Research Presentation Session: Breast

RPS 102 - Breast imaging biomarkers

February 28, 08:00 - 09:00 CET

7 min
MRI radiomics analysis predicting early recurrence in breast cancer patients who are candidates for neoadjuvant chemotherapy
Charlotte Marguerite Lucille Trombadori, Rome / Italy
Author Block: C. M. L. Trombadori, A. D'Angelo, E. Boccia, L. Boldrini, G. Franceschini, D. Giannarelli, A. Franco, A. Fabi, P. Belli; Rome/IT
Purpose: The aim of this study was to assess the role of pure radiomic predictive models and combined models with clinical/radiological variables applied to Magnetic Resonance Imaging (MRI) in predicting early recurrence (ER: disease-free survival <3 years after surgery) in breast cancer patients undergoing neoadjuvant chemotherapy (NAT). Identifying tools for non-invasive pre-treatment predictors of clinical outcomes, particularly recurrence, is necessary for better patient stratification and treatment selection.
Methods or Background: Patients with breast cancer who underwent staging MRI, NAT, and surgery at our centre (from 2012 to 2021) were included. Clinical variables evaluated included pathological complete response, ER, and tumour subtype. Radiological variables included tumour response according to RECIST criteria. Four breast radiologists reviewed MRI, annotated regions of interest, and extracted radiomic features. Pure radiomic models and combined models (clinical-radiological, radiological-radiomic, and clinical-radiomic) were developed. The area under the curve (AUC) was calculated for each model, and the models were compared in terms of accuracy, sensitivity, and specificity.
Results or Findings: A total of 211 patients were included, with an ER prevalence of 11.34%. Patients with complete or partial response to NAT and Luminal tumour subtype had a lower likelihood of developing ER (p = 0.001 and p = 0.037, respectively). Two radiomic features were statistically significant associated with ER: and F_cm_2.5D.joint.entr. The AUC values for combined models were 0.77 (radiological-radiomic model), 0.68 (clinical-radiomic model), and 0.74 (clinical-radiological model). The radiological-radiomic model was significantly more accurate in predicting ER than the pure radiological and radiomic models (p<0.001 and p<0.03, respectively).
Conclusion: The radiological-radiomic model, combining radiomic features and RECIST criteria, showed the most promising results in predicting ER.
Limitations: The small sample size and monocentric nature of the study were identified as limitations.
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: This study was approved by an ethics committee; ID cineca: 6081.
7 min
PET/CT radiomics integrated with clinical indexes as a tool to predict Ki67 in breast cancer: a pilot study
Cong Shen, Xi'an / China
Author Block: C. Shen1, Y. Liao2, X. Yu2; 1Xi'an/CN, 2Shanghai/CN
Purpose: This study aims to assess the predictive value of radiomics features extracted from 18F-FDG PET/CT, in combination with clinical characteristics, for estimating Ki67 expression in patients with breast cancer.
Methods or Background: A total of 114 patients diagnosed with breast cancer and examined using 18F-FDG PET/CT were included. Patients were randomly assigned to a training set (n = 79, including 55 cases of Ki67+ and 24 cases of Ki67-) and a validation set (n = 35, comprising 24 cases of Ki67+ and 11 cases of Ki67-). Thirteen clinical characteristics and 704 radiomics features were extracted. Feature selection involved univariate logistic analysis, Max-Relevance and Min-Redundancy, least absolute shrinkage and selection operator regression, and Spearman test. Three models were developed, including the clinical model, the radiomics model, and the combined mode. Model performance was evaluated using receiver operating characteristic (ROC) curve, and clinical utility was assessed through decision curve analysis (DCA).
Results or Findings: The N stage, tumour morphology, maximal standard uptake, and the longest diameter were significantly differed between Ki67+ and Ki67- groups (all P<0.05). Seven radiomics features were selected for the radiomics model. The area under the ROC curve (AUC) of the combined model in the training and test group was 0.90 (95% CI: 0.82–0.97) and 0.81 (95% CI: 0.64–0.99), respectively. The combined model significantly outperformed both the radiomics model and the clinical model alone (both P<0.05). The DCA curve demonstrated the superior clinical utility of the combined model compared to the clinical model and radiomics model.
Conclusion: PET/CT image-based radiomics features combined with clinical features have the potential to predict Ki67 expression in BC.
Limitations: The retrospective nature of the study and its small sample size were identified as limitations.
Funding for this study: This study was funded by the National Natural Science Foundation of China (No. 82272073), the Key Research and Development Plan of Shaanxi Province (2023-YBSF-480), and the Natural Science and basic research project of Shaanxi Province (2023-JC-QN-0903).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This retrospective study was conducted at the First Affiliated Hospital of Xi 'an Jiaotong University (NCT05826197), and the study protocol was approved by the Ethics Committee of Xi 'an Jiaotong University (IRB-SOP-AF-16).
