<title>Individual 3D region-of-interest atlas of the human brain: neural-network-based tissue classification with automatic training point extraction</title>

Abstract

The purpose of individual 3D region-of-interest atlas extraction is to automatically define anatomically meaningful regions in 3D MRT images for quantification of functional parameters (PET, SPECT: rMRG1u, rCBF). The first step of atlas extraction is to automatically classify brain tissue types into gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), scalp/bone (SB) and background (BG). A feed-forward neural network with back-propagation training algorithm is used and compared to other numerical classifiers. It can be trained by a sample from the individual patient data set in question.Classification is done by a "winner takes all' decision. Automatic extraction of a user-specified number of training points is done in a cross-sectional slice. Background separation is done by simple region growing. The most homogeneous voxels define the region for WM training point extraction (TPE). Non-white-matter and non-background regions are analyzed for GM and CSF training points. For SB TPE, the distance from the BG region is one feature. For each class, spatially uniformly distributed training points are extracted by a random generator from these regions. Simulated and real 3D MRI images are analyzed and error rates for TPE and classification calculated. The resulting class images can be analyzed for extraction of anatomical ROIs.

MRI-Based Individual 3D Region-of-Interest Atlases of the Human Brain

“…The only way to avoid warping while still meeting these requirements is developing a method that generates a 3D atlas directly by automated segmentation of individual MRI images. Atlases thus generated are referred to MRI-based individual 3D region-of-interest atlases (3D-IROI atlases) [18][19][20].…”

“…Since functional parameters are mainly determined in regions of gray and white matter, the exact segmentation of these tissues is essential, as is their differentiation from neighboring tissues. Hence, the first step is classifying the T1-weighted MRI grayscale images into gray and white matter, cerebrospinal fluid, adipose tissue as part of scalp/bone and background (lowlevel processing) [18,20].…”

“…This method allows the implicit segmentation and classification of tissue regions while taking into account different features (vectors). Parts of the method are described in [18,20], including pre-processing steps for reducing artifacts, if necessary. Therefore, only the main issues are dis-cussed here.…”

<title>Simulation of 3D MRI brain images for quantitative evaluation of image segmentation algorithms</title>

“…Especially the histogram-based segmentation of gray and white matter could be a problem since the histogram curve can show a smooth transition from GM to WM. For this reason, more sophisticated statistical and neural network approaches for classification of individual data sets were developed [2,3]. For development of special software phantoms, however, this method is sufficient.…”

“…Two examples of applications of this software phantom: 1 . evaluation of a dynamic programming (DP) algorithm for contrast correction based on the cumulative grayscale value histograms of neighboring slices of gradient-spoiled T1-weighted 3D-FLASH image data [1]; 2. evaluation of the first main steps in individual region-of-interest (ROT) atlas extraction, which are the automatic detection of training points and supervised classification of T1-weighted MRI grayscale images into brain tissue types (GM, WM, CSF, SB, BG) [2,3]. Error rates for these examples are calculated based on this software phantom.…”