Research Presentation Session: Breast

RPS 1402 - Predictive and prognostic models in breast imaging

March 4, 12:30 - 13:30 CET

8 min
Combining DCE-MRI pharmacokinetic parameters at early time points with prognostic factors improves the prediction of pathologic response to neoadjuvant chemotherapy in breast cancer
Fiona Gilbert, Cambridge / United Kingdom
Learning Objectives
Author Block: G. C. Baxter, J. C. Carmona-Bozo, R. Manavaki, A. Colarieti, R. Woitek, R. Bedair, J. Abraham, M. J. Graves, F. Gilbert; Cambridge/UK
Purpose or Learning Objective: To explore the additional value of pharmacokinetic parameters from DCE-MRI at early time-points in predicting pathologic complete response (pCR) to neo-adjuvant chemotherapy (NACT) in breast cancer.
Methods or Background: Women >18 years receiving NACT prior to surgery for breast cancer underwent baseline and post cycle 1 DCE-MRI examinations at 3T. DCE-MRI series were analysed using the extended Tofts’ model to derive Ktrans, kep, ve and hotspot Ktrans (hs-Ktrans). pCR was defined as no residual invasive cancer in the breast at surgery but allowing for the presence of in situ carcinoma. The area under the curve (AUC) was calculated to evaluate the predictive performance of logistic regression models including standard prognostic factors (histology, grade, molecular subtype) with and without the addition of DCE-MRI parameters.
Results or Findings: Data from 82 patients (86 lesions) were analysed. The majority were invasive ductal carcinomas (ductal: 71/86, 83%; lobular: 3/86, 3%; other: 12/86, 14%), hormone receptor (HR)-positive (57/86, 66%), with 31% HER2-positive. All tumours were either grade 2 or 3. 27/86 (31%) lesions showed pCR. Across all cancers, adding baseline hs-Ktrans increased AUC from 0.77 to 0.80, while the inclusion of Ktrans after 1 treatment cycle yielded the highest increase in AUC (0.72 to 0.76). For the HR+ group, the largest increase in AUC was observed for baseline hs-Ktrans (0.80 to 0.85). The addition of baseline hs-Ktrans and post cycle-1 kep showed the best predictive performance in triple-negative cancers (hs-Ktrans: 0.76 vs 0.59; kep: 0.93 vs 0.70).
Conclusion: The addition of DCE-MRI pharmacokinetic parameters at early time-points to standard prognostic factors can improve pCR prediction in HR+ and triple-negative breast cancer.
Limitations: Relatively small sample size from single site.
Ethics committee approval: NRES Committee South East (13/LO/0411).
Funding for this study: NIHR Cambridge Biomedical Research Centre.
8 min
Intra- and peritumoural radiomics based on dynamic contrast-enhanced MRI for a preoperative prediction of the intraductal component in invasive breast cancer
Hao Xu, chengdu / China
Learning Objectives
Author Block: H. Xu, L. hongbing, P. Zhou, J. k. Liu, J. Ren; Chengdu/CN
Purpose or Learning Objective: To develop and validate radiomic models for the preoperative prediction of the intraductal component in invasive breast cancer (IBC-IC) using the intra- and peritumoural features derived from dynamic contrast-enhanced MRI.
Methods or Background: The prediction models were developed in a primary cohort of 183 consecutive patients from September 2017 to December 2018. The validation cohort of 111 patients from February 2019 to January 2020 was enrolled to test the prediction models. A total of 208 radiomic features were extracted from the intra- and peritumoural regions of MRI-visible tumours. Then the radiomic features were selected and combined with clinical characteristics to construct predicting models using multivariate logistic regression. The AUC of receiver operating characteristics, sensitivity, and specificity were used to evaluate the performance of the radiomic models.
Results or Findings: Four radiomic models for the prediction of IBC-IC were built, including intratumoural radiomic signature, intratumoural radiomic nomogram, peritumoural radiomic signature, combined intra- and peritumoural radiomic signature. The combined intra- and peritumoural radiomic signature had the optimal diagnostic performance, with the AUC, sensitivity, and specificity of 0.802 (0.737-0.857), 0.733 (0.580-0.854), and 0.746 (0.665-0.817) in the primary cohort and 0.817 (0.732-0.884), 0.741 (0.537-0.889), and 0.750 (0.644-0.838) in the validation cohort.