7 min
Automatic breast segmentation-based radiomics for classifying breast composition and detecting neoplastic lesions on chest CT
Giridhar Dasegowda, Little Rock / United States
Author Block: G. Dasegowda1, M. Frölich2, S. Dalal3, S. R. Digumarthy3, P. Kaviani3, L. Karout3, R. Fahimi3, E. Garza Frias3, M. K. Kalra3; 1Little Rock, AR/US, 2Munich/DE, 3Boston, MA/US
Purpose: The aim of this study was to evaluate if automatic breast segmentation-based radiomics can differentiate between benign and malignant breast lesions and classify the breast based on tissue composition on contrast-enhanced chest CT.
Methods or Background: Our retrospective study included 882 female patients (mean age 55 ± 13 years) who underwent both contrast-enhanced chest CT and mammography within one year. Patients with surgical clips, prior breast surgeries, and those with artifacts projecting over the breast tissue on CT images were excluded. The tissue composition (dense, fibroglandular, fatty) and BIRADS score reported on mammography examinations were recorded. Furthermore, when suspicious for malignancy, the pathology report was used as the gold standard for classifying benign and malignant breast tissues. Thin-section CT images (1-1.25 mm) were reconstructed and processed with a Radiomics software prototype (Frontier, Siemens Healthineers) for segmentation and feature extractions of the left and right breast (1688 radiomic features) were analysed with multiple logistic regression and area under the curve for precision-recall curve analysis (R Statistical software).
Results or Findings: Automated segmentation-based radiomics differentiated the breast tissue as dense (n=779), fibroglandular (n=876), and fatty (n=108) with an 0.90 AUC (p<0.001). Of the 1764 breasts with a BIRADS score and pathology confirmation of malignancy, there were 1545 benign and 219 malignant breast tissues. To differentiate benign and malignant lesions in all breast tissue, radiomics had an AUC of 0.78 (p<0.001). To differentiate benign and malignant lesions in fatty and fibro glandular breast tissue (excluding dense breast tissue), radiomics had an AUC of 0.82 (p<0.001).
Conclusion: Radiomics can reliably differentiate breast tissue composition as well as predict malignancy in fibroglandular and fatty breast tissues with high accuracy. Radiomics can help assess breast tissues and suspicious breast lesions on contrast-enhanced chest CT examinations.
Limitations: This was a single centre study.
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: Mass General Brigham IRB approved this study.
7 min
Prediction of human epidermal growth factor receptor 2 (HER2) status in breast cancer by mammographic radiomics features and clinical characteristics: a multicentre study
Yalan Deng, Shanghai / China
Author Block: Y. Deng, Y. Lu; Shanghai/CN
Purpose: The study aimed to preoperatively evaluate the human epidermal growth factor 2 (HER2) status in breast cancer using radiomics features extracted from digital mammography (DM) and clinical characteristics.
Methods or Background: This study included a cohort of 1512 Chinese women with invasive ductal carcinoma of no special type (IDC-NST) from two different hospitals. 1332 from Institution A, used for training and testing the models, and 180 women from Institution B, as the external validation cohort. The Gradient Boosting Machine (GBM) algorithm was employed to establish a radiomics model and multiomics model using radiomics features and clinical characteristics. Model efficacy was evaluated by the area under the curve (AUC).
Results or Findings: The number of HER2-positive patients in the training, testing and external validation cohort was 245 (26.3%), 105 (26.3.8%) and 51 (28.3%) respectively, with no statistical differences among the three cohorts (P = 0.842, Chi-square test). The radiomics model, based solely on radiomics features, achieved an AUC of 0.814 (95% CI: 0.784-0.844) in the training cohort, 0.776 (95% CI: 0.727-0.825) in the testing cohort and 0.702 (95% CI: 0.614-0.790) in the external validation cohort. The multiomics model, incorporating radiomics features with clinical characteristics, consistently outperformed the radiomics model with AUC values of 0.838 (95% CI: 0.810-0.866) in the training cohort, 0.788 (95% CI: 0.741-0.835) in the testing cohort, and 0.722 (95% CI: 0.637-0.811) in the external validation cohort.
Conclusion: Our study demonstrates that a model based on radiomics features and clinical characteristics has the potential to accurately predict the HER2 status of breast cancer patients across different centres.
Limitations: The limitations were (1) ROIs were manually delineated, (2) the focus was exclusively on the relationship between radiomics features and HER2 status without analyzing other prognostic factors, and (3) the relationship between imaging equipment and radiomics model efficacy was not explored.
