RPS 402 - Advances in breast imaging: innovations shaping the future of cancer care

Author Block: J. Ferda, E. Ferdova, T. Barakova, S. Vokurka; Plzen/CZ
Purpose: 18F-fluoroestradiol is a novel radiopharmaceutical useful in breast carcinoma, the indications in clinical scenarios are under development. The purpose of the study is to assess the clinical impact of the imaging of the breast carcinoma with estrogen-positive receptors (ER+) using 18F-fluoroestradiol (18F-FES) PET/CT or PET/MRI according to its impact to the treatment decision making. The study is concerned in the different preference of PET/CT and PET/MRI in the staging and restaging.
Methods or Background: 40 patients with estrogen positive breast carcinoma underwent the hybrid imaging after intravenous application of 18F-FES, in 25 cases it was used PET/CT, in 15 cases PE/MRI. The radiopharmaceutical was injected in the activity of 2,5 MBA/kg and the imaging was performed the imaging. In 10 patient PET/MRI was used as restaging method, PET/MRI was performed in the 5 cases of the staging before surgery with targeted full diagnostic MRI imaging of the breast in prone position, followed by the trunk imaging in supine position, in five to seven steps. All PET/MRI were performed with the gadolinium contrast material, the imaging included brain imaging in T1 STARVIBE. PET/CT was performed using the continuous PET acquisition after CT with the administration of the iodinated contrast material, in 5 cases was performed in staging, in 20 cases in rest-aging
Results or Findings: The most important information was detection of ER+ metastases when 18F-FDG-PET was negative (12x) - including brain and liver metastases, the persistent ER+ of the metastases (7x), staging of the disease (10x), the loss of the ER (4x) and the negative finding for metastases (2), no added information was found in 5 examinations.
Conclusion: 18F-FES-PET provided the important clinical information to treatment strategy, PET/MRI improves the imaging of brain and liver.
Limitations: Small sample
Funding for this study: No
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: according to the Helsinky declaration

Author Block: L. Corradini, D. Ballerini, A. Bonanomi, E. D'Ascoli, G. Della Pepa, C. De Berardinis, E. Ancona, C. Depretto, G. P. Scaperrotta; Milan/IT
Purpose: Lesion conspicuity, defined as the "degree of enhancement" relative to the background, was the focus of this retrospective monocentric study, which aimed to explore its correlation with malignancy of lesions, histology, receptor profile and grading in breast cancer patients.
Methods or Background: Two breast radiologist and one radiology resident evaluated all CEDM performed in our oncological center from January 2023 to April 2024, assigning degrees of conspicuity to breast lesions, and evaluating a possible correlation with Ki-67 values (≤ 20% or > 20%), HER-2 status, estrogen (ER) and progesterone (PGR) receptor positivity, molecular subtype, and histological grade. Statistical analysis employed the Cramer’s V test.
Results or Findings: Out of 352 patients included (median age=54, IQR=18), 100 were excluded due to chemotherapy controls and 53 due to B3 lesion. In the 199 remaining patients we observed a moderate to strong association between conspicuity and ER expression (V=0.534) and Ki-67 value (V=0.36). A moderate association was found between conspicuity and PGR expression (V=0.31). No significant correlation was noted between conspicuity and histological grade (V=0.184) or HER2 status (V=0.2).
Conclusion: Conspicuity, a recently incorporated descriptor in CEDM BI-RADS lexicon, was validated by our findings, which are in line with the initial evidence in the literature.
Limitations: Retrospective monocentric study with limited number of patients.
Conspicuity is a subjective descriptor, potentially introducing variability in the data and affecting the findings.
Funding for this study: None
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: None

