Noninvasive assessment of cerebral vasculature that is characterized by significantly increased microvessel density is of considerable clinical interest, in particular for evaluation of tumor invasiveness. Due to its high resolution, the use of magnetic resonance imaging at 8 T (tesla) has great potential to contribute to predicting tumor grade. With the advantages of high field imaging comes the disadvantages (ie. inhomogeneity of the image) that must be overcome. The primary goal of this proposal is to provide automation that improves the throughput and reliability of the microvessel identification and quantification for 8 T high-resolution images. This goal is made difficult because high field images have severe variability of image signal-to-noise ratio and contrast. Thus, standard image processing algorithms are inadequate. For this proposal, three-dimensional (3-D) texture segmentation using co-occurrence matrices will be employed. A secondary goal of this proposal is the comparison of the MRI vessel density to the microvessel density on matching microtome photomicrographs of cadaveric brain tissue sections. To achieve this goal the watershed algorithm can be applied multiple times allowing for the merger of the segment regions. ? ?
