Characterizing imaging data

Abstract

Imaging represents a frequent, non-invasive, longitudinal, in vivo sampling technique for acquiring objective insight into normal and disease phenomenon. Imaging is increasingly used to document complex patient conditions, for diagnostic purposes as well as for assessment of therapeutic interventions (e.g., drug, surgery, radiation therapy) [81]. Imaging can capture structural, compositional, and functional information across multiple scales of evidence, including manifestations of disease processes at the molecular, genetic, cellular, tissue, and organ level [47]. Imaging allows both global assessment of disease extent as well as the characterization of disease micro-environments. Advances in imaging during the past decade have provided an unparalleled view into the human body; and in all likelihood these advances will continue in the foreseeable future. There has been considerable research directed to developing imaging biomarkers, defined as, ...anatomic, physiologic, biochemical, or molecular parameters detectable with imaging methods used to establish the presence or severity of disease which offers the prospect of improved early medical product development and preclinical testing [188]. Yet the full utility of image data is not realized, with prevailing interpretation methods that almost entirely rely on conventional subjective interpretation of images. Quantitative methods to extract the underlying tissue specific parameters that change with pathology will provide a better understanding of pathological processes. The interdisciplinary field of imaging informatics addresses many issues that have prevented the systematic, scientific understanding of radiological evidence and the creation of comprehensive diagnostic models from which the most plausible explanation can be considered for decision making tasks. In this chapter, we explore issues and approaches directed to understanding the process of extracting information from imaging data. We will cover methods for improving procedural information, improving patient assessment, and creating statistical models of normality and disease. Specifically, we want to ascertain what type of knowledge a medical image represents, and what its constituent elements mean. What do contrast and brightness represent in an image? Why are there different presentations of images even when the patient state has not changed? How do we ground a particular pixel measurement to an originating (biological) process? Understanding of the data generation process will permit more effective top-down and bottom-up processing approaches to image analysis. © Springer Science + Business Media, LLC 2010.