Author Block: M. Rivlin¹, R. Sivan Hoffmann², V. Hadar², S. Sukhotnik², N. E. Weisenberg², O. Shmain-Naydenov², M. Zaiss³, S. Weinmüller³, G. Navon¹; ¹Tel Aviv/IL, ²Kfar Saba/IL, ³Erlangen/DE
Purpose: Molecular imaging with 18F-fluorodeoxyglucose positron emission tomography (18FDG-PET) is a powerful and well-established tool in breast cancer management, as increased glucose uptake is a known cancer hallmark.However, it carries the risk of repeated radiation exposure.
We have recently discovered that glucosamine (GlcN), a non-toxic, biocompatible glucose-based material can be detected using a unique MRI contrast mechanism termed chemical exchange saturation transfer (CEST). CEST has emerged as an attractive molecular imaging approach for providing valuable metabolic information. Our goal is to develop an innovative molecular imaging modality based on CEST-MRI of GlcN to visualize and measure breast tumors while also distinguishing between benign and malignant tumors.
Methods or Background: Breast cancer patients and control group were scanned using the CEST-MRI pulse sequence on a 3T scanner (VIDA, Siemens, Germany) equipped with breast coil. The protocol included CEST scans before and two hours after drinking a solution of GlcN (184 mg/kg). The data were evaluated using magnetization transfer asymmetry ratio (MTRasym) and area under curve (AUC) analysis.
Results or Findings: GlcN treatment resulted in higher CEST values in tumor regions of interest (ROIs), with maximum net MTRasym signal (at 2 ppm) of 6.3±2.6% and averaged AUC (2-5 ppm) increase ratio of 3.3±2.1% (N=7). Yet, no significant GlcN CEST signal enhancement was detected in healthy volunteers (N=9). GlcN CEST signal values were highly correlated with the BI-RADS category.
Conclusion: These findings suggest that the GlcN CEST MRI technique can detect breast cancer while also providing molecular-level diagnostic tools for discriminating between benign and malignant breast tumors. These promissing results pave the way to a future metabolic imaging of additional diseases.
Limitations: Due to the size of the study group we were not able to clarify into sub-groups.
Funding for this study: Funding by the ISF No. 1689/18
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: approval number MMC0201-21

Author Block: D. Förnvik¹, S. Zackrisson¹, I. Skarping²; ¹Malmö/SE, ²Lund/SE
Purpose: To predict neoadjuvant chemotherapy (NACT) treatment response by applying artificial intelligence (AI) to digital mammography (DM) and serial breast tomosynthesis (DBT) images. Explainable AI (XAI) for enhanced clinical credibility is explored.
Methods or Background: NACT for early-stage breast cancer (BC) has recently become an attractive approach to patients who are eligible for chemotherapy. MRI is the imaging modality of choice for evaluating tumor response but not as readily available as DM and lately DBT. Nevertheless, predicting residual cancer, as assessed by radiologist, after NACT using imaging is challenging; thus, AI could be an alternative. Between 2005/2014 - 2019, 453 (DM) and 149 (DBT) patients, respectively, at Skane University Hospital, Sweden, comprised the cohorts. Two deep learning architectures (DM: ResNet24, DBT: backbone 3D ResNet) applied to images from both the cancer and contralateral healthy breasts acquired at three time points: pre-NACT (DM and DBT), mid-NACT (DBT) and post-NACT (DBT) were used to predict pathological complete response (pCR). For DBT, GradCAM was used to produce saliency maps to obtain insights into the model-based decisions.
Results or Findings: The DM and the DBT AI models predicted pCR as represented by the area under the ROC curve of 0.71 (95% CI: 0.53–0.90; p = 0.035) and 0.83 (95% CI: 0.63–1.00; p = 0.008), respectively. The spatial correlation of saliency maps for DBT volumes from the same patient but at different time points was high, likely indicating that the model focuses on the same areas during decision-making.
Conclusion: The DBT model demonstrates a high discriminative performance for predicting pCR/non-pCR, possibly outperforming radiologists' assessment.
Limitations: Availability of larger datasets and inclusion of clinicopathological variables would permit more comprehensive training of the models and more rigorous evaluation of their prediction performance for future patients.
Funding for this study: Swedish Breast Cancer Group (BRO), Allmänna Sjukhusets i Malmö Stiftelse för bekämpande av cancer, and the Governmental Funding of Clinical Research within the National Health Services.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Regional Ethics Committee in Lund, Sweden (committee reference numbers: 2014/13, 2014/521, and 2016/521, and 2021-05637-02).

Author Block: H. S. Ahn, S. H. Kim, M. J. Hong, H-S. Lee; Seoul/KR
Purpose: To evaluate the image quality and diagnostic value of single-shot echo-planar imaging (ss-EPI) with deep learning reconstruction (DLR) versus simultaneous multi-slice echo-planar imaging (SMS rs-EPI) in breast MRI.
Methods or Background: This study included 77 cases of breast cancer from 74 patients who underwent preoperative breast MRI. All patients underwent breast MRI that included the ss-EPI sequence combined with post-processing using DLR, as well as the SMS rs-EPI sequence. Two radiologists independently assessed qualitative image quality factors and determined their preferences, while the cancer detection rate (CDR) was calculated. Quantitative analysis included measurements of apparent diffusion coefficient (ADC), signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and lesion contrast, including phantom measurements.
Results or Findings: Regarding qualitative image quality parameters, ss-EPI with DLR demonstrated significantly higher scores in fat suppression and overall image quality as assessed by both radiologists, with comparable scores in artifact presence and lesion conspicuity to the SMS rs-EPI sequence. The CDR showed no significant difference between the two sequences. Both radiologists preferred ss-EPI with DLR (Reader 1: 78.4%, Reader 2: 89.2%). For quantitative parameters, ss-EPI with DLR exhibited significantly higher CNR (p = 0.002) and lesion contrast (p < 0.001), while ADC and SNR values were comparable. In phantom measurements, mean ADC was lower for ss-EPI with DLR (DLR: 1.08 ± 0.58 vs. SMS: 1.12 ± 0.59, p = 0.007), but SNR was not significantly different (DLR: 607.45 ± 346.1 and SMS: 630.03 ± 624.51, p = 0.911). The acquisition time was shorter for ss-EPI with DLR (2:06 min) compared to SMS rs-EPI (3:29 min).
Conclusion: ss-EPI with DLR provided superior image quality and greater reader preference compared to SMS rs-EPI.
Limitations: This is a retrospective study which performed at single center.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: None