Funding for this study: Funding for this study was received from:
1. Clinical Research Plan of SHDC (No. SHDC2020CR4069)
2. Medical Engineering Fund of Fudan University (No. yg2021-029)
3. Shanghai Sailing Program (No. 21YF1404800)
4. Youth Medical Talents –Medical Imaging Practitioner Program (No. 3030256001)
5. Shanghai Municipal Science and Technology Major Project (No. 2018SHZDZX01)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Ethics Committee of Fudan University Affiliated Cancer Hospital.
7 min
Performance of radiomic features in STIR sequences in predicting histopathological outcomes of breast cancer
Günay Rona, Istanbul / Turkey
Author Block: G. Rona1, M. Arifoğlu1, T. A. Serel2, Ö. Adıgüzel Karaoysal1, Ş. Kökten1; 1Istanbul/TR, 2Isparta/TR
Purpose: This study aimed to investigate the performance of radiomic features in STIR sequences in predicting the results of histopathological outcomes of invasive breast cancer.
Methods or Background: Patients who underwent MRI before treatment were evaluated retrospectively. Histologic grade, ER, PR, HER-2, Ki-67 expressions and molecular subtypes were noted. Lesions were manually segmented from STIR sequences in the 3D Slicer program and volume of interest (VOI) was obtained. Machine learning (ML) analysis was performed using Python 2.3, the Pycaret library program. Datasets were randomly divided into training (70%) and independent testing set (30%). The performances of ML algorithms were evaluated by area under curve (AUC), accuracy, recall and precision values.
Results or Findings: 197 patients with a mean age of 50.72±46 (range 28-82) years were included in the study. The mean lesion size was 23.71±14.86 (5-120) mm. 156 of the patients were luminal A+B (79.2%), 17 were HER-2 positive (8.6%), and 24 were TN BC (12.2%). 156 (79.2%) of the patients were ER +, 41 (20.8%) were ER -, 126 (63.9%) were PR +, 71 (36.1%) were PR -, 58 (29.4%) were HER-2 +, 139 (70.6%) were HER-2 -. 43 (21.8%) of the patients were grade 1, 104 (52.8%) were grade 2, and 50 (25.4%) were grade 3.
The best results were obtained in predicting ER status and luminal A+B tumours. In the test set, AUC, accuracy, recall and precision values in ER+/- discrimination were 0.7518, 0.8048, 0.9628, and 0.8194, respectively. AUC, accuracy, recall and precision values in predicting luminal A+B tumours were 0.7229, 0.7958, 0.7958, and 0.6490 respectively.
Conclusion: Radiomic features obtained from STIR sequences have the potential to predict ER receptor status and luminal A+B tumours.
Limitations: The limitations were that it was a retrospective study and a small patient population.
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 with the approval code: 202351425620.
7 min
Radiogenomics model based on quantitative spatial heterogeneity for predicting pathological complete response and prognosis of triple-negative breast cancer
Jiayin Zhou, Shanghai / China
Author Block: J. Zhou, Y. Chao, Y. Gu; Shanghai/CN
Purpose: The study aimed to characterize the spatial heterogeneity of triple-negative breast cancer (TNBC) on MRI and develop a radiogenomics model for predicting both pathological complete response (pCR) and prognosis.
Methods or Background: In this prospective study, TNBC patients undergoing neoadjuvant chemotherapy were enrolled as the radiomics development cohort (n=315); among these patients, those with genetic data were enrolled as the radiogenomics development cohort (n=98). The external validation cohort (n=50) included patients from the DUKE database. Spatial heterogeneity was characterized using features from the tumour body, intratumoral subregions, and peritumoral region. Three radiomics models were developed by logistic regression after selecting features. Two fusion models were developed by further integrating pathological and genomics features (PRM: pathology-radiomics model; GPRM: genomics-pathology-radiomics model). Model performance was assessed with the AUC and decision curve analysis. Prognostic implications were assessed with Kaplan‒Meier curves and multivariate Cox regression.
Results or Findings: For radiomics models, the multiregional model representing spatial heterogeneity (Model 3) exhibited better pCR prediction with AUCs of 0.87, 0.79, and 0.74 in the training, internal validation, and external validation sets, respectively. GPRM showed the best performance for predicting pCR in the training (AUC=0.97, P=0.015) and validation sets (AUC=0.93, P=0.019). Model 3, PRM and GPRM could stratify patients by disease-free survival, and a predicted non-pCR was associated with poor prognosis (P=0.034, 0.001 and 0.019, respectively).
Conclusion: Imaging spatial heterogeneity could effectively predict pCR and prognosis of TNBC. The radiogenomics model could potentially serve as a valuable biomarker to improve the prediction performance.
Limitations: No limitations were identified.
