Use of Radiomic Texture to Characterize Relationships Across the Parenchymal Field in Women with Breast Cancer On FFDMs and Specimen Radiographs
N Baughan1*, H Li1, L Lan1, M Embury2, G Whitman2, R El-zein3, I Bedrosian2, M Giger1, (1) University of Chicago, Chicago, IL, (2) MD Anderson Cancer Center, Houston, TX, (3) Houston Methodist Research Institute, Houston, TX
Presentations
TU-E-201-6 (Tuesday, 7/12/2022) 11:00 AM - 12:15 PM [Eastern Time (GMT-4)]
Room 201
Purpose: In women with biopsy-proven breast cancer, histologically normal areas of the parenchyma within the ipsilateral breast have shown molecular similarity to the tumor, supporting a potential cancer field effect. The purpose of this work was to investigate relationships of human-engineered radiomic features between multiple regions across the breast in mammographic parenchymal patterns and specimen radiographs.
Methods: This study included 32 patients with at least one identified malignant tumor who had undergone mastectomy and both preoperative mammographic imaging and intraoperative radiographic imaging of the mastectomy specimen. Mammograms were acquired with a Hologic Lorad Selenia system (12-bit quantization, 70-micron pixels), specimen radiographs were acquired with a Fuji imaging system (12-bit quantization, 50-micron pixels), and all images were retrospectively collected under an IRB-approved protocol. 128x128 pixel regions of interest (ROI) were selected from three regions across the craniocaudal mammogram and serially sectioned specimen radiographs: in the identified tumor, adjacent to the tumor, and distant from the tumor. Forty-five radiomic features were extracted from each ROI. Kendall’s tau-b was used to evaluate relationships within mammograms and specimen radiographs separately, and Pearson’s correlation test was used to evaluate correlation between mammogram and specimen features.
Results: Kendall test results indicated statistically significant correlations between the tumor, near, and far regions in mammograms for intensity-based histogram and Fourier-based powerlaw beta features, and significant correlations between intensity-based histogram and GLCM features in specimen radiographs. Pearson correlation results confirmed the statistically significant correlation in intensity-based histogram features between mammograms and specimen radiographs, presenting a clear relationship across the tumor, near, and far regions in both modalities.
Conclusion: Correlations indicated that human-engineered radiomic features from ROIs closer to the tumor tended to show more similarity to the tumor than features from distant ROIs and show strong relationships of these features across the parenchymal field in in- and ex-vivo imaging.
Funding Support, Disclosures, and Conflict of Interest: The work was partially supported by NIH T32 EB002103, U01 CA195564, U01 CA189240, and the University of Chicago Comprehensive Cancer Center. MLG receives royalties from Hologic, GE Medical Systems, MEDIAN Technologies, Riverain Medical, Mitsubishi, and Toshiba, and was a cofounder in Qlarity Imaging. HL and LL receive royalties from Hologic.
Keywords
Texture Analysis, Breast, Mammography
Taxonomy
IM- Breast X-Ray Imaging: Radiomics
Contact Email