We are developing new analysis pipelines to extract meaningful, quantitative data from medical images that can be incorporated into the radiology pathway to aid clinical decision making. Deep learning, and other approaches to automated image analysis, are hampered for most medical applications by the paucity of data. Data quality is a universal issue. Our approach is to introduce context into the learning process, coupling learning with physical models mimicking the natural approach used in human interpretation/assessment of images.
Related to the extraction of properties is their incorporation in continuum models of tissue mechanics. Here, our central interest is in advancing the multiphase representation of soft tissue mechanics to better understand their underlying mechanisms of mechanical function, their link to micro- and miso-mechanics, their sensitivity to structural configuration, and the cycle of development, homeostasis, disease and, potentially, recovery.