Conclusion: The radiomic model based on the combined intra- and peritumoural features from DCE-MRI showed good ability to preoperatively predict IBC-IC, which might facilitate the individualized surgical planning for patients with breast cancer before breast-conserving surgery.
Limitations: This was a single-centre study. Hence, the radiomic models in this study need to be verified in a multicentre study with different imaging equipment in the future.
Ethics committee approval: The retrospective analysis was approved by our institutional review board, and the informed consent was waived.
Funding for this study: This study has received funding by the Sichuan Science and Technology Program (grant numbers 2021YFG0125).
8 min
Breast cancer recurrence risk prediction using breast MRI radiomics analysis with nested 10-fold cross-validation
Kun Sun, Shanghai / China
Learning Objectives
Author Block: K. Sun1, B. Wang2, D. Shen1, F. Yan1; 1Shanghai/CN, 2Harbin/CN
Purpose or Learning Objective: To investigate the value of MRI radiomics with nested-10 fold cross-validation based on T1-weighted (T1W) images in predicting recurrence risk in patients with breast cancer.
Methods or Background: This retrospective study enrolled 220 patients with histopathology-confirmed breast cancer and genomic testing. The patients were divided into low-, intermediate- and high-recurrence score (RS) groups based on their genomic testing results. A total of 788 radiomics features and 18 clinicopathological features were extracted to build a radiomics model, a clinico-pathological model, and a combined model. Univariate statistical tests and a random forest algorithm were performed via a nested 10-fold cross-validation to select the best features for predicting different RS groups. The predictive performance was validated by both the receiver operating characteristic curve (ROC) and a decision curve analysis (DCA).
Results or Findings: The area under the ROC curve (AUC) of the 3-class problem (i.e., classification of low-, intermediate- and high-RS groups) in the testing cohort ranged between 0.73 (clinico-pathological model) and 0.78 (radiomics model). The prediction performance of the radiomics model was superior to that of the clinico-pathological model and the combined model. For the 2-class problems (i.e., classification between low- and intermediate-RS groups, low- and high-RS groups, intermediate- and high-RS groups, and low- and intermediate-high RS groups), the AUCs ranged from 0.70 (radiomics model, intermediate- vs high-RS groups) to 0.88 (radiomics model, low- vs high-RS groups).
Conclusion: Radiomics features of breast MRI T1WI used in a machine learning classifier provided high discriminatory accuracy in predicting the recurrence risk of breast cancer.
Limitations: Our study is a single-centre retrospective study.
Ethics committee approval: This study was approved by our institutional ethics committee of Ruijin Hospital, Shanghai Jiaotong University School of Medicine.
Funding for this study: This study was funded by the National Natural Science Foundation of China (No. 81801651).
8 min
Multiparametric MRI-based radiomics nomogram for preoperative prediction of lymphovascular invasion and clinical outcomes in patients with breast invasive ductal carcinoma
Junjie Zhang, Taiyuan / China
Learning Objectives
Author Block: J. Zhang, Y. Cui, X. Yang; Taiyuan/CN
Purpose or Learning Objective: To develop a multiparametric MRI-based radiomics nomogram for predicting lymphovascular invasion (LVI) status and clinical outcomes in breast invasive ductal carcinoma (IDC) patients.
Methods or Background: 160 patients with pathologically confirmed breast IDC (training cohort: n=112; validation cohort: n=48) who underwent preoperative breast MRI were included. Imaging features were extracted from T2-weighted imaging (T2WI), apparent diffusion coefficient (ADC) maps, and contrast-enhanced T1-weighted imaging (cT1WI) sequences. Four-step procedures were applied for feature selection and radiomics signature building. Univariate and multivariate logistic regression analyses were conducted to identify the features associated with LVI, which were then incorporated into the radiomics nomogram. The performance of the nomogram was evaluated by its discrimination, calibration, and clinical usefulness. Kaplan-Meier survival curves based on the two radiomics models were used to estimate disease-free survival (DFS).
