Multispectral analysis of cerebral ischemia with applications to diagnosis and therapy

Magnetic resonance imaging has become a valuable tool for the evaluation of cerebral ischemia. Diffusion-weighted imaging (DWI) provides a means to view the evolution of the ischemic lesion in vivo. The microcirculation of the brain can be evaluated by perfusion imaging (PI). Methods for treating acute ischemia can be divided into two broad categories: methods that attempt to restore blood flow to adequate levels and methods that attempt to protect neurons from damage during the period of ischemia. DWI and PI can be used to evaluate the performance of these therapies in vivo. Within this dissertation, several studies will be presented which employ MRI methods to evaluate potential therapies for cerebral ischemia.; A major difficulty in staging and predicting ischemic brain injury by magnetic resonance imaging (MRI) is the time-varying nature of magnetic resonance parameters as cerebral ischemia progresses. Renormalization of the average apparent diffusion coefficient {dollar}rm(ADCsb{lcub}av{rcub}),{dollar} partial reperfusion, and increases in {dollar}Tsb2{dollar}-relaxation time due to vasogenic edema make it difficult to assess ischemic tissue damage on the basis of a static view of one or two MR parameters. In this dissertation, a new approach will be described to employ multispectral analysis (MS) to characterize cerebral ischemia in a time-independent fashion, where the burden of ischemic tissue identification is shouldered by several MR parameters over time. This feature would be particularly important in the clinical setting, where the time of the onset of the stroke is frequently unknown and the ability to distinguish potentially salvageable from irreversibly damaged ischemic tissue would be important for acute stroke management.; Multispectral (MS) analysis of multiple MR parameters (ADC{dollar}rmsb{lcub}av{rcub},{dollar} diffusion anisotropy, T{dollar}sb2,{dollar} proton density, and perfusion) has been employed to characterize the ischemic lesion at various times during the progression of permanent and transient focal cerebral ischemia in the rat brain. Classification methods employed to detect acute ischemia include: Multivariate Gaussian (MVG); k-Nearest Neighbor (k-NN); K-means (KM); and Fuzzy c-means (FCM). Unsupervised classifiers (KM, FCM) do not require prior statistics or labeled training data resulting in potentially greater clinical usefulness. In this dissertation, it will be demonstrated that MS analysis provides an improved estimate of ischemic lesion volume over that obtained from conventional methods that employ ADC alone. In addition, KM and FCM have been employed to estimate the ischemic lesion volume at multiple time points during the temporal evolution of the lesion.; A major goal in cerebral ischemia research is the identification of reversible ischemic injury, otherwise known as the ischemic penumbra region. The relationship between diffusion and perfusion has been identified as an important component in the characterization of the penumbra. MS analysis offers a formal method to combine these features for the identification of the penumbra. KM has been employed in a hierarchical manner to perform multi-scale segmentation, where the early acute ischemic pixels are classified as core or penumbra.