| In this dissertation, effective computational tools are developed for improved detection, diagnosis, and monitoring of breast cancer and response to therapy. Firstly, a tactile mapping device (TMD) is tested for the early detection of breast cancer through more objective and quantitative breast palpation. Secondly, we introduce a hybrid source decomposition algorithm, which allows for a computational characterization of tumor microvascular heterogeneity in both spatial and temporal domains. The method is based on a combination of time-activity curve clustering, pixel subset selection, and independent component analysis. The goal is to reveal temporal-spatial patterns for the visualization and quantification of tumor-induced angiogenesis and response to therapy using dynamic contrast-enhanced magnetic resonance imaging. Lastly, an image-based change detection approach is proposed to assess tumor's response to therapy where the main task is non-rigid image registration. The recovery of transformational geometry is initially achieved via a mixture of principal axes registrations, whose parameters are estimated by using the expectation-maximization algorithm. The remaining nonlinear deformation is further corrected by a trained multilayer perceptron mapping. We demonstrate the principle of the approaches on both simulated and real data sets. Based on promising experimental results, we anticipate that functional imaging based computational characterization of tumor heterogeneity and response will be useful in a wide variety of medical imaging studies. |
