The Medical Imaging group is dedicated to analyzing and optimizing imaging systems in general, with particular application to clinical radiology. Throught there is a substantial program in actual design and construction of imaging instruments, only the theoretical and computational aspects are described here. A major goal of this research is to develop a rigorous theoretical and experimental framework for objective assessment of image quality and to apply it to the development of practical, clinically useful algorithms for a gamma-ray imaging modality called single-photon emission computed tomography (SPECT).
Modem tomographic imaging methods such a SPECT rely on sophisticated reconstruction algorithms implemented on a digital computer. Much research in many laboratories has gone into developing new algorithms. Current iterative algorithms incorporate reasonably accurate models of the physics of the imaging process, including attenuation and scatter of radiation of radiation in the patient's body, as well as some prior knowledge of the object being imaged.
In spite of all this effort, however, current clinical practice still uses relatively simple algorithms first derived decades ago. We believe there are two major impediments to bringing the results of modem algorithmic research into clinical practice. One is the computational burden. Even with the rapid increases in computer power, it is still prohibitive to perform a fully three-dimensional reconstruction incorporation all known physics and prior information about the object. In this group we are attempting to reduce this computational time by using powerful parallel computers, and we are developing algorithms that are well suited to parallel computation.
The second major impediment to the use of the most complex and sophisticated algorithms, in our opinion, is the lack of clear-cut methods for determining whether there are any practical clinical benefits to be derived in return for the computational expense. To overcome this obstacle, we are developing new methods for objective assessment and prediction of image quality. The basic approach taken in our laboratory and several others [1-4] is to define image quality in terms of how well an observer can perform sone specific task of clinical interest using the reconstructed images. Since the ultimate observer of a clinical image is always a human, psychophysical studies are an important part of this evaluation [5].
Psychophysical studies are, however, too time consuming and imprecise to be used for the routine evaluation and optimization of algorithms or imaging systems. Therefore we are also investigating various mathematical or model observers. One important goal is to define remain analytically tractable for realistic tasks [6,7].
In addition to increasing our basic understanding of image quality, we also aim to produce practical advances in clinical SPECT. SPECT is a valuable tool for the study of organ function, but the images are degraded by poor spatial and temporal resolution, scatter and attenuation in the patient's body, and image noise arising from poor counting statistics. As a result, it is often difficult to detect subtle lesions or to obtain quantitatively accurate estimates of tracer concentration in a region of interest or of other physiologically relevant parameters. A primary goal of our research is to overcome there limitations and perform a rigorous optimization of the algorithms and imaging systems used in clinical SPECT. Specific clinical studies expected to benefit from these investigations include myocardial imaging, detection of breast cancer and other tumors, and detection and characterization of stroke and epileptogenic foci in brain imaging.