Abstract 4925: Spatial heterogeneity of tumor boundary in mammograms is prognostic of recurrence in triple negative breast cancer (TNBC)

Abstract Introduction: Despite standard-of-care therapies, at least 25% of all TNBC patients will develop recurrence or metastasis within 3-5 years and die due to breast cancer. Early prediction of recurrence in TNBC patients from clinical mammograms would allow more aggressive intervention for higher risk patients. The tumor boundary represents the heterogeneous invasive front where malignant cells interface extracellular matrix (ECM) and contribute to invasion. In this study, we investigated role of the irregular infiltration of the normal tissue by tumor in heterogeneity of radiomics features at the boundary of the tumor. Method: Surgically treated patients with pathologically confirmed diagnosis of TNBC of any tumor size were included. All cases had node negative status and were treated with adjuvant chemotherapy. Patients with node positive or metastatic TNBCs at diagnosis, receiving neo-adjuvant chemotherapy or non-chemo regimens were excluded. Manually segmented breast lesions in mammograms from a cohort of 29 patients with 5-year outcomes were used to automatically determine the tumor boundary as a zone of 1 mm from both intra-tumoral and peritumoral regions in concordance with prior studies. Over 2000 radiomics features from the tumor boundary and the central regions of the tumor were extracted to quantify heterogeneity, obscure margins, spiculations, texture, breast fat percentage, shape, and size of the tumor. Recursive feature elimination using random forest classifier was used to reduce feature dimensionality to prevent over-fitting and predict recurrence in TNBC patients. Results: We observed that tumor boundary heterogeneity as quantified by continuous gradient magnitude features at pixel level was significantly better at predicting lethality for TNBC patients when compared to usual radiomic features (gradient, texture, and shape features) extracted from the entire tumor, as done in prior radiomics analyses. Continuous gradient magnitude features quantifying spatial heterogeneity of the tumor boundary were found to be superior to texture features in providing estimation of local changes in pixel-level heterogeneity. In a three-fold validation framework, continuous gradient features were associated with a mean AUC of 0.82 (std 0.17) and better differentiated patient lethality risk as compared to gradient, texture, and shape features of the entire tumor (mean AUC of 0.27 (std 0.11)). Conclusions: To the best of our knowledge, this is the first time that heterogeneity features of a tumor quantified by continuous gradient magnitude features from tumor boundary, as observed in routine clinical mammograms, have been shown to predict TNBC lethality. This model will be validated in a larger cohort in future studies. Citation Format: Soumya Ghose, Cynthia Davis, Sanghee Cho, Yesim Polar, Fiona Ginty, Sunil Badve. Spatial heterogeneity of tumor boundary in mammograms is prognostic of recurrence in triple negative breast cancer (TNBC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4925.