| Technological advances in medical imaging have created an explosion in the amount of data that is currently available to researchers and clinicians. The typical approach to analyzing brain image data focuses on small subject groups and manual, tedious and time-consuming image analysis techniques. Recent approaches to handling the large amount of image data in a database environment have only taken into account storage, security, and access needs. Our focus is in the area of psychiatric research and diagnosis, with an emphasis on patients in the early stages of Alzheimer's disease. Researchers need the ability to work with larger subject groups and analyze many images and sets of images, involving more comprehensive analyses without having to rely on manual processing and expert system support. The clinician, who must consult with patients on important topics such as future prognosis and treatment, cannot yet fully utilize the information in the patient's functional brain image, and must often rely solely on visual inspection of the image instead of objective analytical interpretation.; The contributions of this work, to both computer science and psychiatry, are: (1) the design and implementation of a suite of image, operators in the form of database extensions, including primitive, non-primitive and relational image operators that greatly reduce the work of analyzing brain images; (2) we show the value of the operators by developing two application interfaces---one built for the clinical environment and the second for the research environment---and present results of user acceptance trials for each application. The clinical interface was tested by clinicians on real patients seen for dementia; the research application was tested by researchers studying cognitive and drug behaviors in study populations; and (3) we studied automated image classification with support vector machines of MCI patient images and normal control images. We discuss the results of the classification trials, in addition to their implications for disease categorization and patient diagnosis. This work not only provides researchers with much needed tools to do more complex analyses than is currently possible, but enhances the diagnostic process by enabling and prototyping the systematic inclusion of neuroimaging data. |