Funding for this study: No information provided by submitter.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: No information provided by submitter.
7 min
Prediction of cancer aggressiveness based on breast MRI features
Veronica Magni, Milan / Italy
Author Block: V. Magni1, A. Benedek2, A. Colarieti2, F. Sardanelli1; 1Milan/IT, 2San Donato Milanese/IT
Purpose: This study aimed to investigate the value of breast MRI features for the prediction of tumour aggressiveness, considering the association of Ki67 expression and tumour grade with perilesional oedema, rim enhancement, necrosis sign, and adjacent vessel sign.
Methods or Background: Patients with histologically confirmed malignant breast lesions at preoperative breast MRI and available results of final pathology on surgical specimens were included in this retrospective study. Exclusion criteria were incomplete or suboptimal MRI examinations, incomplete histopathological data, breast implants, and neoadjuvant therapy. Pearson correlation coefficient was calculated to evaluate the strength of association between variables, dichotomising Ki67 expression as low (when <20% positive cells) and as high (when ≥20% positive cells). Multivariable binary logistic regression was then performed to identify significant predictors of Ki67 expression and histological tumour grade.
Results or Findings: Among 50 malignant lesions included in the study, 23/50 (48%) showed high-Ki67 expression, while 27/50 (54%) showed low-Ki67 expression. Seven (14%) lesions were grade 1, 26 (52%) were grade 2, and 17 (34%) were grade 3. Ki67 expression showed a positive association with perilesional edema (ρ=0.729, p<0.001), rim enhancement (ρ=0.382, p=0.006), necrosis sign (ρ=0.341, p=0.015), and adjacent vessel sign (ρ=0.327, p=0.020). At multivariable binary logistic regression, perilesional oedema and rim enhancement were significant predictors for high-Ki67 expression, showing odds ratios of 39.7 (p=0.002) and 13.6 (p=0.040) respectively. Perilesional oedema was significantly correlated with histological tumour grade (ρ=0.465, p<0.001).
Conclusion: Breast MRI features may have the potential to predict tumour aggressiveness, serving as prognostic and predictive biomarkers usable in clinical practice. The systematic and standardised reporting of these findings in radiological reports should be encouraged to obtain an initial assessment of tumour biological behaviour.
Limitations: The limitations were that it was a single-centre retrospective study with a small sample size.
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: This study was approved by the Ethics Committee of IRCCS Ospedale San Raffaele. The protocol code SenoRetro was approved on November 9th, 2017 and amended on April 4th, 2021.
7 min
Quantitative radiomic analysis in contrast-enhanced mammography for breast lesions characterisation
Gianmarco Della Pepa, Milan / Italy
Author Block: G. Della Pepa1, C. Depretto1, W. Carli1, G. Irmici1, E. D'Ascoli1, C. De Berardinis1, C. Cazzella2, D. Ballerini1, G. P. Scaperrotta1; 1Milan/IT, 2Bergamo/IT
Purpose: This study aimed to investigate the potential of radiomic quantitative texture analysis for characterising breast lesions on contrast-enhanced mammography (CEM) and correlate them with their biological phenotypes.
Methods or Background: Patients who underwent CEM procedures at our institution since 2018 were considered. Among them, all CEM-detected malignant lesions, confirmed via core needle biopsy and surgical intervention, were included in our study.
These lesions were firstly subjected to a semi-automatic segmentation, and then 93 radiomic features were extracted for each of them, using the open-source Python package Py-Radiomics.
The association between each feature and the predetermined endpoints was evaluated through univariate logistic regression analysis. The correlation was performed either with the singular molecular characteristics: the presence of estrogen (ER) and progesterone (PR) receptors, HER2 status, Ki67 level either with the specific immunophenotype: Luminal A, Luminal B, HER2+ and Triple Negative (TN).
Results or Findings: In our preliminary results, 86 patients were selected, with a total of 89 breast lesions analysed. The logistic regression isolated a subset of radiomic features correlating robustly with the biological phenotype. Second-order statistics textural features of Neighbouring Gray Tone Difference Matrix (NGTDM) demonstrated a stronger correlation with the presence of both ER and PR receptors, and multiple combinations of them resulted in a better correlation with Luminal A and Luminal B immunophenotype. The Gray Level Run Length Matrix GLSZM contrast and first-order uniformity both correlate with the TN immunophenotype.
Conclusion: Radiomic quantitative texture analysis of breast lesions on CEM demonstrates promising capability in characterising biological phenotype. Looking forward, it could lead to the construction of a nomogram to be used in clinical practice, potentially helping decision-making processes before biopsy.
Limitations: The study is constrained by a limited sample size and by the lack of a distinct validation cohort.
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: No information provided by the submitter.

This session will not be streamed, nor will it be available on-demand!