Sho Araki, Presenter: Nothing to Disclose

Hidetaka Arimura PhD, Abstract Co-Author: Nothing to Disclose

Takashi Yoshiura MD, PhD, Abstract Co-Author: Nothing to Disclose

Chiaki Tokunaga, Abstract Co-Author: Nothing to Disclose

Yasuo Yamashita RT, Abstract Co-Author: Nothing to Disclose

Masafumi Ohki PhD, Abstract Co-Author: Nothing to Disclose

Hiroshi Honda MD, Abstract Co-Author: Nothing to Disclose

Hideki Hirata, Abstract Co-Author: Nothing to Disclose

Fukai Toyofuku PhD, Abstract Co-Author: Nothing to Disclose

Our results showed that this method could be useful in evaluating the hippocampal volume loss due to AD.

One of important structures in Alzheimer’s disease (AD) is the hippocampus, especially for detection of the AD at early stages. We have developed a novel automated segmentation method for three-dimensional (3D) hippocampal regions in magnetic resonance (MR) images based on Bayes’ theorem and probabilistic atlases and to apply it to evaluation of hippocampal volumes loss due to AD.

Hippocampal probabilistic atlases, which denote the prior existence probability of the hippocampus, were derived from the hippocampal regions in 3D brain MR images of 10 control subjects, which were used as a training data set. The left and right hippocampal contours of the 10 cases were manually determined by consensus of a neuroradiologist and a medical physicist. The hippocampal probabilistic atlases (Fig. 1) were produced by registering all images to a reference image by using a non-linear free-form deformation (FFD) following a linear affine transformation. Next, the pixel value probability distributions of the hippocampal regions and their surroundings were obtained from the corresponding regions separately in all training cases. Finally, the posteriori existence probability of the hippocampus of each case was estimated based on the Bayes’ theorem with the probabilistic atlas and the pixel value probability distributions of the hippocampal regions. For evaluation of the proposed method, we applied it to 20 test cases including 10 control subjects and 10 AD patients, and calculated the hippocampal volumes for each case.

Figure 2 shows an example of segmentation of the left and right hippocampal regions in a control subject. Similarity index, i.e., the degree of the agreement between the hippocampal regions determined by the manual method and the proposed method, was 0.49 ± 0.17 for control subjects. The hippocampal mean volume for the 10 AD patients (2,694 ± 553 mm3) was significantly smaller than that for the 10 control subjects (4,166 ± 1,333 mm3) (p < 0.05).

Araki, S, Arimura, H, Yoshiura, T, Tokunaga, C, Yamashita, Y, Ohki, M, Honda, H, Hirata, H, Toyofuku, F, A Novel Segmentation Method for Three-dimensional Hippocampal Regions Based on Bayes Theorem and Probabilistic Atlases: Development and Application to Alzheimers Disease.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11008921.html