
A systematic review of the bibliography from the earliest possible date through March 2024 was performed.
Results or Findings: Perspectives in AI differ and are more complex for IR than for diagnostic radiology because IR encompasses diagnostic imaging, imaging guidance, and early imaging evaluation as well as therapeutic tools.
The integration of artificial intelligence in IR is an emerging field with transformative potential, aiming to make a great contribution to the health domain.
Based on the overview, the integration of AI in IR presents significant opportunities to enhance precision, efficiency, and personalization of procedures. AI automates tasks like catheter manipulation and needle placement, improving accuracy and reducing variability. It also integrates multiple imaging modalities, optimizing treatment planning and outcomes. AI aids intra-procedural guidance with advanced needle tracking and real-time image fusion.
Conclusion: The integration of these technologies makes interventional radiology more personalized, patient-centered and safe.
Limitations: No limitations were identified in this study.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: S. Kolahi, M. Shakiba, S. Rahmani, S. Nosrat Sheybani, D. Seyedi, H. Abdelmalik, S. Parviz, M. Malek, M. Tahamtan; Tehran/IR
Purpose: Diffuse idiopathic skeletal hyperostosis (DISH) is a systemic condition characterized by ligamentous ossification along the spine. While its prevalence has been well described in the general population, data on its occurrence in oncology patients remain limited. This study aimed to assess the prevalence and distribution of DISH and early-phase DISH in newly diagnosed cancer patients undergoing initial staging with Computed Tomography (CT).
Methods or Background: In this retrospective cross-sectional study, 1,053 adult oncology patients who underwent thoraco-abdominopelvic CT for initial staging were evaluated. DISH and early-phase DISH were diagnosed using established radiologic criteria. Vertebral body densities were measured, and associated extraspinal enthesopathies and ligamentous ossifications were documented.
Results or Findings: DISH was present in 30.3% of patients, including 13.8% with established DISH and 16.5% with early-phase DISH. Prevalence was higher in older patients and males (p < 0.01). Notably, renal (43.2%), gastric (37.5%), and colorectal (33.7%) cancers demonstrated significantly higher DISH rates, whereas esophageal cancer showed a lower prevalence (13.4%). DISH was associated with decreased vertebral bone density and frequent extraspinal enthesopathies. No significant correlations were found with BMI, diabetes, or hypertension.
Conclusion: DISH is common among oncology patients and often coexists with extraspinal enthesopathies and reduced bone density. These findings suggest possible shared pathogenic mechanisms and underscore the importance of further studies exploring the relationship between DISH and malignancy.
Limitations: The retrospective design introduced potential selection and information biases, as reliance on medical records limited access to comprehensive clinical histories and prior treatments, which may have confounded our results, including the DISH prevalence estimate and associations with enthesopathies and bone density.
Unbalanced cancer subgroups with small sample sizes limit the generalizability of some analyses.
The lack of serum biomarker data (e.g., FGF-23, DKK-1) precluded direct mechanistic investigation.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The research protocol was approved by our Institutional Review Board (Approval Code: IR.TUMS.IKHC.REC.1404.019).

Author Block: S. Halder, K. Chaudhury, S. Mandal; Kharagpur/IN
Purpose: Photoacoustic (PA) imaging, when combined with ultrasound (US), offers a powerful hybrid modality providing high-resolution anatomical details and functional information such as oxygen saturation and tissue perfusion. However, challenges in delineating tissue boundaries, reconstruction, and functional quantification have limited its translational potential. This study aims to integrate deep learning to enhance segmentation accuracy while incorporating prior information from co-registered ultrasound images to enable accurate quantitative analysis and model optical fluence distribution in PA images.
Methods or Background: A multi-modal imaging framework was developed, combining anatomical US data with functional PA signals. Deep learning models such as Unet and nnUnet were trained to segment anatomical and minute vascular structures from US images of a mouse kidney. Evaluation metrics focusing on segmentation accuracy were co-validated using Doppler images. These segmentation priors were used to quantify the perfusion profiles in the intra-organ and inter-organ anatomical structures. Further, these priors have assisted in correcting the fluence in the PA images.
Results or Findings: The deep learning pipeline achieved precise anatomical segmentation with very high Dice coefficients of more than 90%, enabling clearer visualization of soft tissue boundaries. Functional imaging revealed accurate mapping of blood perfusion and oxygenation parameters, with strong agreement compared to Doppler-based measurements. Incorporation of multimodal priors in the pipeline reduced quantitative errors and could accurately estimate the fluence in the PA images, even at depths exceeding several centimetres.
Conclusion: The combination of deep learning–driven segmentation and prior-informed imaging pipeline could significantly enhance the performance of US-PA imaging, improving both anatomical and functional information.
Limitations: Current validation is restricted to preclinical studies. For broader adoption (Magnetic Resonance Imaging) MRI images could be integrated in this pipeline.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The animal ethical had been taken for the studies.

Author Block: X. Zhang, H. F. Zhao, H. Zhang; Lanzhou/CN
Purpose: This study aims to predict ΔKi-67 after NAT in breast cancer patients using radiomic features from DCE-MRI.
Methods or Background: Ki-67 is a key biomarker of tumor proliferation in breast cancer. A reduction in Ki-67 following neoadjuvant therapy (NAT) reflects chemosensitivity and holds significant prognostic value. Therefore, pre-treatment assessment of Ki-67 dynamics during NAT is crucial for evaluating patient prognosis. This retrospective study of 148 patients (7:3 training/test split) aimed to predict ΔKi-67 (post- minus pre-NAT Ki-67), with the response cutoff determined by ROC analysis. Independent clinical predictors were identified using multivariable logistic regression (P<0.05). Radiomics features were extracted from pre-treatment DCE-MRI scans from the early, peak, and delayed phases, along with the corresponding phase differences (delayed–early, delayed–peak, peak–early). Following feature selection with PCA and RFE, models were built using SVM. The optimal radiomics model selected by DeLong test was combined with clinical factors into a hybrid model, evaluated via AUC, calibration, and DCA.
Results or Findings: In the testing cohort, the peak-to-early delta-radiomics model (10 features) achieved the best performance (AUC=0.802, 95% CI: 0.666–0.938), significantly outperforming the delayed-to-early (AUC=0.679, 95% CI: 0.519–0.839) and peak-phase (AUC=0.625, 95% CI: 0.458–0.793) models. Integration with significant clinical predictors (HER2 status, histological grade) yielded a combined model with a superior AUC of 0.858 (95% CI: 0.753–0.965), significantly exceeding both the clinical (AUC=0.785, 95% CI: 0.645–0.925) and radiomics models alone (p<0.001).
Conclusion: The integration of MRI delta-radiomics and clinical parameters offers a superior, non-invasive method for predicting Ki-67 downstaging after NAT, aiding personalized strategies and prognostic evaluation.
Limitations: (i) Single-center, retrospective nature and small sample limit generalizability, mandating multi-center prospective validation; (ii) Subtype-specific analyses were absent, warranting future validation; (iii) Semi-automated VOI delineation risks inter-observer bias, addressable by deep learning-based auto-segmentation.
Funding for this study: Not applicable
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: No.LDYYLL2025-914