Summary The overall aim of this project is to investigate and better understand the repeatability and robustness of radiomics in breast cancer imaging. Radiomics from medical images can provide information about lesion features such as size, irregularity, and texture, which can be used to produce quantitative image-based phenotypes that can assist in diagnosis of cancer and assessment of treatment. Using previously acquired radiomics measurements of breast cancer imaged by full-field digital mammography (FFDM) and magnetic resonance (MR), Aim 1 of this study is to assess their repeatability using three classifiers (linear discriminant analysis, support vector machines, and Bayesian neural network methods), bootstrapping for variability assessment, and receiver operating characteristics (ROC) methods. By doing so, we will be able to evaluate how radiomics may be expected to vary in their output and performance on FFDM and MR. In Aim 2, we endeavor to understand the cross-modality performance of radiomics measurements of lesion cases imaged by both FFDM and MR. This work will provide a new understanding of the robustness of radiomics tumor descriptors compared across two modalities and fulfill a currently unmet need ? thus being both novel and significant. Statistical analysis will be conducted using superiority and non-inferiority testing. These studies will provide a better understanding of the repeatability and robustness of radiomics of breast lesion images, an important step in establishing their utility in disease diagnosis and treatment assessment.

R15CA227948

2019-01-16

2021-12-31