Author Block: M. Balbino¹, M. Montatore², F. Masino³, F. A. Carpagnano⁴, G. Capuano², G. Guglielmi⁵; ¹Triggiano/IT, ²Barletta/IT, ³Bari/IT, ⁴Foggia/IT, ⁵Andria/IT
Purpose: This study, one of the first in Italy, aims to evaluate the efficacy of performing CT and CEM consecutively using the same contrast medium in a single imaging session. The goals include reducing the amount of contrast agent injected into oncological patients and enhancing the detection of metastases in various districts through CT, while also providing a precise diagnosis of breast extension and identifying additional foci within the breasts through CEM.
Methods or Background: A cohort of female patients with confirmed primary breast cancer and suspected metastatic disease were enrolled. Each patient underwent a CT scan followed immediately by a CEM using the same contrast medium. The CT was performed to identify visceral metastases, while the CEM targeted the detection of additional breast lesions and regional lymph node involvement. Both imaging modalities utilized iodine-based contrast agents, administered intravenously. The diagnostic outcomes were compared with those from conventional imaging techniques, including standard mammography, ultrasound, and MRI.
Results or Findings: The combined CT and CEM approach demonstrated a higher sensitivity and specificity in detecting metastatic sites compared to traditional imaging methods. In particular, CEM revealed additional lesions in the breast and regional lymph nodes that were not identified by CT alone. The concurrent use of the same contrast medium was found to be safe and well-tolerated, with no significant increase in adverse reactions. The integrated imaging protocol provided comprehensive anatomical and functional information, leading to more accurate staging and better-informed treatment decisions.
Conclusion: The integration of CT and CEM using the same contrast agent offers a promising advancement in the diagnostic imaging of metastatic breast cancer. This combined approach enhances the detection of metastatic lesions, providing a more comprehensive assessment of disease spread.
Limitations: No
Funding for this study: No
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: No

Author Block: M. Bobowicz, M. Kosno, K. P. Brzozowski, M. Rygusik; Gdańsk/PL
Purpose: The breast density visual assessment in mammography is a subjective process prone to errors but impacting diagnostic decisions. To overcome this problem, we developed a robust and reliable AI model that employs convolutional neural network-based transfer learning, specifically ResNet, DenseNet, and EfficientNet architectures, to predict breast density. Our research benefits from the Tree-structured Parzen Estimator (TPE) algorithm, an advanced tool for hyperparameter optimisation.
Methods or Background: A dataset of 2101 digital MLO mammography images performed at the Medical University of Gdansk from 2014 to 2022 was selected for analysis. The images were acquired using various devices from SIEMENS, GE HEALTHCARE, and HOLOGIC to ensure a high degree of image characteristics variability. The dataset was divided into 80% training and 20% validation sets. ResNet, DenseNet, and EfficientNet architectures were trained using the TPE algorithm. The assembly model comprises three five-fold cross-validated convolutional networks.
Results or Findings: An ensemble model resulted in good performance metrics: AUC-ROC (0.99), accuracy (0.91), F1-score (0.91), and recall values (0.90) for the test dataset. The TPE algorithm facilitates the development of high-performance models on a relatively small dataset, eliminating the need for image segmentation to extract the skin and pectoral muscle opacities, which is challenging to implement and often burdened with significant errors.
Conclusion: Our methodology enables the straightforward training of a robust model that can provide highly precise breast density predictions, reducing and automating the burden of required density reporting. Furthermore, our findings demonstrate the efficacy of advanced hyperparameter numerical optimisation methods in enhancing the efficiency of transfer deep learning models in the context of breast density prediction.
Limitations: The study's limitations are its relatively small dataset, single-centre design, and lack of external validation.
Funding for this study: Funding was provided by the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952103 (EuCanImage project) and was co-funded by the Digital Europe programme under grant agreement No 101100633 (EUCAIM project).
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: This retrospective study uses fully anonymised data from the EuCanImage project under the global ethics committee agreement for MUG.