RPS 1216 - New perspectives in breast and gynaecological cancer

Author Block: C. Pizzi¹, C. Sattin¹, F. Arnone¹, D. Berloco¹, P. Hoxha¹, P. Summers¹, R. Maggioni¹, A. R. R. Padhani², G. Petralia¹; ¹Milan/IT, ²Northwood/UK
Purpose: To investigate the potential of the response assessment category (RAC) from MET-RADS-P guidelines as prognostic biomarker in metastatic breast cancer (MBC) patients.
Methods or Background: We enrolled MBC patients who underwent whole-body MRI at baseline and at each time point (every 12 weeks disease until progression) after systemic anti-cancer therapy (SACT). We correlated the maximum RAC at time point 1 (TP1) with overall survival (OS). Patients were divided in two groups: those with a maximum RAC 1-2 (highly likely or likely to be responding, respectively) and those with a maximum RAC 3-4-5 (stable disease, likely or highly likely to be progressing) at TP1. Survival curves were depicted in Kaplan-Meier plots and compared via a log-rank test and hazard ratio (HR) using Cox regression model, with point comparisons of three-year survival and median survival duration, using R.
Results or Findings: Out of 45 MBC patients enrolled, a higher OS was observed in patients with a maximum RAC 1-2 (N=18) than in those with a maximum RAC 3-4-5 (N=27) at TP1 (log-rank test p=0.007). Because more than 50% of the maximum RAC 1-2 patients are still living, the median survival duration could not be determined, median survival in the maximum RAC 3-4-5 group was 36 months (upper limit of 95%CI not available). The HR for the maximum RAC 3-4-5 patients was 2.28 (95%CI 1.24 – 3.33). Three years OS was 88.9% for RAC1-2 vs 42.6% for RAC 3-4-5; for a difference of 46.2% (95%CI 12.7%-79.8%, p=0.0068).
Conclusion: Our observations support the potential of RAC after TP1 as a prognostic biomarker in MBC patients undergoing SACT.
Limitations: Retrospective and monocentric 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: Not applicable

Author Block: P. Morrone, C. Zampieri, C. Esposito, E. Barone, I. Capitoni, F. Gentili, G. Bagnacci, S. Guerrini, M. A. Mazzei; Siena/IT
Purpose: To identify a machine learning (ML) model with morphological and dual-energy (DE) data, to characterize lymph node’s (LN) status during breast cancer (BC) staging.
Methods or Background: From a cohort of 636 patients who undergone whole-body DE-CT and subsequent surgery with axillary lymphadenectomy between April 2015 to July 2023, 117 patients were included. Exclusion criteria: previous ipsilateral breast or axillary surgery, or chemo-radiotherapy; poor quality CT; lack of anatomopathological data.
For the morphological analysis, the main diameter of the neoplasm and location, long and short axis and morphological features (fat hilum, cortical area status, extranodal extension-ENE) of the ipsilateral axillary LNs were recorded.
For quantitative analysis regions of interest (ROIs) were placed on the neoplasm and axillary LNs encompassing an area of post-contrast enhancement as large and homogeneous as possible. An attempt was made to place the ROIs on the entire LN excluding the fat hilum and surrounding structures, setting a HU displaying threshold to suppress negative HU values. For each ROI, mean attenuation value at 40 and 70keV, iodine concentration (IC), water concentration (WC) and effective-Z value were recorded.
Results or Findings: 116 BC and 375 LNs were analyzed, 180 pathological and 195 non-pathological.
On univariate analysis the attenuation (HU) at 40 and 70keV, slope, IC, WC, long and short LNs axis showed statistically significant differences between histologically proven pathological and non-pathological LNs (p<0.001).
There were statistically significant differences (p<0.001) according to the cortical status and ENE.
The logistic regression-based ML model included IC, short axis, fat hilum, cortical status and ENE; the ROC curve showed an AUC of 0.881, demonstrating good model accuracy.
Conclusion: The ML model provides a good discriminatory ability to differentiate pathological from non-pathological axillary LNs in patients with BC.
Limitations: Not applicable
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: Waived from our etical committe due to the retrospective nature of this study.

Author Block: H-J. Meyer¹, A-K. Höhn¹, A. Surov²; ¹Leipzig/DE, ²Minden/DE
Purpose: The complex interactions of the tumor micromilieu could be reflected by diffusion-weighted imaging (DWI) derived from the magnetic resonance imaging (MRI). The present study investigated the association between apparent diffusion coefficient (ADC) values and histopathological features in uterine cervical cancer.
Methods or Background: This retrospective study used the prebiopsy MRI to analyze histogram ADC-parameters. The biopsy specimens were stained for Ki 67, E-cadherine, vimentin and tumor-infiltrating lymphocytes (TIL, all CD45 positive cells). Tumor-stroma ratio (TSR) was calculated on routine H&E specimen. Spearman’s correlation analysis and receiver-operating characteristics curves were used as statistical analyses.
Results or Findings: The patient sample comprised 70 female patients (age range 32-79 years; mean age 55.4 years) with squamous cell cervical carcinoma. The interreader agreement was high ranging from intraclass coefficient (ICC)=0.71 for entropy to ICC=0.96 for ADCmedian. Several ADC-histogram parameters correlated strongly with the TSR. The highest correlation coefficient achieved p10 (r=-0.81, p<0.0001). ADCmean can predict tumors with high TSR, AUC: 0.91, sensitivity: 0.91 (95%CI 0.77;0.96), specificity: 0.91 (95%CI 0.78;0.97). Also, several ADC-histogram parameters correlated slightly with the proliferation index Ki 67. No associations were found with TIL, E-Cadherin and vimentin. In well and moderately differentiated cancers, ADC histogram values showed stronger correlations with Ki 67 and TSR than in poorly differentiated tumors.
Conclusion: ADC values are strongly associated with tumor-stroma ratio. ADC mean can be used for prediction of tumors with high TSR. Associations between histopathology and ADC values depend on tumor differentiation. ADC values show only weak associations with Ki 67 and none with TIL, vimentin and E-Cadherin.
Limitations: First, it is a retrospective study with known inherent bias. Second, the patient sample is comprised from a single center.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethics commitee University of Leipzig (Ethical code: 012/13–28012013)