Results or Findings: The fusion radiomics signature of the T2WI, cT1WI, and ADC maps achieved a better predictive efficacy for LVI than either of them alone. The proposed radiomics nomogram, incorporating the fusion radiomics signature and MRI-reported peritumoral oedema, showed satisfactory capabilities of calibration and discrimination in both training and validation datasets, with AUCs of 0.919 (95% CI: 0.871-0.967) and 0.863 (95% CI: 0.726-0.999), respectively. The radiomics signature and nomogram-defined high-risk groups had a shorter DFS than those in the low-risk groups (both P<0.05). Higher Rad-scores were independently associated with a worse DFS in the whole cohort (P<0.05).
Conclusion: The proposed nomogram, incorporating multiparametric MRI-based radiomics signature and MRI-reported peritumoral oedema achieved a satisfactory preoperative prediction of LVI and clinical outcomes in IDC patients.
Limitations: No external validation.
Ethics committee approval: Institutional Review Board approval was obtained.
Funding for this study: The National Natural Science Foundation of China (No. 82001789 and 81802479).
8 min
Quantitative assessment of contrast-enhanced cone-beam breast computed tomography for the prediction of pathologic response to neoadjuvant chemotherapy in breast cancer: a prospective study
Shen Chen, Guangzhou / China
Learning Objectives
Author Block: S. Chen, C. Zhou, N. He, Y. Wu, s. li; Guangzhou/CN
Purpose or Learning Objective: To study the predictive value of contrast-enhanced cone-beam breast computed tomography (CE-CBBCT) for pathologic response to neoadjuvant chemotherapy (NAC) in breast cancer patients.
Methods or Background: This prospective study comprised eighty-one women with breast cancer who underwent NAC treatment from August 2017 to January 2021. All of them underwent CE-CBBCT at the pre-treatment; 55 and 66 patients had an examination during the mid-treatment (3 weeks) and the late-treatment (12 weeks) for NAC. Clinical information and quantitative parameters like diameter, volume, surface area and CT value at each enhancement stage of NAC were recorded. T-test, Mann-Whitney and chi-square tests were performed to evaluate each phase. The performance of each parameter was evaluated by area under the receiver operating characteristic curves (AUC), sensitivity, specificity and cut-off value.
Results or Findings: There was no significant difference between pCR and non-pCR in the baseline characteristics. However, mid- and late-treatment showed a predictive value for pCR with statistical significance (p<0.050). Segmented volume, surface area, washin rate, enhancement value and ratio during the treatment showed strong results (AUC=0.70~0.90). Segmented volume reduction (SVR) and maximum enhancement ratio (MER) showed as best predictors for pCR with a combinative AUC value of 0.874.
Conclusion: The best predictors in our study are SVR and MER in late-treatment. The results should be considered in assisting monitoring therapeutic response during NAC process.
Limitations: (1) We perform multivariate regression analysis on late-treatment parameters to construct a predictive model. Future studies with a larger population are warranted for validation. (2) Volumetric measurement became more challenging in the late period of NAC therapy due to the decreasing blood supply of the tumour.
Ethics committee approval: The Institutional Ethics Committee of Sun Yat-sen University Cancer Center (No.B2019-016) approved all experimental protocols of this study.
Funding for this study: This research was supported by the National Key Research and Development Program of China (grant numbers 2107YFC0112600.
8 min
The value of diffusion-weighted imaging (DWI) in pathological complete response (pCR) prediction in addition to dynamic contrast-enhanced (DCE) MRI in HER2-positive breast cancer patients
/
Learning Objectives
Author Block: A. van der Voort1, K. van der Hoogt1, R. Wessels2, R-J. Schipper3, G. Sonke1, R. M. Mann4; 1Amsterdam/NL, 2The Hague/NL, 3Eindhoven/NL, 4Nijmegen/NL
Purpose or Learning Objective: To investigate the added value of DWI to identify pCR in stage I-III HER2+ breast cancer patients with radiological complete response (rCR) after neoadjuvant chemotherapy (NAC) on DCE-MRI.
