Toru Higaki, Presenter: Nothing to Disclose

Daisuke Komoto, Abstract Co-Author: Nothing to Disclose

Koji Iida, Abstract Co-Author: Nothing to Disclose

Yoko Kaichi, Abstract Co-Author: Nothing to Disclose

Yutaka Hirokawa, Abstract Co-Author: Nothing to Disclose

Kazuo Awai MD, Abstract Co-Author: Research Grant, Toshiba Corporation Research Grant, Hitachi Ltd Research Grant, Bayer AG Research Consultant, DAIICHI SANKYO Group Research Grant, AZE, Ltd

Shuji Date, Abstract Co-Author: Nothing to Disclose

One of the methods for detecting abnormalities on brain PET images is by using the asymmetry index (AI), which estimates the laterality of a symmetric object. For the computerized automatic calculation of the AI we deform an input image into a symmetric image and calculate the AI between corresponding pixels on both sides. In cases where the brain is not completely symmetric it it is difficult to calculate the AI accurately because of mis-correspondences. We developed an AI computation method that uses brain anatomical regions of interest (ROI) and investigated its accuracy for detecting brain laterality.

We first symmetrize an input image by registering it to a statistical parametric mapping (SPM) template image using a non-rigid registration technique. Then we apply the anatomic ROI to the symmetrized input image and calculate the average value for each ROI. Lastly we calculated the AI for each ROI using the formula: AI = (Vhere - Vopp) / (Vhere + Vopp) x 100 [%], where Vhere indicates the averaged value of the focus ROI and Vopp the averaged value of the contralateral ROI. To confirm the validity of our method we created model data based on an SPM template image. We added simulated spherical 30-, 40-, and 50-mm diameter cold-spots to the template image, generated 36 asymmetric images by applying translation and rotation to the simulated images, and applied our method to these images. To compare the accuracy of our- and the conventional method we calculated laterality with the conventional method using grid-like ROIs containing 1-, 2-, 4-, 8-, and 16 pixels on one side.  

With the conventional method the averaged errors for each grid size were 2.46-2.74% per voxel compared with the true value. With our method they were 1.01% per voxel.

Comparison of the conventional- and our method showed that our method using anatomic ROIs was highly accurate for the detection of brain laterality.

Our newly developed method using anatomical ROIs may improve the accuracy of detecting abnormalities on brain images acquired by PET, MRI, and CT.

Higaki, T, Komoto, D, Iida, K, Kaichi, Y, Hirokawa, Y, Awai, K, Date, S, Automatic Laterality Detection Using Anatomic Regions of Interest (ROI) on Brain PET Images.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12024833.html