Validation of tissue segmentation based on 3D feature map in an animal model of a brain tumor

Abstract

The purpose of this study was to validate our tissue segmentation technique by comparing its results with the composition of living biological tissues. A multispectral approach with three inputs was used. Volumetric MR images were obtained with steady state free procession, gradient echo, with RF spoiling and inversion recovery gradient echo techniques. The animal model used was brain tumors in hamsters. Immediately after imaging, animals were sacrificed and underwent thorough histological examination. Pre-segmentation image processing included our technique for correction of image non-uniformity, application of non-linear diffusion type filters, and, after collecting training points, cluster optimization. Finally, k-NN segmentation was used and a stack of color-coded segmented images was created. Results indicated that good quality of a small subject, such as a hamster brain MRI, can be obtained. Secondly, pre-processing steps vastly improved the results of segmentation-in particular, sharpness. We were able to identify up to eleven tissues. Most importantly, our findings were in full accord with histological exams.