Methods or Background: We retrospectively identified patients treated with trastuzumab-containing NAC between January 2015 until September 2019 who had rCR (absence of pathologic enhancement) on post-chemotherapy DCE-MRI-breast in the Netherlands Cancer Institute. Baseline and post-NAC MRI’s (Philips 1.5/3.0T) were evaluated by a dedicated breast radiologist blinded for the pathological outcome. We re-evaluated rCR on DCE-MRI and visually evaluated response on high b-value DW-images (b800 and higher). ADC values were measured within the original tumour region. We calculated the negative predictive value (NPV) for pCR (ypT0/is) with a corresponding 95% standard logit confidence interval. Fisher’s exact and Mann-Whitney’s U test were used for comparison between groups.
Results or Findings: DCE showed rCR in 102 patients of whom 76 had a pCR. A pCR was more common in HR+/HER2+ than HR-/HER2+ patients (40 of 46 vs 36 of 56, p=0.01). Residual DWI signal was visible in 7 patients. NPVs for DCE and for DWI among patients with rCR on DCE, were respectively 74.5% and 77.9% (95%CI: 75.4-80.2%) overall, 64.3% and 70.0% (95%CI: 64.3-75.1%) in HR+ and 86.9% and 86.7 (95%CI: 86.1-87.2%) in HR- patients. Within HR+ patients with visual residual DWI signal only 1 of 6 had a pCR (16.7%, 95%CI: 2.5-61.5%). The relative mean ADC-difference in HR+ patients was 80.1% (IQR 41.1-128.6%) and 114.7%, respectively with and without pCR (IQR 25.1-191.7%; p=0.36).
Conclusion: Standardised DWI evaluation after NAC could potentially help to identify more HR+/HER2+ patients with residual invasive disease.
Limitations: Double reader analysis will be performed before ECR. Multiple DWI-scan protocols were used.
Ethics committee approval: Approved by the IRB.
Funding for this study: No funding was received for this work.
8 min
MRI morphological criteria and ADC value in predicting axillary lymph node (ALN) response after neoadjuvant chemotherapy (NAC): are we nearly there?
Maria Clotilde Sciandrello, Torino / Italy
Learning Objectives
Author Block: M. C. Sciandrello, M. Durando, G. Bartoli, E. Regini, A. Santonocito, A. Pittaro, I. Castellano, P. Fonio; Turin/IT
Purpose or Learning Objective: To identify which MRI criteria can predict residual ALN disease in breast cancer patients undergone NAC.
Methods or Background: From 2014 to 2021, pre-and post-NAC 1,5 T MRIs of 164 patients with locally advanced breast cancer were retrospectively analysed by two dedicated radiologists in consensus, blinded to histological results. We evaluated both quantitative (number, diameter) and qualitative (irregular margins, absence of fatty hilum, cortical thickness>3mm, perifocal oedema, rim enhancement, asymmetry comparing with contralateral side) criteria and ADC value related to ALNs before and after NAC. ALNs status was compared before NAC with ALN biopsy and with sentinel ALN biopsy or axillary dissection after NAC; nodal pathological response is classified according to Pinder’s criteria [complete response (pCR) versus no-complete response(no-pCR)]. Statistical analysis (Chi-square or Fisher’s exact tests for categorical variables, non-parametric Mann-Whitney test for continuous variables) was performed.
Results or Findings: At pre-therapy MRI, the two parameters that best correlated with positive ALN biopsy were irregular margins and the absence of fatty hilum (p= 0,0003 and p=0,0014 respectively), while, after NAC, relating the different parameters with pCR or no-pCR, the only statistically significant data was the irregularity of margins (p= 0,0003). The other variables, although at the univariate analysis they seemed to demonstrate a statistically significant correlation, did not confirm this data at the multivariate analysis.
Conclusion: Based on our results, irregular ALNs margins seem to be the most reliable parameter associated to pre-therapy ALNs disease and no-pCR after NAC.
Limitations: Retrospective study.
Ethics committee approval: Not required.
Funding for this study: No funding was provided for this study.