Author Block: X. Cai¹, J. Han², G. Zhang², F. Yang¹, Y. Wang¹, J. Liu¹, R. Li¹; ¹Shijiazhuang/CN, ²Shanghai/CN
Purpose: To test the feasibility of low-dose abdominopelvic CT with an artificial intelligence iterative reconstruction (AIIR) for diagnosing peritoneal invasion in pre-operative imaging of ovarian tumor.
Methods or Background: In this prospective study, 88 patients with pathology-confirmed ovarian tumors were enrolled, where the routine-dose CT scan at portal venous phase (120 kV/ref. 200 mAs) was followed immediately with a low-dose scan (120 kV/ref. 40 mAs). Images at routine dose were reconstructed with hybrid iterative reconstruction (HIR) and images at low dose were reconstructed with AIIR. Two radiologists independently diagnosed the peritoneal invasion using a 5-point confidence scale (1: definitely absent, 5: definitely present). In case of disagreement, the consensus was obtained through a third radiologist. The diagnostic performance was assessed using receiver operating characteristic (ROC) analysis with pathological results serving as the reference. The inter-observer agreement was assessed by Cohen’s Kappa test.
Results or Findings: The 88 patients consisted of 37 patients with benign/borderline ovarian tumors and 51 patients with ovarian carcinomas. The effective dose of low-dose CT at portal venous phase was 79.8% lower than that of routine-dose scan (2.64 ± 0.46 mSv vs. 13.04 ± 2.25 mSv, p < 0.001). In diagnosing peritoneal invasion, the area under the ROC curve (AUC) of low-dose AIIR and routine-dose HIR images was 0.961 and 0.960, respectively (p = 0.734). The sensitivity, specificity, and accuracy were 86.1%, 92.3%, and 89.8%, respectively, for low-dose AIIR images, and 86.1%, 90.4%, and 88.6%, respectively, for routine-dose HIR images. The interobserver agreement was good for diagnosing peritoneal invasion (κ = 0.694).
Conclusion: In low-dose pre-operative CT of ovarian tumor with 80% dose reduction, AIIR delivers similar diagnostic accuracy for peritoneal invasion as compared to routine abdominopelvic CT.
Limitations: Not applicable.
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 ethics committee notification can be found under the number 2024KS138.

Author Block: K. Shapriya, A. Jackson, X. Li, S. Das, N. Bharwani, A. G. Rockall; London/UK
Purpose: Preoperative differentiation between leiomyosarcoma (LMS) and atypical benign leiomyoma (LM) is diagnostically challenging. This study aims to develop and validate a machine learning (ML) model using MRI-based clinical and radiomic features to distinguish LMS from LM.
Methods or Background: This retrospective study included 214 patients with atypical myometrial lesions who underwent surgery between 2013 and 2023. All subjects had preoperative full blood count (FBC) and MR imaging. Among 214 cases, 193 were LM and 21 were LMS. T2-weighted sagittal MRI sequences were manually segmented then optimized using nonuniformity correction method to ensure image stability. 4114 radiomic features were extracted using TexLab (version 2) and IBSI compliant MATLAB™ software. These radiomic features and 11 clinical variables (including age and FBC) were incorporated into several ML models, with the dependent variable being LMS (binary). Data was split 70:30 into training and testing sets. To address data imbalance, an ensemble classification model was employed with unequal classification costs, penalizing the misclassification of LMS. The model’s performance was evaluated using area under the curve (AUC), sensitivity, specificity, accuracy, F1 score, and confusion matrix.
Results or Findings: The final ensemble model included four clinical and six radiomic features. The test dataset demonstrated a sensitivity, specificity, accuracy and AUC of 1, 0.8, 0.84 and 0.90, and F1 score of 0.50.
Conclusion: This study presents a promising ML model for preoperative differentiation of LMS from LM, achieving high accuracy (84%). As sarcoma subjects are uncommon, the ML model was developed to take data imbalance into account. High sensitivity was achieved, but with some loss of specificity. Future research will focus on validating this model using larger datasets to enhance its reliability and clinical application.
Limitations: None
Funding for this study: No funding
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: HRA and Health and Care Research Wales (HCRW)
Reference: 20/HRA/4925